1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Virtio PCI driver - modern (virtio 1.0) device support
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.
8 * Copyright IBM Corp. 2007
9 * Copyright Red Hat, Inc. 2014
17 #include <linux/delay.h>
18 #define VIRTIO_PCI_NO_LEGACY
19 #include "virtio_pci_common.h"
22 * Type-safe wrappers for io accesses.
23 * Use these to enforce at compile time the following spec requirement:
25 * The driver MUST access each field using the “natural” access
26 * method, i.e. 32-bit accesses for 32-bit fields, 16-bit accesses
27 * for 16-bit fields and 8-bit accesses for 8-bit fields.
29 static inline u8 vp_ioread8(u8 __iomem *addr)
33 static inline u16 vp_ioread16 (__le16 __iomem *addr)
35 return ioread16(addr);
38 static inline u32 vp_ioread32(__le32 __iomem *addr)
40 return ioread32(addr);
43 static inline void vp_iowrite8(u8 value, u8 __iomem *addr)
45 iowrite8(value, addr);
48 static inline void vp_iowrite16(u16 value, __le16 __iomem *addr)
50 iowrite16(value, addr);
53 static inline void vp_iowrite32(u32 value, __le32 __iomem *addr)
55 iowrite32(value, addr);
58 static void vp_iowrite64_twopart(u64 val,
59 __le32 __iomem *lo, __le32 __iomem *hi)
61 vp_iowrite32((u32)val, lo);
62 vp_iowrite32(val >> 32, hi);
65 static void __iomem *map_capability(struct pci_dev *dev, int off,
75 pci_read_config_byte(dev, off + offsetof(struct virtio_pci_cap,
78 pci_read_config_dword(dev, off + offsetof(struct virtio_pci_cap, offset),
80 pci_read_config_dword(dev, off + offsetof(struct virtio_pci_cap, length),
83 if (length <= start) {
85 "virtio_pci: bad capability len %u (>%u expected)\n",
90 if (length - start < minlen) {
92 "virtio_pci: bad capability len %u (>=%zu expected)\n",
99 if (start + offset < offset) {
101 "virtio_pci: map wrap-around %u+%u\n",
108 if (offset & (align - 1)) {
110 "virtio_pci: offset %u not aligned to %u\n",
121 if (minlen + offset < minlen ||
122 minlen + offset > pci_resource_len(dev, bar)) {
124 "virtio_pci: map virtio %zu@%u "
125 "out of range on bar %i length %lu\n",
127 bar, (unsigned long)pci_resource_len(dev, bar));
131 p = pci_iomap_range(dev, bar, offset, length);
134 "virtio_pci: unable to map virtio %u@%u on bar %i\n",
135 length, offset, bar);
139 /* virtio config->get_features() implementation */
140 static u64 vp_get_features(struct virtio_device *vdev)
142 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
145 vp_iowrite32(0, &vp_dev->common->device_feature_select);
146 features = vp_ioread32(&vp_dev->common->device_feature);
147 vp_iowrite32(1, &vp_dev->common->device_feature_select);
148 features |= ((u64)vp_ioread32(&vp_dev->common->device_feature) << 32);
153 static void vp_transport_features(struct virtio_device *vdev, u64 features)
155 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
156 struct pci_dev *pci_dev = vp_dev->pci_dev;
158 if ((features & BIT_ULL(VIRTIO_F_SR_IOV)) &&
159 pci_find_ext_capability(pci_dev, PCI_EXT_CAP_ID_SRIOV))
160 __virtio_set_bit(vdev, VIRTIO_F_SR_IOV);
163 /* virtio config->finalize_features() implementation */
164 static int vp_finalize_features(struct virtio_device *vdev)
166 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
