2 * Copyright (C) 2012 Red Hat, Inc. All rights reserved.
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License version 2 as
7 * published by the Free Software Foundation.
9 * Derived from original vfio:
10 * Copyright 2010 Cisco Systems, Inc. All rights reserved.
14 #include <linux/device.h>
15 #include <linux/eventfd.h>
16 #include <linux/file.h>
17 #include <linux/interrupt.h>
18 #include <linux/iommu.h>
19 #include <linux/module.h>
20 #include <linux/mutex.h>
21 #include <linux/notifier.h>
22 #include <linux/pci.h>
23 #include <linux/pm_runtime.h>
24 #include <linux/slab.h>
25 #include <linux/types.h>
26 #include <linux/uaccess.h>
27 #include <linux/vfio.h>
29 #include "vfio_pci_private.h"
31 #define DRIVER_VERSION "0.2"
33 #define DRIVER_DESC "VFIO PCI - User Level meta-driver"
35 static bool nointxmask;
36 module_param_named(nointxmask, nointxmask, bool, S_IRUGO | S_IWUSR);
37 MODULE_PARM_DESC(nointxmask,
38 "Disable support for PCI 2.3 style INTx masking. If this resolves problems for specific devices, report lspci -vvvxxx to
[email protected] so the device can be fixed automatically via the broken_intx_masking flag.");
40 static DEFINE_MUTEX(driver_lock);
42 static void vfio_pci_try_bus_reset(struct vfio_pci_device *vdev);
44 static int vfio_pci_enable(struct vfio_pci_device *vdev)
46 struct pci_dev *pdev = vdev->pdev;
51 /* Don't allow our initial saved state to include busmaster */
52 pci_clear_master(pdev);
54 ret = pci_enable_device(pdev);
58 vdev->reset_works = (pci_reset_function(pdev) == 0);
60 vdev->pci_saved_state = pci_store_saved_state(pdev);
61 if (!vdev->pci_saved_state)
62 pr_debug("%s: Couldn't store %s saved state\n",
63 __func__, dev_name(&pdev->dev));
65 ret = vfio_config_init(vdev);
67 kfree(vdev->pci_saved_state);
68 vdev->pci_saved_state = NULL;
69 pci_disable_device(pdev);
73 if (likely(!nointxmask))
74 vdev->pci_2_3 = pci_intx_mask_supported(pdev);
76 pci_read_config_word(pdev, PCI_COMMAND, &cmd);
77 if (vdev->pci_2_3 && (cmd & PCI_COMMAND_INTX_DISABLE)) {
78 cmd &= ~PCI_COMMAND_INTX_DISABLE;
79 pci_write_config_word(pdev, PCI_COMMAND, cmd);
82 msix_pos = pdev->msix_cap;
87 pci_read_config_word(pdev, msix_pos + PCI_MSIX_FLAGS, &flags);
88 pci_read_config_dword(pdev, msix_pos + PCI_MSIX_TABLE, &table);
90 vdev->msix_bar = table & PCI_MSIX_TABLE_BIR;
91 vdev->msix_offset = table & PCI_MSIX_TABLE_OFFSET;
92 vdev->msix_size = ((flags & PCI_MSIX_FLAGS_QSIZE) + 1) * 16;
94 vdev->msix_bar = 0xFF;
96 #ifdef CONFIG_VFIO_PCI_VGA
97 if ((pdev->class >> 8) == PCI_CLASS_DISPLAY_VGA)
104 static void vfio_pci_disable(struct vfio_pci_device *vdev)
106 struct pci_dev *pdev = vdev->pdev;
109 /* Stop the device from further DMA */
110 pci_clear_master(pdev);
112 vfio_pci_set_irqs_ioctl(vdev, VFIO_IRQ_SET_DATA_NONE |
113 VFIO_IRQ_SET_ACTION_TRIGGER,
114 vdev->irq_type, 0, 0, NULL);
116 vdev->virq_disabled = false;
118 vfio_config_free(vdev);
120 for (bar = PCI_STD_RESOURCES; bar <= PCI_STD_RESOURCE_END; bar++) {
121 if (!vdev->barmap[bar])
123 pci_iounmap(pdev, vdev->barmap[bar]);
124 pci_release_selected_regions(pdev, 1 << bar);
125 vdev->barmap[bar] = NULL;
128 vdev->needs_reset = true;
131 * If we have saved state, restore it. If we can reset the device,
132 * even better. Resetting with current state seems better than
133 * nothing, but saving and restoring current state without reset
136 if (pci_load_and_free_saved_state(pdev, &vdev->pci_saved_state)) {
137 pr_info("%s: Couldn't reload %s saved state\n",
138 __func__, dev_name(&pdev->dev));
140 if (!vdev->reset_works)
143 pci_save_state(pdev);
147 * Disable INTx and MSI, presumably to avoid spurious interrupts
148 * during reset. Stolen from pci_reset_function()
150 pci_write_config_word(pdev, PCI_COMMAND, PCI_COMMAND_INTX_DISABLE);
153 * Try to reset the device. The success of this is dependent on
154 * being able to lock the device, which is not always possible.
