2 * Virtio PCI driver - modern (virtio 1.0) device support
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.
7 * Copyright IBM Corp. 2007
8 * Copyright Red Hat, Inc. 2014
11 * Anthony Liguori <aliguori@us.ibm.com>
12 * Rusty Russell <rusty@rustcorp.com.au>
13 * Michael S. Tsirkin <mst@redhat.com>
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.
20 #include <linux/delay.h>
21 #define VIRTIO_PCI_NO_LEGACY
22 #include "virtio_pci_common.h"
25 * Type-safe wrappers for io accesses.
26 * Use these to enforce at compile time the following spec requirement:
28 * The driver MUST access each field using the “natural” access
29 * method, i.e. 32-bit accesses for 32-bit fields, 16-bit accesses
30 * for 16-bit fields and 8-bit accesses for 8-bit fields.
32 static inline u8 vp_ioread8(u8 __iomem *addr)
36 static inline u16 vp_ioread16 (u16 __iomem *addr)
38 return ioread16(addr);
41 static inline u32 vp_ioread32(u32 __iomem *addr)
43 return ioread32(addr);
46 static inline void vp_iowrite8(u8 value, u8 __iomem *addr)
48 iowrite8(value, addr);
51 static inline void vp_iowrite16(u16 value, u16 __iomem *addr)
53 iowrite16(value, addr);
56 static inline void vp_iowrite32(u32 value, u32 __iomem *addr)
58 iowrite32(value, addr);
61 static void vp_iowrite64_twopart(u64 val,
62 __le32 __iomem *lo, __le32 __iomem *hi)
64 vp_iowrite32((u32)val, lo);
65 vp_iowrite32(val >> 32, hi);
68 static void __iomem *map_capability(struct pci_dev *dev, int off,
78 pci_read_config_byte(dev, off + offsetof(struct virtio_pci_cap,
81 pci_read_config_dword(dev, off + offsetof(struct virtio_pci_cap, offset),
83 pci_read_config_dword(dev, off + offsetof(struct virtio_pci_cap, length),
86 if (length <= start) {
88 "virtio_pci: bad capability len %u (>%u expected)\n",
93 if (length - start < minlen) {
95 "virtio_pci: bad capability len %u (>=%zu expected)\n",
102 if (start + offset < offset) {
104 "virtio_pci: map wrap-around %u+%u\n",
111 if (offset & (align - 1)) {
113 "virtio_pci: offset %u not aligned to %u\n",
124 if (minlen + offset < minlen ||
125 minlen + offset > pci_resource_len(dev, bar)) {
127 "virtio_pci: map virtio %zu@%u "
128 "out of range on bar %i length %lu\n",
130 bar, (unsigned long)pci_resource_len(dev, bar));
134 p = pci_iomap_range(dev, bar, offset, length);
137 "virtio_pci: unable to map virtio %u@%u on bar %i\n",
138 length, offset, bar);
142 /* virtio config->get_features() implementation */
143 static u64 vp_get_features(struct virtio_device *vdev)
145 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
148 vp_iowrite32(0, &vp_dev->common->device_feature_select);
149 features = vp_ioread32(&vp_dev->common->device_feature);
150 vp_iowrite32(1, &vp_dev->common->device_feature_select);
151 features |= ((u64)vp_ioread32(&vp_dev->common->device_feature) << 32);
156 /* virtio config->finalize_features() implementation */
157 static int vp_finalize_features(struct virtio_device *vdev)
159 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