167 u64 features = vdev->features;
169 /* Give virtio_ring a chance to accept features. */
170 vring_transport_features(vdev);
172 /* Give virtio_pci a chance to accept features. */
173 vp_transport_features(vdev, features);
175 if (!__virtio_test_bit(vdev, VIRTIO_F_VERSION_1)) {
176 dev_err(&vdev->dev, "virtio: device uses modern interface "
177 "but does not have VIRTIO_F_VERSION_1\n");
181 vp_iowrite32(0, &vp_dev->common->guest_feature_select);
182 vp_iowrite32((u32)vdev->features, &vp_dev->common->guest_feature);
183 vp_iowrite32(1, &vp_dev->common->guest_feature_select);
184 vp_iowrite32(vdev->features >> 32, &vp_dev->common->guest_feature);
189 /* virtio config->get() implementation */
190 static void vp_get(struct virtio_device *vdev, unsigned offset,
191 void *buf, unsigned len)
193 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
198 BUG_ON(offset + len > vp_dev->device_len);
202 b = ioread8(vp_dev->device + offset);
203 memcpy(buf, &b, sizeof b);
206 w = cpu_to_le16(ioread16(vp_dev->device + offset));
207 memcpy(buf, &w, sizeof w);
210 l = cpu_to_le32(ioread32(vp_dev->device + offset));
211 memcpy(buf, &l, sizeof l);
214 l = cpu_to_le32(ioread32(vp_dev->device + offset));
215 memcpy(buf, &l, sizeof l);
216 l = cpu_to_le32(ioread32(vp_dev->device + offset + sizeof l));
217 memcpy(buf + sizeof l, &l, sizeof l);
224 /* the config->set() implementation. it's symmetric to the config->get()
226 static void vp_set(struct virtio_device *vdev, unsigned offset,
227 const void *buf, unsigned len)
229 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
234 BUG_ON(offset + len > vp_dev->device_len);
238 memcpy(&b, buf, sizeof b);
239 iowrite8(b, vp_dev->device + offset);
242 memcpy(&w, buf, sizeof w);
243 iowrite16(le16_to_cpu(w), vp_dev->device + offset);
246 memcpy(&l, buf, sizeof l);
247 iowrite32(le32_to_cpu(l), vp_dev->device + offset);
250 memcpy(&l, buf, sizeof l);
251 iowrite32(le32_to_cpu(l), vp_dev->device + offset);
252 memcpy(&l, buf + sizeof l, sizeof l);
253 iowrite32(le32_to_cpu(l), vp_dev->device + offset + sizeof l);
260 static u32 vp_generation(struct virtio_device *vdev)
262 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
263 return vp_ioread8(&vp_dev->common->config_generation);
266 /* config->{get,set}_status() implementations */
267 static u8 vp_get_status(struct virtio_device *vdev)
269 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
270 return vp_ioread8(&vp_dev->common->device_status);
273 static void vp_set_status(struct virtio_device *vdev, u8 status)
275 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
276 /* We should never be setting status to 0. */
278 vp_iowrite8(status, &vp_dev->common->device_status);
281 static void vp_reset(struct virtio_device *vdev)
283 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
284 /* 0 status means a reset. */
285 vp_iowrite8(0, &vp_dev->common->device_status);
286 /* After writing 0 to device_status, the driver MUST wait for a read of
287 * device_status to return 0 before reinitializing the device.
288 * This will flush out the status write, and flush in device writes,
289 * including MSI-X interrupts, if any.
