2 * Copyright (C) 2012 Red Hat, Inc. All rights reserved.
3 * Author: Alex Williamson <alex.williamson@redhat.com>
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.
11 * Author: Tom Lyon, pugs@cisco.com
14 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
16 #include <linux/device.h>
17 #include <linux/eventfd.h>
18 #include <linux/file.h>
19 #include <linux/interrupt.h>
20 #include <linux/iommu.h>
21 #include <linux/module.h>
22 #include <linux/mutex.h>
23 #include <linux/notifier.h>
24 #include <linux/pci.h>
25 #include <linux/pm_runtime.h>
26 #include <linux/slab.h>
27 #include <linux/types.h>
28 #include <linux/uaccess.h>
29 #include <linux/vfio.h>
30 #include <linux/vgaarb.h>
32 #include "vfio_pci_private.h"
34 #define DRIVER_VERSION "0.2"
35 #define DRIVER_AUTHOR "Alex Williamson <alex.williamson@redhat.com>"
36 #define DRIVER_DESC "VFIO PCI - User Level meta-driver"
38 static char ids[1024] __initdata;
39 module_param_string(ids, ids, sizeof(ids), 0);
40 MODULE_PARM_DESC(ids, "Initial PCI IDs to add to the vfio driver, format is \"vendor:device[:subvendor[:subdevice[:class[:class_mask]]]]\" and multiple comma separated entries can be specified");
42 static bool nointxmask;
43 module_param_named(nointxmask, nointxmask, bool, S_IRUGO | S_IWUSR);
44 MODULE_PARM_DESC(nointxmask,
45 "Disable support for PCI 2.3 style INTx masking. If this resolves problems for specific devices, report lspci -vvvxxx to linux-pci@vger.kernel.org so the device can be fixed automatically via the broken_intx_masking flag.");
47 #ifdef CONFIG_VFIO_PCI_VGA
48 static bool disable_vga;
49 module_param(disable_vga, bool, S_IRUGO);
50 MODULE_PARM_DESC(disable_vga, "Disable VGA resource access through vfio-pci");
53 static bool disable_idle_d3;
54 module_param(disable_idle_d3, bool, S_IRUGO | S_IWUSR);
55 MODULE_PARM_DESC(disable_idle_d3,
56 "Disable using the PCI D3 low power state for idle, unused devices");
58 static DEFINE_MUTEX(driver_lock);
60 static inline bool vfio_vga_disabled(void)
62 #ifdef CONFIG_VFIO_PCI_VGA
70 * Our VGA arbiter participation is limited since we don't know anything
71 * about the device itself. However, if the device is the only VGA device
72 * downstream of a bridge and VFIO VGA support is disabled, then we can
73 * safely return legacy VGA IO and memory as not decoded since the user
74 * has no way to get to it and routing can be disabled externally at the
77 static unsigned int vfio_pci_set_vga_decode(void *opaque, bool single_vga)
79 struct vfio_pci_device *vdev = opaque;
80 struct pci_dev *tmp = NULL, *pdev = vdev->pdev;
81 unsigned char max_busnr;
84 if (single_vga || !vfio_vga_disabled() || pci_is_root_bus(pdev->bus))
85 return VGA_RSRC_NORMAL_IO | VGA_RSRC_NORMAL_MEM |
86 VGA_RSRC_LEGACY_IO | VGA_RSRC_LEGACY_MEM;
88 max_busnr = pci_bus_max_busnr(pdev->bus);
89 decodes = VGA_RSRC_NORMAL_IO | VGA_RSRC_NORMAL_MEM;
91 while ((tmp = pci_get_class(PCI_CLASS_DISPLAY_VGA << 8, tmp)) != NULL) {
93 pci_domain_nr(tmp->bus) != pci_domain_nr(pdev->bus) ||
94 pci_is_root_bus(tmp->bus))
97 if (tmp->bus->number >= pdev->bus->number &&
98 tmp->bus->number <= max_busnr) {
100 decodes |= VGA_RSRC_LEGACY_IO | VGA_RSRC_LEGACY_MEM;
108 static inline bool vfio_pci_is_vga(struct pci_dev *pdev)