161 /* Give virtio_ring a chance to accept features. */
162 vring_transport_features(vdev);
164 if (!__virtio_test_bit(vdev, VIRTIO_F_VERSION_1)) {
165 dev_err(&vdev->dev, "virtio: device uses modern interface "
166 "but does not have VIRTIO_F_VERSION_1\n");
170 vp_iowrite32(0, &vp_dev->common->guest_feature_select);
171 vp_iowrite32((u32)vdev->features, &vp_dev->common->guest_feature);
172 vp_iowrite32(1, &vp_dev->common->guest_feature_select);
173 vp_iowrite32(vdev->features >> 32, &vp_dev->common->guest_feature);
178 /* virtio config->get() implementation */
179 static void vp_get(struct virtio_device *vdev, unsigned offset,
180 void *buf, unsigned len)
182 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
187 BUG_ON(offset + len > vp_dev->device_len);
191 b = ioread8(vp_dev->device + offset);
192 memcpy(buf, &b, sizeof b);
195 w = cpu_to_le16(ioread16(vp_dev->device + offset));
196 memcpy(buf, &w, sizeof w);
199 l = cpu_to_le32(ioread32(vp_dev->device + offset));
200 memcpy(buf, &l, sizeof l);
203 l = cpu_to_le32(ioread32(vp_dev->device + offset));
204 memcpy(buf, &l, sizeof l);
205 l = cpu_to_le32(ioread32(vp_dev->device + offset + sizeof l));
206 memcpy(buf + sizeof l, &l, sizeof l);
213 /* the config->set() implementation. it's symmetric to the config->get()
215 static void vp_set(struct virtio_device *vdev, unsigned offset,
216 const void *buf, unsigned len)
218 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
223 BUG_ON(offset + len > vp_dev->device_len);
227 memcpy(&b, buf, sizeof b);
228 iowrite8(b, vp_dev->device + offset);
231 memcpy(&w, buf, sizeof w);
232 iowrite16(le16_to_cpu(w), vp_dev->device + offset);
235 memcpy(&l, buf, sizeof l);
236 iowrite32(le32_to_cpu(l), vp_dev->device + offset);
239 memcpy(&l, buf, sizeof l);
240 iowrite32(le32_to_cpu(l), vp_dev->device + offset);
241 memcpy(&l, buf + sizeof l, sizeof l);
242 iowrite32(le32_to_cpu(l), vp_dev->device + offset + sizeof l);
249 static u32 vp_generation(struct virtio_device *vdev)
251 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
252 return vp_ioread8(&vp_dev->common->config_generation);
255 /* config->{get,set}_status() implementations */
256 static u8 vp_get_status(struct virtio_device *vdev)
258 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
259 return vp_ioread8(&vp_dev->common->device_status);
262 static void vp_set_status(struct virtio_device *vdev, u8 status)
264 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
265 /* We should never be setting status to 0. */
267 vp_iowrite8(status, &vp_dev->common->device_status);
270 static void vp_reset(struct virtio_device *vdev)
272 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
273 /* 0 status means a reset. */
274 vp_iowrite8(0, &vp_dev->common->device_status);
275 /* After writing 0 to device_status, the driver MUST wait for a read of
276 * device_status to return 0 before reinitializing the device.
277 * This will flush out the status write, and flush in device writes,
278 * including MSI-X interrupts, if any.