291 while (vp_ioread8(&vp_dev->common->device_status))
293 /* Flush pending VQ/configuration callbacks. */
294 vp_synchronize_vectors(vdev);
297 static u16 vp_config_vector(struct virtio_pci_device *vp_dev, u16 vector)
299 /* Setup the vector used for configuration events */
300 vp_iowrite16(vector, &vp_dev->common->msix_config);
301 /* Verify we had enough resources to assign the vector */
302 /* Will also flush the write out to device */
303 return vp_ioread16(&vp_dev->common->msix_config);
306 static struct virtqueue *setup_vq(struct virtio_pci_device *vp_dev,
307 struct virtio_pci_vq_info *info,
309 void (*callback)(struct virtqueue *vq),
314 struct virtio_pci_common_cfg __iomem *cfg = vp_dev->common;
315 struct virtqueue *vq;
319 if (index >= vp_ioread16(&cfg->num_queues))
320 return ERR_PTR(-ENOENT);
322 /* Select the queue we're interested in */
323 vp_iowrite16(index, &cfg->queue_select);
325 /* Check if queue is either not available or already active. */
326 num = vp_ioread16(&cfg->queue_size);
327 if (!num || vp_ioread16(&cfg->queue_enable))
328 return ERR_PTR(-ENOENT);
330 if (num & (num - 1)) {
331 dev_warn(&vp_dev->pci_dev->dev, "bad queue size %u", num);
332 return ERR_PTR(-EINVAL);
335 /* get offset of notification word for this vq */
336 off = vp_ioread16(&cfg->queue_notify_off);
338 info->msix_vector = msix_vec;
340 /* create the vring */
341 vq = vring_create_virtqueue(index, num,
342 SMP_CACHE_BYTES, &vp_dev->vdev,
344 vp_notify, callback, name);
346 return ERR_PTR(-ENOMEM);
348 /* activate the queue */
349 vp_iowrite16(virtqueue_get_vring_size(vq), &cfg->queue_size);
350 vp_iowrite64_twopart(virtqueue_get_desc_addr(vq),
351 &cfg->queue_desc_lo, &cfg->queue_desc_hi);
352 vp_iowrite64_twopart(virtqueue_get_avail_addr(vq),
353 &cfg->queue_avail_lo, &cfg->queue_avail_hi);
354 vp_iowrite64_twopart(virtqueue_get_used_addr(vq),
355 &cfg->queue_used_lo, &cfg->queue_used_hi);
357 if (vp_dev->notify_base) {
358 /* offset should not wrap */
359 if ((u64)off * vp_dev->notify_offset_multiplier + 2
360 > vp_dev->notify_len) {
361 dev_warn(&vp_dev->pci_dev->dev,
362 "bad notification offset %u (x %u) "
363 "for queue %u > %zd",
364 off, vp_dev->notify_offset_multiplier,
365 index, vp_dev->notify_len);
369 vq->priv = (void __force *)vp_dev->notify_base +
370 off * vp_dev->notify_offset_multiplier;
372 vq->priv = (void __force *)map_capability(vp_dev->pci_dev,
373 vp_dev->notify_map_cap, 2, 2,
374 off * vp_dev->notify_offset_multiplier, 2,
383 if (msix_vec != VIRTIO_MSI_NO_VECTOR) {
384 vp_iowrite16(msix_vec, &cfg->queue_msix_vector);
385 msix_vec = vp_ioread16(&cfg->queue_msix_vector);
386 if (msix_vec == VIRTIO_MSI_NO_VECTOR) {
388 goto err_assign_vector;
395 if (!vp_dev->notify_base)
396 pci_iounmap(vp_dev->pci_dev, (void __iomem __force *)vq->priv);
398 vring_del_virtqueue(vq);
402 static int vp_modern_find_vqs(struct virtio_device *vdev, unsigned nvqs,
403 struct virtqueue *vqs[],
404 vq_callback_t *callbacks[],
405 const char * const names[], const bool *ctx,
406 struct irq_affinity *desc)
408 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
409 struct virtqueue *vq;
410 int rc = vp_find_vqs(vdev, nvqs, vqs, callbacks, names, ctx, desc);
415 /* Select and activate all queues. Has to be done last: once we do
416 * this, there's no way to go back except reset.