156 if (vdev->reset_works) {
157 int ret = pci_try_reset_function(pdev);
159 pr_warn("%s: Failed to reset device %s (%d)\n",
160 __func__, dev_name(&pdev->dev), ret);
162 vdev->needs_reset = false;
165 pci_restore_state(pdev);
167 pci_disable_device(pdev);
169 vfio_pci_try_bus_reset(vdev);
172 static void vfio_pci_release(void *device_data)
174 struct vfio_pci_device *vdev = device_data;
176 mutex_lock(&driver_lock);
178 if (!(--vdev->refcnt)) {
179 vfio_spapr_pci_eeh_release(vdev->pdev);
180 vfio_pci_disable(vdev);
183 mutex_unlock(&driver_lock);
185 module_put(THIS_MODULE);
188 static int vfio_pci_open(void *device_data)
190 struct vfio_pci_device *vdev = device_data;
193 if (!try_module_get(THIS_MODULE))
196 mutex_lock(&driver_lock);
199 ret = vfio_pci_enable(vdev);
203 vfio_spapr_pci_eeh_open(vdev->pdev);
207 mutex_unlock(&driver_lock);
209 module_put(THIS_MODULE);
213 static int vfio_pci_get_irq_count(struct vfio_pci_device *vdev, int irq_type)
215 if (irq_type == VFIO_PCI_INTX_IRQ_INDEX) {
217 pci_read_config_byte(vdev->pdev, PCI_INTERRUPT_PIN, &pin);
218 if (IS_ENABLED(CONFIG_VFIO_PCI_INTX) && pin)
221 } else if (irq_type == VFIO_PCI_MSI_IRQ_INDEX) {
225 pos = vdev->pdev->msi_cap;
227 pci_read_config_word(vdev->pdev,
228 pos + PCI_MSI_FLAGS, &flags);
229 return 1 << ((flags & PCI_MSI_FLAGS_QMASK) >> 1);
231 } else if (irq_type == VFIO_PCI_MSIX_IRQ_INDEX) {
235 pos = vdev->pdev->msix_cap;
237 pci_read_config_word(vdev->pdev,
238 pos + PCI_MSIX_FLAGS, &flags);
240 return (flags & PCI_MSIX_FLAGS_QSIZE) + 1;
242 } else if (irq_type == VFIO_PCI_ERR_IRQ_INDEX) {
243 if (pci_is_pcie(vdev->pdev))
245 } else if (irq_type == VFIO_PCI_REQ_IRQ_INDEX) {
252 static int vfio_pci_count_devs(struct pci_dev *pdev, void *data)
258 struct vfio_pci_fill_info {
261 struct vfio_pci_dependent_device *devices;
264 static int vfio_pci_fill_devs(struct pci_dev *pdev, void *data)
266 struct vfio_pci_fill_info *fill = data;
267 struct iommu_group *iommu_group;
269 if (fill->cur == fill->max)
270 return -EAGAIN; /* Something changed, try again */
272 iommu_group = iommu_group_get(&pdev->dev);
274 return -EPERM; /* Cannot reset non-isolated devices */
276 fill->devices[fill->cur].group_id = iommu_group_id(iommu_group);
277 fill->devices[fill->cur].segment = pci_domain_nr(pdev->bus);
278 fill->devices[fill->cur].bus = pdev->bus->number;
279 fill->devices[fill->cur].devfn = pdev->devfn;
281 iommu_group_put(iommu_group);
285 struct vfio_pci_group_entry {
286 struct vfio_group *group;
290 struct vfio_pci_group_info {
292 struct vfio_pci_group_entry *groups;
295 static int vfio_pci_validate_devs(struct pci_dev *pdev, void *data)
297 struct vfio_pci_group_info *info = data;
298 struct iommu_group *group;
301 group = iommu_group_get(&pdev->dev);
305 id = iommu_group_id(group);
307 for (i = 0; i < info->count; i++)
308 if (info->groups[i].id == id)
311 iommu_group_put(group);
313 return (i == info->count) ? -EINVAL : 0;
316 static bool vfio_pci_dev_below_slot(struct pci_dev *pdev, struct pci_slot *slot)
