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 (__le16 __iomem *addr)
38 return ioread16(addr);
41 static inline u32 vp_ioread32(__le32 __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, __le16 __iomem *addr)
53 iowrite16(value, addr);
56 static inline void vp_iowrite32(u32 value, __le32 __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 struct virtqueue *setup_vq(struct virtio_pci_device *vp_dev,
296 struct virtio_pci_vq_info *info,
298 void (*callback)(struct virtqueue *vq),
303 struct virtio_pci_common_cfg __iomem *cfg = vp_dev->common;
304 struct virtqueue *vq;
308 if (index >= vp_ioread16(&cfg->num_queues))
309 return ERR_PTR(-ENOENT);
311 /* Select the queue we're interested in */
312 vp_iowrite16(index, &cfg->queue_select);
314 /* Check if queue is either not available or already active. */
315 num = vp_ioread16(&cfg->queue_size);
316 if (!num || vp_ioread16(&cfg->queue_enable))
317 return ERR_PTR(-ENOENT);
319 if (num & (num - 1)) {
320 dev_warn(&vp_dev->pci_dev->dev, "bad queue size %u", num);
321 return ERR_PTR(-EINVAL);
324 /* get offset of notification word for this vq */
325 off = vp_ioread16(&cfg->queue_notify_off);
327 info->msix_vector = msix_vec;
329 /* create the vring */
330 vq = vring_create_virtqueue(index, num,
331 SMP_CACHE_BYTES, &vp_dev->vdev,
333 vp_notify, callback, name);
335 return ERR_PTR(-ENOMEM);
337 /* activate the queue */
338 vp_iowrite16(virtqueue_get_vring_size(vq), &cfg->queue_size);
339 vp_iowrite64_twopart(virtqueue_get_desc_addr(vq),
340 &cfg->queue_desc_lo, &cfg->queue_desc_hi);
341 vp_iowrite64_twopart(virtqueue_get_avail_addr(vq),
342 &cfg->queue_avail_lo, &cfg->queue_avail_hi);
343 vp_iowrite64_twopart(virtqueue_get_used_addr(vq),
344 &cfg->queue_used_lo, &cfg->queue_used_hi);
346 if (vp_dev->notify_base) {
347 /* offset should not wrap */
348 if ((u64)off * vp_dev->notify_offset_multiplier + 2
349 > vp_dev->notify_len) {
350 dev_warn(&vp_dev->pci_dev->dev,
351 "bad notification offset %u (x %u) "
352 "for queue %u > %zd",
353 off, vp_dev->notify_offset_multiplier,
354 index, vp_dev->notify_len);
358 vq->priv = (void __force *)vp_dev->notify_base +
359 off * vp_dev->notify_offset_multiplier;
361 vq->priv = (void __force *)map_capability(vp_dev->pci_dev,
362 vp_dev->notify_map_cap, 2, 2,
363 off * vp_dev->notify_offset_multiplier, 2,
372 if (msix_vec != VIRTIO_MSI_NO_VECTOR) {
373 vp_iowrite16(msix_vec, &cfg->queue_msix_vector);
374 msix_vec = vp_ioread16(&cfg->queue_msix_vector);
375 if (msix_vec == VIRTIO_MSI_NO_VECTOR) {
377 goto err_assign_vector;
384 if (!vp_dev->notify_base)
385 pci_iounmap(vp_dev->pci_dev, (void __iomem __force *)vq->priv);
387 vring_del_virtqueue(vq);
391 static int vp_modern_find_vqs(struct virtio_device *vdev, unsigned nvqs,
392 struct virtqueue *vqs[],
393 vq_callback_t *callbacks[],
394 const char * const names[], const bool *ctx,
395 struct irq_affinity *desc)
397 struct virtio_pci_device *vp_dev = to_vp_device(vdev);
398 struct virtqueue *vq;
399 int rc = vp_find_vqs(vdev, nvqs, vqs, callbacks, names, ctx, desc);
404 /* Select and activate all queues. Has to be done last: once we do
405 * this, there's no way to go back except reset.