280 while (vp_ioread8(&vp_dev->common->device_status))
282 /* Flush pending VQ/configuration callbacks. */
283 vp_synchronize_vectors(vdev);
286 static u16 vp_config_vector(struct virtio_pci_device *vp_dev, u16 vector)
288 /* Setup the vector used for configuration events */
289 vp_iowrite16(vector, &vp_dev->common->msix_config);
290 /* Verify we had enough resources to assign the vector */
291 /* Will also flush the write out to device */
292 return vp_ioread16(&vp_dev->common->msix_config);
295 static size_t vring_pci_size(u16 num)
297 /* We only need a cacheline separation. */
298 return PAGE_ALIGN(vring_size(num, SMP_CACHE_BYTES));
301 static void *alloc_virtqueue_pages(int *num)
305 /* TODO: allocate each queue chunk individually */
306 for (; *num && vring_pci_size(*num) > PAGE_SIZE; *num /= 2) {
307 pages = alloc_pages_exact(vring_pci_size(*num),
308 GFP_KERNEL|__GFP_ZERO|__GFP_NOWARN);
316 /* Try to get a single page. You are my only hope! */
317 return alloc_pages_exact(vring_pci_size(*num), GFP_KERNEL|__GFP_ZERO);
320 static struct virtqueue *setup_vq(struct virtio_pci_device *vp_dev,
321 struct virtio_pci_vq_info *info,
323 void (*callback)(struct virtqueue *vq),
327 struct virtio_pci_common_cfg __iomem *cfg = vp_dev->common;
328 struct virtqueue *vq;
332 if (index >= vp_ioread16(&cfg->num_queues))
333 return ERR_PTR(-ENOENT);
335 /* Select the queue we're interested in */
336 vp_iowrite16(index, &cfg->queue_select);
338 /* Check if queue is either not available or already active. */
339 num = vp_ioread16(&cfg->queue_size);
340 if (!num || vp_ioread16(&cfg->queue_enable))
341 return ERR_PTR(-ENOENT);
343 if (num & (num - 1)) {
344 dev_warn(&vp_dev->pci_dev->dev, "bad queue size %u", num);
345 return ERR_PTR(-EINVAL);
348 /* get offset of notification word for this vq */
349 off = vp_ioread16(&cfg->queue_notify_off);
352 info->msix_vector = msix_vec;
354 info->queue = alloc_virtqueue_pages(&info->num);
355 if (info->queue == NULL)
356 return ERR_PTR(-ENOMEM);
358 /* create the vring */
359 vq = vring_new_virtqueue(index, info->num,
360 SMP_CACHE_BYTES, &vp_dev->vdev,
361 true, info->queue, vp_notify, callback, name);
367 /* activate the queue */
368 vp_iowrite16(num, &cfg->queue_size);
369 vp_iowrite64_twopart(virt_to_phys(info->queue),
370 &cfg->queue_desc_lo, &cfg->queue_desc_hi);
371 vp_iowrite64_twopart(virt_to_phys(virtqueue_get_avail(vq)),
372 &cfg->queue_avail_lo, &cfg->queue_avail_hi);
373 vp_iowrite64_twopart(virt_to_phys(virtqueue_get_used(vq)),
374 &cfg->queue_used_lo, &cfg->queue_used_hi);
376 if (vp_dev->notify_base) {
377 /* offset should not wrap */
378 if ((u64)off * vp_dev->notify_offset_multiplier + 2
379 > vp_dev->notify_len) {
380 dev_warn(&vp_dev->pci_dev->dev,
381 "bad notification offset %u (x %u) "
382 "for queue %u > %zd",
383 off, vp_dev->notify_offset_multiplier,
384 index, vp_dev->notify_len);
388 vq->priv = (void __force *)vp_dev->notify_base +
389 off * vp_dev->notify_offset_multiplier;
391 vq->priv = (void __force *)map_capability(vp_dev->pci_dev,
392 vp_dev->notify_map_cap, 2, 2,
393 off * vp_dev->notify_offset_multiplier, 2,
402 if (msix_vec != VIRTIO_MSI_NO_VECTOR) {
403 vp_iowrite16(msix_vec, &cfg->queue_msix_vector);
404 msix_vec = vp_ioread16(&cfg->queue_msix_vector);
405 if (msix_vec == VIRTIO_MSI_NO_VECTOR) {
407 goto err_assign_vector;
414 if (!vp_dev->notify_base)
415 pci_iounmap(vp_dev->pci_dev, (void __iomem __force *)vq->priv);
417 vring_del_virtqueue(vq);
419 free_pages_exact(info->queue, vring_pci_size(info->num));
423 static int vp_modern_find_vqs(struct virtio_device *vdev, unsigned nvqs,
424 struct virtqueue *vqs[],
425 vq_callback_t *callbacks[],
428 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
429 struct virtqueue *vq;
430 int rc = vp_find_vqs(vdev, nvqs, vqs, callbacks, names);
435 /* Select and activate all queues. Has to be done last: once we do
436 * this, there's no way to go back except reset.