318 for (; pdev; pdev = pdev->bus->self)
319 if (pdev->bus == slot->bus)
320 return (pdev->slot == slot);
324 struct vfio_pci_walk_info {
325 int (*fn)(struct pci_dev *, void *data);
327 struct pci_dev *pdev;
332 static int vfio_pci_walk_wrapper(struct pci_dev *pdev, void *data)
334 struct vfio_pci_walk_info *walk = data;
336 if (!walk->slot || vfio_pci_dev_below_slot(pdev, walk->pdev->slot))
337 walk->ret = walk->fn(pdev, walk->data);
342 static int vfio_pci_for_each_slot_or_bus(struct pci_dev *pdev,
343 int (*fn)(struct pci_dev *,
344 void *data), void *data,
347 struct vfio_pci_walk_info walk = {
348 .fn = fn, .data = data, .pdev = pdev, .slot = slot, .ret = 0,
351 pci_walk_bus(pdev->bus, vfio_pci_walk_wrapper, &walk);
356 static long vfio_pci_ioctl(void *device_data,
357 unsigned int cmd, unsigned long arg)
359 struct vfio_pci_device *vdev = device_data;
362 if (cmd == VFIO_DEVICE_GET_INFO) {
363 struct vfio_device_info info;
365 minsz = offsetofend(struct vfio_device_info, num_irqs);
367 if (copy_from_user(&info, (void __user *)arg, minsz))
370 if (info.argsz < minsz)
373 info.flags = VFIO_DEVICE_FLAGS_PCI;
375 if (vdev->reset_works)
376 info.flags |= VFIO_DEVICE_FLAGS_RESET;
378 info.num_regions = VFIO_PCI_NUM_REGIONS;
379 info.num_irqs = VFIO_PCI_NUM_IRQS;
381 return copy_to_user((void __user *)arg, &info, minsz);
383 } else if (cmd == VFIO_DEVICE_GET_REGION_INFO) {
384 struct pci_dev *pdev = vdev->pdev;
385 struct vfio_region_info info;
387 minsz = offsetofend(struct vfio_region_info, offset);
389 if (copy_from_user(&info, (void __user *)arg, minsz))
392 if (info.argsz < minsz)
395 switch (info.index) {
396 case VFIO_PCI_CONFIG_REGION_INDEX:
397 info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index);
398 info.size = pdev->cfg_size;
399 info.flags = VFIO_REGION_INFO_FLAG_READ |
400 VFIO_REGION_INFO_FLAG_WRITE;
402 case VFIO_PCI_BAR0_REGION_INDEX ... VFIO_PCI_BAR5_REGION_INDEX:
403 info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index);
404 info.size = pci_resource_len(pdev, info.index);
410 info.flags = VFIO_REGION_INFO_FLAG_READ |
411 VFIO_REGION_INFO_FLAG_WRITE;
412 if (IS_ENABLED(CONFIG_VFIO_PCI_MMAP) &&
413 pci_resource_flags(pdev, info.index) &
414 IORESOURCE_MEM && info.size >= PAGE_SIZE)
415 info.flags |= VFIO_REGION_INFO_FLAG_MMAP;
417 case VFIO_PCI_ROM_REGION_INDEX:
422 info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index);
425 /* Report the BAR size, not the ROM size */
426 info.size = pci_resource_len(pdev, info.index);
430 /* Is it really there? */
431 io = pci_map_rom(pdev, &size);
436 pci_unmap_rom(pdev, io);
438 info.flags = VFIO_REGION_INFO_FLAG_READ;
441 case VFIO_PCI_VGA_REGION_INDEX:
445 info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index);
447 info.flags = VFIO_REGION_INFO_FLAG_READ |
448 VFIO_REGION_INFO_FLAG_WRITE;
455 return copy_to_user((void __user *)arg, &info, minsz);
457 } else if (cmd == VFIO_DEVICE_GET_IRQ_INFO) {