407 list_for_each_entry(vq, &vdev->vqs, list) {
408 vp_iowrite16(vq->index, &vp_dev->common->queue_select);
409 vp_iowrite16(1, &vp_dev->common->queue_enable);
415 static void del_vq(struct virtio_pci_vq_info *info)
417 struct virtqueue *vq = info->vq;
418 struct virtio_pci_device *vp_dev = to_vp_device(vq->vdev);
420 vp_iowrite16(vq->index, &vp_dev->common->queue_select);
422 if (vp_dev->msix_enabled) {
423 vp_iowrite16(VIRTIO_MSI_NO_VECTOR,
424 &vp_dev->common->queue_msix_vector);
425 /* Flush the write out to device */
426 vp_ioread16(&vp_dev->common->queue_msix_vector);
429 if (!vp_dev->notify_base)
430 pci_iounmap(vp_dev->pci_dev, (void __force __iomem *)vq->priv);
432 vring_del_virtqueue(vq);
435 static const struct virtio_config_ops virtio_pci_config_nodev_ops = {
438 .generation = vp_generation,
439 .get_status = vp_get_status,
440 .set_status = vp_set_status,
442 .find_vqs = vp_modern_find_vqs,
443 .del_vqs = vp_del_vqs,
444 .get_features = vp_get_features,
445 .finalize_features = vp_finalize_features,
446 .bus_name = vp_bus_name,
447 .set_vq_affinity = vp_set_vq_affinity,
448 .get_vq_affinity = vp_get_vq_affinity,
451 static const struct virtio_config_ops virtio_pci_config_ops = {
454 .generation = vp_generation,
455 .get_status = vp_get_status,
456 .set_status = vp_set_status,
458 .find_vqs = vp_modern_find_vqs,
459 .del_vqs = vp_del_vqs,
460 .get_features = vp_get_features,
461 .finalize_features = vp_finalize_features,
462 .bus_name = vp_bus_name,
463 .set_vq_affinity = vp_set_vq_affinity,
464 .get_vq_affinity = vp_get_vq_affinity,
468 * virtio_pci_find_capability - walk capabilities to find device info.
469 * @dev: the pci device
470 * @cfg_type: the VIRTIO_PCI_CAP_* value we seek
471 * @ioresource_types: IORESOURCE_MEM and/or IORESOURCE_IO.
473 * Returns offset of the capability, or 0.
475 static inline int virtio_pci_find_capability(struct pci_dev *dev, u8 cfg_type,
476 u32 ioresource_types, int *bars)
480 for (pos = pci_find_capability(dev, PCI_CAP_ID_VNDR);
482 pos = pci_find_next_capability(dev, pos, PCI_CAP_ID_VNDR)) {
484 pci_read_config_byte(dev, pos + offsetof(struct virtio_pci_cap,
487 pci_read_config_byte(dev, pos + offsetof(struct virtio_pci_cap,
491 /* Ignore structures with reserved BAR values */
495 if (type == cfg_type) {
496 if (pci_resource_len(dev, bar) &&
497 pci_resource_flags(dev, bar) & ioresource_types) {
506 /* This is part of the ABI. Don't screw with it. */
507 static inline void check_offsets(void)
509 /* Note: disk space was harmed in compilation of this function. */
510 BUILD_BUG_ON(VIRTIO_PCI_CAP_VNDR !=
511 offsetof(struct virtio_pci_cap, cap_vndr));
512 BUILD_BUG_ON(VIRTIO_PCI_CAP_NEXT !=
513 offsetof(struct virtio_pci_cap, cap_next));
514 BUILD_BUG_ON(VIRTIO_PCI_CAP_LEN !=
515 offsetof(struct virtio_pci_cap, cap_len));
516 BUILD_BUG_ON(VIRTIO_PCI_CAP_CFG_TYPE !=
517 offsetof(struct virtio_pci_cap, cfg_type));
518 BUILD_BUG_ON(VIRTIO_PCI_CAP_BAR !=
519 offsetof(struct virtio_pci_cap, bar));
520 BUILD_BUG_ON(VIRTIO_PCI_CAP_OFFSET !=