438 list_for_each_entry(vq, &vdev->vqs, list) {
439 vp_iowrite16(vq->index, &vp_dev->common->queue_select);
440 vp_iowrite16(1, &vp_dev->common->queue_enable);
446 static void del_vq(struct virtio_pci_vq_info *info)
448 struct virtqueue *vq = info->vq;
449 struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
451 vp_iowrite16(vq->index, &vp_dev->common->queue_select);
453 if (vp_dev->msix_enabled) {
454 vp_iowrite16(VIRTIO_MSI_NO_VECTOR,
455 &vp_dev->common->queue_msix_vector);
456 /* Flush the write out to device */
457 vp_ioread16(&vp_dev->common->queue_msix_vector);
460 if (!vp_dev->notify_base)
461 pci_iounmap(vp_dev->pci_dev, (void __force __iomem *)vq->priv);
463 vring_del_virtqueue(vq);
465 free_pages_exact(info->queue, vring_pci_size(info->num));
468 static const struct virtio_config_ops virtio_pci_config_nodev_ops = {
471 .generation = vp_generation,
472 .get_status = vp_get_status,
473 .set_status = vp_set_status,
475 .find_vqs = vp_modern_find_vqs,
476 .del_vqs = vp_del_vqs,
477 .get_features = vp_get_features,
478 .finalize_features = vp_finalize_features,
479 .bus_name = vp_bus_name,
480 .set_vq_affinity = vp_set_vq_affinity,
483 static const struct virtio_config_ops virtio_pci_config_ops = {
486 .generation = vp_generation,
487 .get_status = vp_get_status,
488 .set_status = vp_set_status,
490 .find_vqs = vp_modern_find_vqs,
491 .del_vqs = vp_del_vqs,
492 .get_features = vp_get_features,
493 .finalize_features = vp_finalize_features,
494 .bus_name = vp_bus_name,
495 .set_vq_affinity = vp_set_vq_affinity,
499 * virtio_pci_find_capability - walk capabilities to find device info.
500 * @dev: the pci device
501 * @cfg_type: the VIRTIO_PCI_CAP_* value we seek
502 * @ioresource_types: IORESOURCE_MEM and/or IORESOURCE_IO.
504 * Returns offset of the capability, or 0.
506 static inline int virtio_pci_find_capability(struct pci_dev *dev, u8 cfg_type,
507 u32 ioresource_types, int *bars)
511 for (pos = pci_find_capability(dev, PCI_CAP_ID_VNDR);
513 pos = pci_find_next_capability(dev, pos, PCI_CAP_ID_VNDR)) {
515 pci_read_config_byte(dev, pos + offsetof(struct virtio_pci_cap,
518 pci_read_config_byte(dev, pos + offsetof(struct virtio_pci_cap,
522 /* Ignore structures with reserved BAR values */
526 if (type == cfg_type) {
527 if (pci_resource_len(dev, bar) &&
528 pci_resource_flags(dev, bar) & ioresource_types) {
537 /* This is part of the ABI. Don't screw with it. */
538 static inline void check_offsets(void)
540 /* Note: disk space was harmed in compilation of this function. */
541 BUILD_BUG_ON(VIRTIO_PCI_CAP_VNDR !=
542 offsetof(struct virtio_pci_cap, cap_vndr));
543 BUILD_BUG_ON(VIRTIO_PCI_CAP_NEXT !=
544 offsetof(struct virtio_pci_cap, cap_next));
545 BUILD_BUG_ON(VIRTIO_PCI_CAP_LEN !=
546 offsetof(struct virtio_pci_cap, cap_len));
547 BUILD_BUG_ON(VIRTIO_PCI_CAP_CFG_TYPE !=
548 offsetof(struct virtio_pci_cap, cfg_type));
549 BUILD_BUG_ON(VIRTIO_PCI_CAP_BAR !=
550 offsetof(struct virtio_pci_cap, bar));
551 BUILD_BUG_ON(VIRTIO_PCI_CAP_OFFSET !=