418 list_for_each_entry(vq, &vdev->vqs, list) {
419 vp_iowrite16(vq->index, &vp_dev->common->queue_select);
420 vp_iowrite16(1, &vp_dev->common->queue_enable);
426 static void del_vq(struct virtio_pci_vq_info *info)
428 struct virtqueue *vq = info->vq;
429 struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
431 vp_iowrite16(vq->index, &vp_dev->common->queue_select);
433 if (vp_dev->msix_enabled) {
434 vp_iowrite16(VIRTIO_MSI_NO_VECTOR,
435 &vp_dev->common->queue_msix_vector);
436 /* Flush the write out to device */
437 vp_ioread16(&vp_dev->common->queue_msix_vector);
440 if (!vp_dev->notify_base)
441 pci_iounmap(vp_dev->pci_dev, (void __force __iomem *)vq->priv);
443 vring_del_virtqueue(vq);
446 static const struct virtio_config_ops virtio_pci_config_nodev_ops = {
449 .generation = vp_generation,
450 .get_status = vp_get_status,
451 .set_status = vp_set_status,
453 .find_vqs = vp_modern_find_vqs,
454 .del_vqs = vp_del_vqs,
455 .get_features = vp_get_features,
456 .finalize_features = vp_finalize_features,
457 .bus_name = vp_bus_name,
458 .set_vq_affinity = vp_set_vq_affinity,
459 .get_vq_affinity = vp_get_vq_affinity,
462 static const struct virtio_config_ops virtio_pci_config_ops = {
465 .generation = vp_generation,
466 .get_status = vp_get_status,
467 .set_status = vp_set_status,
469 .find_vqs = vp_modern_find_vqs,
470 .del_vqs = vp_del_vqs,
471 .get_features = vp_get_features,
472 .finalize_features = vp_finalize_features,
473 .bus_name = vp_bus_name,
474 .set_vq_affinity = vp_set_vq_affinity,
475 .get_vq_affinity = vp_get_vq_affinity,
479 * virtio_pci_find_capability - walk capabilities to find device info.
480 * @dev: the pci device
481 * @cfg_type: the VIRTIO_PCI_CAP_* value we seek
482 * @ioresource_types: IORESOURCE_MEM and/or IORESOURCE_IO.
484 * Returns offset of the capability, or 0.
486 static inline int virtio_pci_find_capability(struct pci_dev *dev, u8 cfg_type,
487 u32 ioresource_types, int *bars)
491 for (pos = pci_find_capability(dev, PCI_CAP_ID_VNDR);
493 pos = pci_find_next_capability(dev, pos, PCI_CAP_ID_VNDR)) {
495 pci_read_config_byte(dev, pos + offsetof(struct virtio_pci_cap,
498 pci_read_config_byte(dev, pos + offsetof(struct virtio_pci_cap,
502 /* Ignore structures with reserved BAR values */
506 if (type == cfg_type) {
507 if (pci_resource_len(dev, bar) &&
508 pci_resource_flags(dev, bar) & ioresource_types) {
517 /* This is part of the ABI. Don't screw with it. */
518 static inline void check_offsets(void)
520 /* Note: disk space was harmed in compilation of this function. */
521 BUILD_BUG_ON(VIRTIO_PCI_CAP_VNDR !=
522 offsetof(struct virtio_pci_cap, cap_vndr));
523 BUILD_BUG_ON(VIRTIO_PCI_CAP_NEXT !=
524 offsetof(struct virtio_pci_cap, cap_next));
525 BUILD_BUG_ON(VIRTIO_PCI_CAP_LEN !=
526 offsetof(struct virtio_pci_cap, cap_len));
527 BUILD_BUG_ON(VIRTIO_PCI_CAP_CFG_TYPE !=
528 offsetof(struct virtio_pci_cap, cfg_type));
529 BUILD_BUG_ON(VIRTIO_PCI_CAP_BAR !=
530 offsetof(struct virtio_pci_cap, bar));
531 BUILD_BUG_ON(VIRTIO_PCI_CAP_OFFSET !=