521 offsetof(struct virtio_pci_cap, offset));
522 BUILD_BUG_ON(VIRTIO_PCI_CAP_LENGTH !=
523 offsetof(struct virtio_pci_cap, length));
524 BUILD_BUG_ON(VIRTIO_PCI_NOTIFY_CAP_MULT !=
525 offsetof(struct virtio_pci_notify_cap,
526 notify_off_multiplier));
527 BUILD_BUG_ON(VIRTIO_PCI_COMMON_DFSELECT !=
528 offsetof(struct virtio_pci_common_cfg,
529 device_feature_select));
530 BUILD_BUG_ON(VIRTIO_PCI_COMMON_DF !=
531 offsetof(struct virtio_pci_common_cfg, device_feature));
532 BUILD_BUG_ON(VIRTIO_PCI_COMMON_GFSELECT !=
533 offsetof(struct virtio_pci_common_cfg,
534 guest_feature_select));
535 BUILD_BUG_ON(VIRTIO_PCI_COMMON_GF !=
536 offsetof(struct virtio_pci_common_cfg, guest_feature));
537 BUILD_BUG_ON(VIRTIO_PCI_COMMON_MSIX !=
538 offsetof(struct virtio_pci_common_cfg, msix_config));
539 BUILD_BUG_ON(VIRTIO_PCI_COMMON_NUMQ !=
540 offsetof(struct virtio_pci_common_cfg, num_queues));
541 BUILD_BUG_ON(VIRTIO_PCI_COMMON_STATUS !=
542 offsetof(struct virtio_pci_common_cfg, device_status));
543 BUILD_BUG_ON(VIRTIO_PCI_COMMON_CFGGENERATION !=
544 offsetof(struct virtio_pci_common_cfg, config_generation));
545 BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_SELECT !=
546 offsetof(struct virtio_pci_common_cfg, queue_select));
547 BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_SIZE !=
548 offsetof(struct virtio_pci_common_cfg, queue_size));
549 BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_MSIX !=
550 offsetof(struct virtio_pci_common_cfg, queue_msix_vector));
551 BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_ENABLE !=
552 offsetof(struct virtio_pci_common_cfg, queue_enable));
553 BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_NOFF !=
554 offsetof(struct virtio_pci_common_cfg, queue_notify_off));
555 BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_DESCLO !=
556 offsetof(struct virtio_pci_common_cfg, queue_desc_lo));
557 BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_DESCHI !=
558 offsetof(struct virtio_pci_common_cfg, queue_desc_hi));
559 BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_AVAILLO !=
560 offsetof(struct virtio_pci_common_cfg, queue_avail_lo));
561 BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_AVAILHI !=
562 offsetof(struct virtio_pci_common_cfg, queue_avail_hi));
563 BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_USEDLO !=
564 offsetof(struct virtio_pci_common_cfg, queue_used_lo));
565 BUILD_BUG_ON(VIRTIO_PCI_COMMON_Q_USEDHI !=
566 offsetof(struct virtio_pci_common_cfg, queue_used_hi));
569 /* the PCI probing function */
570 int virtio_pci_modern_probe(struct virtio_pci_device *vp_dev)
572 struct pci_dev *pci_dev = vp_dev->pci_dev;
573 int err, common, isr, notify, device;
579 /* We only own devices >= 0x1000 and <= 0x107f: leave the rest. */
580 if (pci_dev->device < 0x1000 || pci_dev->device > 0x107f)
583 if (pci_dev->device < 0x1040) {
584 /* Transitional devices: use the PCI subsystem device id as
585 * virtio device id, same as legacy driver always did.