458 struct vfio_irq_info info;
460 minsz = offsetofend(struct vfio_irq_info, count);
462 if (copy_from_user(&info, (void __user *)arg, minsz))
465 if (info.argsz < minsz || info.index >= VFIO_PCI_NUM_IRQS)
468 switch (info.index) {
469 case VFIO_PCI_INTX_IRQ_INDEX ... VFIO_PCI_MSIX_IRQ_INDEX:
470 case VFIO_PCI_REQ_IRQ_INDEX:
472 case VFIO_PCI_ERR_IRQ_INDEX:
473 if (pci_is_pcie(vdev->pdev))
475 /* pass thru to return error */
480 info.flags = VFIO_IRQ_INFO_EVENTFD;
482 info.count = vfio_pci_get_irq_count(vdev, info.index);
484 if (info.index == VFIO_PCI_INTX_IRQ_INDEX)
485 info.flags |= (VFIO_IRQ_INFO_MASKABLE |
486 VFIO_IRQ_INFO_AUTOMASKED);
488 info.flags |= VFIO_IRQ_INFO_NORESIZE;
490 return copy_to_user((void __user *)arg, &info, minsz);
492 } else if (cmd == VFIO_DEVICE_SET_IRQS) {
493 struct vfio_irq_set hdr;
497 minsz = offsetofend(struct vfio_irq_set, count);
499 if (copy_from_user(&hdr, (void __user *)arg, minsz))
502 if (hdr.argsz < minsz || hdr.index >= VFIO_PCI_NUM_IRQS ||
503 hdr.flags & ~(VFIO_IRQ_SET_DATA_TYPE_MASK |
504 VFIO_IRQ_SET_ACTION_TYPE_MASK))
507 if (!(hdr.flags & VFIO_IRQ_SET_DATA_NONE)) {
509 int max = vfio_pci_get_irq_count(vdev, hdr.index);
511 if (hdr.flags & VFIO_IRQ_SET_DATA_BOOL)
512 size = sizeof(uint8_t);
513 else if (hdr.flags & VFIO_IRQ_SET_DATA_EVENTFD)
514 size = sizeof(int32_t);
518 if (hdr.argsz - minsz < hdr.count * size ||
519 hdr.start >= max || hdr.start + hdr.count > max)
522 data = memdup_user((void __user *)(arg + minsz),
525 return PTR_ERR(data);
528 mutex_lock(&vdev->igate);
530 ret = vfio_pci_set_irqs_ioctl(vdev, hdr.flags, hdr.index,
531 hdr.start, hdr.count, data);
533 mutex_unlock(&vdev->igate);
538 } else if (cmd == VFIO_DEVICE_RESET) {
539 return vdev->reset_works ?
540 pci_try_reset_function(vdev->pdev) : -EINVAL;
542 } else if (cmd == VFIO_DEVICE_GET_PCI_HOT_RESET_INFO) {
543 struct vfio_pci_hot_reset_info hdr;
544 struct vfio_pci_fill_info fill = { 0 };
545 struct vfio_pci_dependent_device *devices = NULL;
549 minsz = offsetofend(struct vfio_pci_hot_reset_info, count);
551 if (copy_from_user(&hdr, (void __user *)arg, minsz))
554 if (hdr.argsz < minsz)
559 /* Can we do a slot or bus reset or neither? */
560 if (!pci_probe_reset_slot(vdev->pdev->slot))
562 else if (pci_probe_reset_bus(vdev->pdev->bus))
565 /* How many devices are affected? */
566 ret = vfio_pci_for_each_slot_or_bus(vdev->pdev,
572 WARN_ON(!fill.max); /* Should always be at least one */
575 * If there's enough space, fill it now, otherwise return
576 * -ENOSPC and the number of devices affected.
578 if (hdr.argsz < sizeof(hdr) + (fill.max * sizeof(*devices))) {
580 hdr.count = fill.max;
581 goto reset_info_exit;
584 devices = kcalloc(fill.max, sizeof(*devices), GFP_KERNEL);
588 fill.devices = devices;
590 ret = vfio_pci_for_each_slot_or_bus(vdev->pdev,
595 * If a device was removed between counting and filling,
596 * we may come up short of fill.max. If a device was
597 * added, we'll have a return of -EAGAIN above.