532 offsetof(struct virtio_pci_cap, offset));
533 BUILD_BUG_ON(VIRTIO_PCI_CAP_LENGTH !=
534 offsetof(struct virtio_pci_cap, length));
535 BUILD_BUG_ON(VIRTIO_PCI_NOTIFY_CAP_MULT !=
536 offsetof(struct virtio_pci_notify_cap,
537 notify_off_multiplier));
538 BUILD_BUG_ON(VIRTIO_PCI_COMMON_DFSELECT !=
539 offsetof(struct virtio_pci_common_cfg,
540 device_feature_select));
541 BUILD_BUG_ON(VIRTIO_PCI_COMMON_DF !=
542 offsetof(struct virtio_pci_common_cfg, device_feature));
543 BUILD_BUG_ON(VIRTIO_PCI_COMMON_GFSELECT !=
544 offsetof(struct virtio_pci_common_cfg,
545 guest_feature_select));
546 BUILD_BUG_ON(VIRTIO_PCI_COMMON_GF !=
547 offsetof(struct virtio_pci_common_cfg, guest_feature));
548 BUILD_BUG_ON(VIRTIO_PCI_COMMON_MSIX !=
549 offsetof(struct virtio_pci_common_cfg, msix_config));
550 BUILD_BUG_ON(VIRTIO_PCI_COMMON_NUMQ !=
551 offsetof(struct virtio_pci_common_cfg, num_queues));
552 BUILD_BUG_ON(VIRTIO_PCI_COMMON_STATUS !=
553 offsetof(struct virtio_pci_common_cfg, device_status));
554 BUILD_BUG_ON(VIRTIO_PCI_COMMON_CFGGENERATION !=
555 offsetof(struct virtio_pci_common_cfg, config_generation));
556 BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_SELECT !=
557 offsetof(struct virtio_pci_common_cfg, queue_select));
558 BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_SIZE !=
559 offsetof(struct virtio_pci_common_cfg, queue_size));
560 BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_MSIX !=
561 offsetof(struct virtio_pci_common_cfg, queue_msix_vector));
562 BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_ENABLE !=
563 offsetof(struct virtio_pci_common_cfg, queue_enable));
564 BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_NOFF !=
565 offsetof(struct virtio_pci_common_cfg, queue_notify_off));
566 BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_DESCLO !=
567 offsetof(struct virtio_pci_common_cfg, queue_desc_lo));
568 BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_DESCHI !=
569 offsetof(struct virtio_pci_common_cfg, queue_desc_hi));
570 BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_AVAILLO !=
571 offsetof(struct virtio_pci_common_cfg, queue_avail_lo));
572 BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_AVAILHI !=
573 offsetof(struct virtio_pci_common_cfg, queue_avail_hi));
574 BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_USEDLO !=
575 offsetof(struct virtio_pci_common_cfg, queue_used_lo));
576 BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_USEDHI !=
577 offsetof(struct virtio_pci_common_cfg, queue_used_hi));
580 /* the PCI probing function */
581 int virtio_pci_modern_probe(struct virtio_pci_device *vp_dev)
583 struct pci_dev *pci_dev = vp_dev->pci_dev;
584 int err, common, isr, notify, device;
590 /* We only own devices >= 0x1000 and <= 0x107f: leave the rest. */
591 if (pci_dev->device < 0x1000 || pci_dev->device > 0x107f)
594 if (pci_dev->device < 0x1040) {
595 /* Transitional devices: use the PCI subsystem device id as
596 * virtio device id, same as legacy driver always did.