552 offsetof(struct virtio_pci_cap, offset));
553 BUILD_BUG_ON(VIRTIO_PCI_CAP_LENGTH !=
554 offsetof(struct virtio_pci_cap, length));
555 BUILD_BUG_ON(VIRTIO_PCI_NOTIFY_CAP_MULT !=
556 offsetof(struct virtio_pci_notify_cap,
557 notify_off_multiplier));
558 BUILD_BUG_ON(VIRTIO_PCI_COMMON_DFSELECT !=
559 offsetof(struct virtio_pci_common_cfg,
560 device_feature_select));
561 BUILD_BUG_ON(VIRTIO_PCI_COMMON_DF !=
562 offsetof(struct virtio_pci_common_cfg, device_feature));
563 BUILD_BUG_ON(VIRTIO_PCI_COMMON_GFSELECT !=
564 offsetof(struct virtio_pci_common_cfg,
565 guest_feature_select));
566 BUILD_BUG_ON(VIRTIO_PCI_COMMON_GF !=
567 offsetof(struct virtio_pci_common_cfg, guest_feature));
568 BUILD_BUG_ON(VIRTIO_PCI_COMMON_MSIX !=
569 offsetof(struct virtio_pci_common_cfg, msix_config));
570 BUILD_BUG_ON(VIRTIO_PCI_COMMON_NUMQ !=
571 offsetof(struct virtio_pci_common_cfg, num_queues));
572 BUILD_BUG_ON(VIRTIO_PCI_COMMON_STATUS !=
573 offsetof(struct virtio_pci_common_cfg, device_status));
574 BUILD_BUG_ON(VIRTIO_PCI_COMMON_CFGGENERATION !=
575 offsetof(struct virtio_pci_common_cfg, config_generation));
576 BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_SELECT !=
577 offsetof(struct virtio_pci_common_cfg, queue_select));
578 BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_SIZE !=
579 offsetof(struct virtio_pci_common_cfg, queue_size));
580 BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_MSIX !=
581 offsetof(struct virtio_pci_common_cfg, queue_msix_vector));
582 BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_ENABLE !=
583 offsetof(struct virtio_pci_common_cfg, queue_enable));
584 BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_NOFF !=
585 offsetof(struct virtio_pci_common_cfg, queue_notify_off));
586 BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_DESCLO !=
587 offsetof(struct virtio_pci_common_cfg, queue_desc_lo));
588 BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_DESCHI !=
589 offsetof(struct virtio_pci_common_cfg, queue_desc_hi));
590 BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_AVAILLO !=
591 offsetof(struct virtio_pci_common_cfg, queue_avail_lo));
592 BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_AVAILHI !=
593 offsetof(struct virtio_pci_common_cfg, queue_avail_hi));
594 BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_USEDLO !=
595 offsetof(struct virtio_pci_common_cfg, queue_used_lo));
596 BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_USEDHI !=
597 offsetof(struct virtio_pci_common_cfg, queue_used_hi));
600 /* the PCI probing function */
601 int virtio_pci_modern_probe(struct virtio_pci_device *vp_dev)
603 struct pci_dev *pci_dev = vp_dev->pci_dev;
604 int err, common, isr, notify, device;
610 /* We only own devices >= 0x1000 and <= 0x107f: leave the rest. */
611 if (pci_dev->device < 0x1000 || pci_dev->device > 0x107f)
614 if (pci_dev->device < 0x1040) {
615 /* Transitional devices: use the PCI subsystem device id as
616 * virtio device id, same as legacy driver always did.