600 hdr.count = fill.cur;
603 if (copy_to_user((void __user *)arg, &hdr, minsz))
607 if (copy_to_user((void __user *)(arg + minsz), devices,
608 hdr.count * sizeof(*devices)))
615 } else if (cmd == VFIO_DEVICE_PCI_HOT_RESET) {
616 struct vfio_pci_hot_reset hdr;
618 struct vfio_pci_group_entry *groups;
619 struct vfio_pci_group_info info;
621 int i, count = 0, ret = 0;
623 minsz = offsetofend(struct vfio_pci_hot_reset, count);
625 if (copy_from_user(&hdr, (void __user *)arg, minsz))
628 if (hdr.argsz < minsz || hdr.flags)
631 /* Can we do a slot or bus reset or neither? */
632 if (!pci_probe_reset_slot(vdev->pdev->slot))
634 else if (pci_probe_reset_bus(vdev->pdev->bus))
638 * We can't let userspace give us an arbitrarily large
639 * buffer to copy, so verify how many we think there
640 * could be. Note groups can have multiple devices so
641 * one group per device is the max.
643 ret = vfio_pci_for_each_slot_or_bus(vdev->pdev,
649 /* Somewhere between 1 and count is OK */
650 if (!hdr.count || hdr.count > count)
653 group_fds = kcalloc(hdr.count, sizeof(*group_fds), GFP_KERNEL);
654 groups = kcalloc(hdr.count, sizeof(*groups), GFP_KERNEL);
655 if (!group_fds || !groups) {
661 if (copy_from_user(group_fds, (void __user *)(arg + minsz),
662 hdr.count * sizeof(*group_fds))) {
669 * For each group_fd, get the group through the vfio external
670 * user interface and store the group and iommu ID. This
671 * ensures the group is held across the reset.
673 for (i = 0; i < hdr.count; i++) {
674 struct vfio_group *group;
675 struct fd f = fdget(group_fds[i]);
681 group = vfio_group_get_external_user(f.file);
684 ret = PTR_ERR(group);
688 groups[i].group = group;
689 groups[i].id = vfio_external_user_iommu_id(group);
694 /* release reference to groups on error */
696 goto hot_reset_release;
698 info.count = hdr.count;
699 info.groups = groups;
702 * Test whether all the affected devices are contained
703 * by the set of groups provided by the user.
705 ret = vfio_pci_for_each_slot_or_bus(vdev->pdev,
706 vfio_pci_validate_devs,
709 /* User has access, do the reset */
710 ret = slot ? pci_try_reset_slot(vdev->pdev->slot) :
711 pci_try_reset_bus(vdev->pdev->bus);
714 for (i--; i >= 0; i--)
715 vfio_group_put_external_user(groups[i].group);
724 static ssize_t vfio_pci_rw(void *device_data, char __user *buf,
725 size_t count, loff_t *ppos, bool iswrite)
727 unsigned int index = VFIO_PCI_OFFSET_TO_INDEX(*ppos);
728 struct vfio_pci_device *vdev = device_data;
730 if (index >= VFIO_PCI_NUM_REGIONS)
734 case VFIO_PCI_CONFIG_REGION_INDEX:
735 return vfio_pci_config_rw(vdev, buf, count, ppos, iswrite);
737 case VFIO_PCI_ROM_REGION_INDEX:
740 return vfio_pci_bar_rw(vdev, buf, count, ppos, false);
742 case VFIO_PCI_BAR0_REGION_INDEX ... VFIO_PCI_BAR5_REGION_INDEX:
743 return vfio_pci_bar_rw(vdev, buf, count, ppos, iswrite);
745 case VFIO_PCI_VGA_REGION_INDEX:
746 return vfio_pci_vga_rw(vdev, buf, count, ppos, iswrite);
752 static ssize_t vfio_pci_read(void *device_data, char __user *buf,
753 size_t count, loff_t *ppos)
758 return vfio_pci_rw(device_data, buf, count, ppos, false);
761 static ssize_t vfio_pci_write(void *device_data, const char __user *buf,
762 size_t count, loff_t *ppos)
767 return vfio_pci_rw(device_data, (char __user *)buf, count, ppos, true);
770 static int vfio_pci_mmap(void *device_data, struct vm_area_struct *vma)