598 vp_dev->vdev.id.device = pci_dev->subsystem_device;
600 /* Modern devices: simply use PCI device id, but start from 0x1040. */
601 vp_dev->vdev.id.device = pci_dev->device - 0x1040;
603 vp_dev->vdev.id.vendor = pci_dev->subsystem_vendor;
605 /* check for a common config: if not, use legacy mode (bar 0). */
606 common = virtio_pci_find_capability(pci_dev, VIRTIO_PCI_CAP_COMMON_CFG,
607 IORESOURCE_IO | IORESOURCE_MEM,
608 &vp_dev->modern_bars);
610 dev_info(&pci_dev->dev,
611 "virtio_pci: leaving for legacy driver\n");
615 /* If common is there, these should be too... */
616 isr = virtio_pci_find_capability(pci_dev, VIRTIO_PCI_CAP_ISR_CFG,
617 IORESOURCE_IO | IORESOURCE_MEM,
618 &vp_dev->modern_bars);
619 notify = virtio_pci_find_capability(pci_dev, VIRTIO_PCI_CAP_NOTIFY_CFG,
620 IORESOURCE_IO | IORESOURCE_MEM,
621 &vp_dev->modern_bars);
622 if (!isr || !notify) {
623 dev_err(&pci_dev->dev,
624 "virtio_pci: missing capabilities %i/%i/%i\n",
625 common, isr, notify);
629 err = dma_set_mask_and_coherent(&pci_dev->dev, DMA_BIT_MASK(64));
631 err = dma_set_mask_and_coherent(&pci_dev->dev,
634 dev_warn(&pci_dev->dev, "Failed to enable 64-bit or 32-bit DMA. Trying to continue, but this might not work.\n");
636 /* Device capability is only mandatory for devices that have
637 * device-specific configuration.
639 device = virtio_pci_find_capability(pci_dev, VIRTIO_PCI_CAP_DEVICE_CFG,
640 IORESOURCE_IO | IORESOURCE_MEM,
641 &vp_dev->modern_bars);
643 err = pci_request_selected_regions(pci_dev, vp_dev->modern_bars,
644 "virtio-pci-modern");
649 vp_dev->common = map_capability(pci_dev, common,
650 sizeof(struct virtio_pci_common_cfg), 4,
651 0, sizeof(struct virtio_pci_common_cfg),
655 vp_dev->isr = map_capability(pci_dev, isr, sizeof(u8), 1,
661 /* Read notify_off_multiplier from config space. */
662 pci_read_config_dword(pci_dev,
663 notify + offsetof(struct virtio_pci_notify_cap,
664 notify_off_multiplier),
665 &vp_dev->notify_offset_multiplier);
666 /* Read notify length and offset from config space. */
667 pci_read_config_dword(pci_dev,
668 notify + offsetof(struct virtio_pci_notify_cap,
672 pci_read_config_dword(pci_dev,
673 notify + offsetof(struct virtio_pci_notify_cap,
677 /* We don't know how many VQs we'll map, ahead of the time.
678 * If notify length is small, map it all now.
679 * Otherwise, map each VQ individually later.
681 if ((u64)notify_length + (notify_offset % PAGE_SIZE) <= PAGE_SIZE) {
682 vp_dev->notify_base = map_capability(pci_dev, notify, 2, 2,
684 &vp_dev->notify_len);
685 if (!vp_dev->notify_base)
688 vp_dev->notify_map_cap = notify;
691 /* Again, we don't know how much we should map, but PAGE_SIZE
692 * is more than enough for all existing devices.
695 vp_dev->device = map_capability(pci_dev, device, 0, 4,
697 &vp_dev->device_len);
701 vp_dev->vdev.config = &virtio_pci_config_ops;
703 vp_dev->vdev.config = &virtio_pci_config_nodev_ops;
706 vp_dev->config_vector = vp_config_vector;
707 vp_dev->setup_vq = setup_vq;
708 vp_dev->del_vq = del_vq;
713 if (vp_dev->notify_base)
714 pci_iounmap(pci_dev, vp_dev->notify_base);
716 pci_iounmap(pci_dev, vp_dev->isr);
718 pci_iounmap(pci_dev, vp_dev->common);
723 void virtio_pci_modern_remove(struct virtio_pci_device *vp_dev)
725 struct pci_dev *pci_dev = vp_dev->pci_dev;
728 pci_iounmap(pci_dev, vp_dev->device);
729 if (vp_dev->notify_base)
730 pci_iounmap(pci_dev, vp_dev->notify_base);
731 pci_iounmap(pci_dev, vp_dev->isr);
732 pci_iounmap(pci_dev, vp_dev->common);
733 pci_release_selected_regions(pci_dev, vp_dev->modern_bars);