618 vp_dev->vdev.id.device = pci_dev->subsystem_device;
620 /* Modern devices: simply use PCI device id, but start from 0x1040. */
621 vp_dev->vdev.id.device = pci_dev->device - 0x1040;
623 vp_dev->vdev.id.vendor = pci_dev->subsystem_vendor;
625 /* check for a common config: if not, use legacy mode (bar 0). */
626 common = virtio_pci_find_capability(pci_dev, VIRTIO_PCI_CAP_COMMON_CFG,
627 IORESOURCE_IO | IORESOURCE_MEM,
628 &vp_dev->modern_bars);
630 dev_info(&pci_dev->dev,
631 "virtio_pci: leaving for legacy driver\n");
635 /* If common is there, these should be too... */
636 isr = virtio_pci_find_capability(pci_dev, VIRTIO_PCI_CAP_ISR_CFG,
637 IORESOURCE_IO | IORESOURCE_MEM,
638 &vp_dev->modern_bars);
639 notify = virtio_pci_find_capability(pci_dev, VIRTIO_PCI_CAP_NOTIFY_CFG,
640 IORESOURCE_IO | IORESOURCE_MEM,
641 &vp_dev->modern_bars);
642 if (!isr || !notify) {
643 dev_err(&pci_dev->dev,
644 "virtio_pci: missing capabilities %i/%i/%i\n",
645 common, isr, notify);
649 /* Device capability is only mandatory for devices that have
650 * device-specific configuration.
652 device = virtio_pci_find_capability(pci_dev, VIRTIO_PCI_CAP_DEVICE_CFG,
653 IORESOURCE_IO | IORESOURCE_MEM,
654 &vp_dev->modern_bars);
656 err = pci_request_selected_regions(pci_dev, vp_dev->modern_bars,
657 "virtio-pci-modern");
662 vp_dev->common = map_capability(pci_dev, common,
663 sizeof(struct virtio_pci_common_cfg), 4,
664 0, sizeof(struct virtio_pci_common_cfg),
668 vp_dev->isr = map_capability(pci_dev, isr, sizeof(u8), 1,
674 /* Read notify_off_multiplier from config space. */
675 pci_read_config_dword(pci_dev,
676 notify + offsetof(struct virtio_pci_notify_cap,
677 notify_off_multiplier),
678 &vp_dev->notify_offset_multiplier);
679 /* Read notify length and offset from config space. */
680 pci_read_config_dword(pci_dev,
681 notify + offsetof(struct virtio_pci_notify_cap,
685 pci_read_config_dword(pci_dev,
686 notify + offsetof(struct virtio_pci_notify_cap,
690 /* We don't know how many VQs we'll map, ahead of the time.
691 * If notify length is small, map it all now.
692 * Otherwise, map each VQ individually later.
694 if ((u64)notify_length + (notify_offset % PAGE_SIZE) <= PAGE_SIZE) {
695 vp_dev->notify_base = map_capability(pci_dev, notify, 2, 2,
697 &vp_dev->notify_len);
698 if (!vp_dev->notify_base)
701 vp_dev->notify_map_cap = notify;
704 /* Again, we don't know how much we should map, but PAGE_SIZE
705 * is more than enough for all existing devices.
708 vp_dev->device = map_capability(pci_dev, device, 0, 4,
710 &vp_dev->device_len);
714 vp_dev->vdev.config = &virtio_pci_config_ops;
716 vp_dev->vdev.config = &virtio_pci_config_nodev_ops;
719 vp_dev->config_vector = vp_config_vector;
720 vp_dev->setup_vq = setup_vq;
721 vp_dev->del_vq = del_vq;
726 if (vp_dev->notify_base)
727 pci_iounmap(pci_dev, vp_dev->notify_base);
729 pci_iounmap(pci_dev, vp_dev->isr);
731 pci_iounmap(pci_dev, vp_dev->common);
736 void virtio_pci_modern_remove(struct virtio_pci_device *vp_dev)
738 struct pci_dev *pci_dev = vp_dev->pci_dev;
741 pci_iounmap(pci_dev, vp_dev->device);
742 if (vp_dev->notify_base)
743 pci_iounmap(pci_dev, vp_dev->notify_base);
744 pci_iounmap(pci_dev, vp_dev->isr);
745 pci_iounmap(pci_dev, vp_dev->common);
746 pci_release_selected_regions(pci_dev, vp_dev->modern_bars);