772 struct vfio_pci_device *vdev = device_data;
773 struct pci_dev *pdev = vdev->pdev;
775 u64 phys_len, req_len, pgoff, req_start;
778 index = vma->vm_pgoff >> (VFIO_PCI_OFFSET_SHIFT - PAGE_SHIFT);
780 if (vma->vm_end < vma->vm_start)
782 if ((vma->vm_flags & VM_SHARED) == 0)
784 if (index >= VFIO_PCI_ROM_REGION_INDEX)
786 if (!(pci_resource_flags(pdev, index) & IORESOURCE_MEM))
789 phys_len = pci_resource_len(pdev, index);
790 req_len = vma->vm_end - vma->vm_start;
791 pgoff = vma->vm_pgoff &
792 ((1U << (VFIO_PCI_OFFSET_SHIFT - PAGE_SHIFT)) - 1);
793 req_start = pgoff << PAGE_SHIFT;
795 if (phys_len < PAGE_SIZE || req_start + req_len > phys_len)
798 if (index == vdev->msix_bar) {
800 * Disallow mmaps overlapping the MSI-X table; users don't
801 * get to touch this directly. We could find somewhere
802 * else to map the overlap, but page granularity is only
803 * a recommendation, not a requirement, so the user needs
804 * to know which bits are real. Requiring them to mmap
805 * around the table makes that clear.
808 /* If neither entirely above nor below, then it overlaps */
809 if (!(req_start >= vdev->msix_offset + vdev->msix_size ||
810 req_start + req_len <= vdev->msix_offset))
815 * Even though we don't make use of the barmap for the mmap,
816 * we need to request the region and the barmap tracks that.
818 if (!vdev->barmap[index]) {
819 ret = pci_request_selected_regions(pdev,
820 1 << index, "vfio-pci");
824 vdev->barmap[index] = pci_iomap(pdev, index, 0);
827 vma->vm_private_data = vdev;
828 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
829 vma->vm_pgoff = (pci_resource_start(pdev, index) >> PAGE_SHIFT) + pgoff;
831 return remap_pfn_range(vma, vma->vm_start, vma->vm_pgoff,
832 req_len, vma->vm_page_prot);
835 static void vfio_pci_request(void *device_data, unsigned int count)
837 struct vfio_pci_device *vdev = device_data;
839 mutex_lock(&vdev->igate);
841 if (vdev->req_trigger) {
842 dev_dbg(&vdev->pdev->dev, "Requesting device from user\n");
843 eventfd_signal(vdev->req_trigger, 1);
846 mutex_unlock(&vdev->igate);
849 static const struct vfio_device_ops vfio_pci_ops = {
851 .open = vfio_pci_open,
852 .release = vfio_pci_release,
853 .ioctl = vfio_pci_ioctl,
854 .read = vfio_pci_read,
855 .write = vfio_pci_write,
856 .mmap = vfio_pci_mmap,
857 .request = vfio_pci_request,
860 static int vfio_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)
862 struct vfio_pci_device *vdev;
863 struct iommu_group *group;
866 if (pdev->hdr_type != PCI_HEADER_TYPE_NORMAL)
869 group = iommu_group_get(&pdev->dev);
873 vdev = kzalloc(sizeof(*vdev), GFP_KERNEL);
875 iommu_group_put(group);
880 vdev->irq_type = VFIO_PCI_NUM_IRQS;
881 mutex_init(&vdev->igate);
882 spin_lock_init(&vdev->irqlock);
884 ret = vfio_add_group_dev(&pdev->dev, &vfio_pci_ops, vdev);
886 iommu_group_put(group);
893 static void vfio_pci_remove(struct pci_dev *pdev)
895 struct vfio_pci_device *vdev;
897 vdev = vfio_del_group_dev(&pdev->dev);
899 iommu_group_put(pdev->dev.iommu_group);
904 static pci_ers_result_t vfio_pci_aer_err_detected(struct pci_dev *pdev,
905 pci_channel_state_t state)
907 struct vfio_pci_device *vdev;
908 struct vfio_device *device;
910 device = vfio_device_get_from_dev(&pdev->dev);
912 return PCI_ERS_RESULT_DISCONNECT;
914 vdev = vfio_device_data(device);