587 vp_dev->vdev.id.device = pci_dev->subsystem_device;
589 /* Modern devices: simply use PCI device id, but start from 0x1040. */
590 vp_dev->vdev.id.device = pci_dev->device - 0x1040;
592 vp_dev->vdev.id.vendor = pci_dev->subsystem_vendor;
594 /* check for a common config: if not, use legacy mode (bar 0). */
595 common = virtio_pci_find_capability(pci_dev, VIRTIO_PCI_CAP_COMMON_CFG,
596 IORESOURCE_IO | IORESOURCE_MEM,
597 &vp_dev->modern_bars);
599 dev_info(&pci_dev->dev,
600 "virtio_pci: leaving for legacy driver\n");
604 /* If common is there, these should be too... */
605 isr = virtio_pci_find_capability(pci_dev, VIRTIO_PCI_CAP_ISR_CFG,
606 IORESOURCE_IO | IORESOURCE_MEM,
607 &vp_dev->modern_bars);
608 notify = virtio_pci_find_capability(pci_dev, VIRTIO_PCI_CAP_NOTIFY_CFG,
609 IORESOURCE_IO | IORESOURCE_MEM,
610 &vp_dev->modern_bars);
611 if (!isr || !notify) {
612 dev_err(&pci_dev->dev,
613 "virtio_pci: missing capabilities %i/%i/%i\n",
614 common, isr, notify);
618 err = dma_set_mask_and_coherent(&pci_dev->dev, DMA_BIT_MASK(64));
620 err = dma_set_mask_and_coherent(&pci_dev->dev,
623 dev_warn(&pci_dev->dev, "Failed to enable 64-bit or 32-bit DMA. Trying to continue, but this might not work.\n");
625 /* Device capability is only mandatory for devices that have
626 * device-specific configuration.
628 device = virtio_pci_find_capability(pci_dev, VIRTIO_PCI_CAP_DEVICE_CFG,
629 IORESOURCE_IO | IORESOURCE_MEM,
630 &vp_dev->modern_bars);
632 err = pci_request_selected_regions(pci_dev, vp_dev->modern_bars,
633 "virtio-pci-modern");
638 vp_dev->common = map_capability(pci_dev, common,
639 sizeof(struct virtio_pci_common_cfg), 4,
640 0, sizeof(struct virtio_pci_common_cfg),
644 vp_dev->isr = map_capability(pci_dev, isr, sizeof(u8), 1,
650 /* Read notify_off_multiplier from config space. */
651 pci_read_config_dword(pci_dev,
652 notify + offsetof(struct virtio_pci_notify_cap,
653 notify_off_multiplier),
654 &vp_dev->notify_offset_multiplier);
655 /* Read notify length and offset from config space. */
656 pci_read_config_dword(pci_dev,
657 notify + offsetof(struct virtio_pci_notify_cap,
661 pci_read_config_dword(pci_dev,
662 notify + offsetof(struct virtio_pci_notify_cap,
666 /* We don't know how many VQs we'll map, ahead of the time.
667 * If notify length is small, map it all now.
668 * Otherwise, map each VQ individually later.
670 if ((u64)notify_length + (notify_offset % PAGE_SIZE) <= PAGE_SIZE) {
671 vp_dev->notify_base = map_capability(pci_dev, notify, 2, 2,
673 &vp_dev->notify_len);
674 if (!vp_dev->notify_base)
677 vp_dev->notify_map_cap = notify;
680 /* Again, we don't know how much we should map, but PAGE_SIZE
681 * is more than enough for all existing devices.
684 vp_dev->device = map_capability(pci_dev, device, 0, 4,
686 &vp_dev->device_len);
690 vp_dev->vdev.config = &virtio_pci_config_ops;
692 vp_dev->vdev.config = &virtio_pci_config_nodev_ops;
695 vp_dev->config_vector = vp_config_vector;
696 vp_dev->setup_vq = setup_vq;
697 vp_dev->del_vq = del_vq;
702 if (vp_dev->notify_base)
703 pci_iounmap(pci_dev, vp_dev->notify_base);
705 pci_iounmap(pci_dev, vp_dev->isr);
707 pci_iounmap(pci_dev, vp_dev->common);
712 void virtio_pci_modern_remove(struct virtio_pci_device *vp_dev)
714 struct pci_dev *pci_dev = vp_dev->pci_dev;
717 pci_iounmap(pci_dev, vp_dev->device);
718 if (vp_dev->notify_base)
719 pci_iounmap(pci_dev, vp_dev->notify_base);
720 pci_iounmap(pci_dev, vp_dev->isr);
721 pci_iounmap(pci_dev, vp_dev->common);
722 pci_release_selected_regions(pci_dev, vp_dev->modern_bars);