916 vfio_device_put(device);
917 return PCI_ERS_RESULT_DISCONNECT;
920 mutex_lock(&vdev->igate);
922 if (vdev->err_trigger)
923 eventfd_signal(vdev->err_trigger, 1);
925 mutex_unlock(&vdev->igate);
927 vfio_device_put(device);
929 return PCI_ERS_RESULT_CAN_RECOVER;
932 static struct pci_error_handlers vfio_err_handlers = {
933 .error_detected = vfio_pci_aer_err_detected,
936 static struct pci_driver vfio_pci_driver = {
938 .id_table = NULL, /* only dynamic ids */
939 .probe = vfio_pci_probe,
940 .remove = vfio_pci_remove,
941 .err_handler = &vfio_err_handlers,
944 struct vfio_devices {
945 struct vfio_device **devices;
950 static int vfio_pci_get_devs(struct pci_dev *pdev, void *data)
952 struct vfio_devices *devs = data;
953 struct pci_driver *pci_drv = ACCESS_ONCE(pdev->driver);
955 if (pci_drv != &vfio_pci_driver)
958 if (devs->cur_index == devs->max_index)
961 devs->devices[devs->cur_index] = vfio_device_get_from_dev(&pdev->dev);
962 if (!devs->devices[devs->cur_index])
970 * Attempt to do a bus/slot reset if there are devices affected by a reset for
971 * this device that are needs_reset and all of the affected devices are unused
972 * (!refcnt). Callers are required to hold driver_lock when calling this to
973 * prevent device opens and concurrent bus reset attempts. We prevent device
974 * unbinds by acquiring and holding a reference to the vfio_device.
976 * NB: vfio-core considers a group to be viable even if some devices are
977 * bound to drivers like pci-stub or pcieport. Here we require all devices
978 * to be bound to vfio_pci since that's the only way we can be sure they
981 static void vfio_pci_try_bus_reset(struct vfio_pci_device *vdev)
983 struct vfio_devices devs = { .cur_index = 0 };
984 int i = 0, ret = -EINVAL;
985 bool needs_reset = false, slot = false;
986 struct vfio_pci_device *tmp;
988 if (!pci_probe_reset_slot(vdev->pdev->slot))
990 else if (pci_probe_reset_bus(vdev->pdev->bus))
993 if (vfio_pci_for_each_slot_or_bus(vdev->pdev, vfio_pci_count_devs,
998 devs.devices = kcalloc(i, sizeof(struct vfio_device *), GFP_KERNEL);
1002 if (vfio_pci_for_each_slot_or_bus(vdev->pdev,
1003 vfio_pci_get_devs, &devs, slot))
1006 for (i = 0; i < devs.cur_index; i++) {
1007 tmp = vfio_device_data(devs.devices[i]);
1008 if (tmp->needs_reset)
1015 ret = slot ? pci_try_reset_slot(vdev->pdev->slot) :
1016 pci_try_reset_bus(vdev->pdev->bus);
1019 for (i = 0; i < devs.cur_index; i++) {
1021 tmp = vfio_device_data(devs.devices[i]);
1022 tmp->needs_reset = false;
1024 vfio_device_put(devs.devices[i]);
1027 kfree(devs.devices);
1030 static void __exit vfio_pci_cleanup(void)
1032 pci_unregister_driver(&vfio_pci_driver);
1033 vfio_pci_virqfd_exit();
1034 vfio_pci_uninit_perm_bits();
1037 static int __init vfio_pci_init(void)
1041 /* Allocate shared config space permision data used by all devices */
1042 ret = vfio_pci_init_perm_bits();
1046 /* Start the virqfd cleanup handler */
1047 ret = vfio_pci_virqfd_init();
1051 /* Register and scan for devices */
1052 ret = pci_register_driver(&vfio_pci_driver);
1059 vfio_pci_virqfd_exit();
1061 vfio_pci_uninit_perm_bits();
1065 module_init(vfio_pci_init);
1066 module_exit(vfio_pci_cleanup);
1068 MODULE_VERSION(DRIVER_VERSION);
1069 MODULE_LICENSE("GPL v2");
1070 MODULE_AUTHOR(DRIVER_AUTHOR);
1071 MODULE_DESCRIPTION(DRIVER_DESC);