110 return (pdev->class >> 8) == PCI_CLASS_DISPLAY_VGA;
113 static void vfio_pci_try_bus_reset(struct vfio_pci_device *vdev);
115 static int vfio_pci_enable(struct vfio_pci_device *vdev)
117 struct pci_dev *pdev = vdev->pdev;
122 pci_set_power_state(pdev, PCI_D0);
124 /* Don't allow our initial saved state to include busmaster */
125 pci_clear_master(pdev);
127 ret = pci_enable_device(pdev);
131 vdev->reset_works = (pci_reset_function(pdev) == 0);
132 pci_save_state(pdev);
133 vdev->pci_saved_state = pci_store_saved_state(pdev);
134 if (!vdev->pci_saved_state)
135 pr_debug("%s: Couldn't store %s saved state\n",
136 __func__, dev_name(&pdev->dev));
138 ret = vfio_config_init(vdev);
140 kfree(vdev->pci_saved_state);
141 vdev->pci_saved_state = NULL;
142 pci_disable_device(pdev);
146 if (likely(!nointxmask))
147 vdev->pci_2_3 = pci_intx_mask_supported(pdev);
149 pci_read_config_word(pdev, PCI_COMMAND, &cmd);
150 if (vdev->pci_2_3 && (cmd & PCI_COMMAND_INTX_DISABLE)) {
151 cmd &= ~PCI_COMMAND_INTX_DISABLE;
152 pci_write_config_word(pdev, PCI_COMMAND, cmd);
155 msix_pos = pdev->msix_cap;
160 pci_read_config_word(pdev, msix_pos + PCI_MSIX_FLAGS, &flags);
161 pci_read_config_dword(pdev, msix_pos + PCI_MSIX_TABLE, &table);
163 vdev->msix_bar = table & PCI_MSIX_TABLE_BIR;
164 vdev->msix_offset = table & PCI_MSIX_TABLE_OFFSET;
165 vdev->msix_size = ((flags & PCI_MSIX_FLAGS_QSIZE) + 1) * 16;
167 vdev->msix_bar = 0xFF;
169 if (!vfio_vga_disabled() && vfio_pci_is_vga(pdev))
170 vdev->has_vga = true;
175 static void vfio_pci_disable(struct vfio_pci_device *vdev)
177 struct pci_dev *pdev = vdev->pdev;
180 /* Stop the device from further DMA */
181 pci_clear_master(pdev);
183 vfio_pci_set_irqs_ioctl(vdev, VFIO_IRQ_SET_DATA_NONE |
184 VFIO_IRQ_SET_ACTION_TRIGGER,
185 vdev->irq_type, 0, 0, NULL);
187 vdev->virq_disabled = false;
189 vfio_config_free(vdev);
191 for (bar = PCI_STD_RESOURCES; bar <= PCI_STD_RESOURCE_END; bar++) {
192 if (!vdev->barmap[bar])
194 pci_iounmap(pdev, vdev->barmap[bar]);
195 pci_release_selected_regions(pdev, 1 << bar);
196 vdev->barmap[bar] = NULL;
199 vdev->needs_reset = true;
202 * If we have saved state, restore it. If we can reset the device,
203 * even better. Resetting with current state seems better than
204 * nothing, but saving and restoring current state without reset
207 if (pci_load_and_free_saved_state(pdev, &vdev->pci_saved_state)) {
208 pr_info("%s: Couldn't reload %s saved state\n",
209 __func__, dev_name(&pdev->dev));
211 if (!vdev->reset_works)
214 pci_save_state(pdev);
218 * Disable INTx and MSI, presumably to avoid spurious interrupts
219 * during reset. Stolen from pci_reset_function()
221 pci_write_config_word(pdev, PCI_COMMAND, PCI_COMMAND_INTX_DISABLE);
224 * Try to reset the device. The success of this is dependent on
225 * being able to lock the device, which is not always possible.
227 if (vdev->reset_works && !pci_try_reset_function(pdev))
228 vdev->needs_reset = false;
230 pci_restore_state(pdev);
232 pci_disable_device(pdev);
234 vfio_pci_try_bus_reset(vdev);
236 if (!disable_idle_d3)
237 pci_set_power_state(pdev, PCI_D3hot);
240 static void vfio_pci_release(void *device_data)
242 struct vfio_pci_device *vdev = device_data;
244 mutex_lock(&driver_lock);
246 if (!(--vdev->refcnt)) {
247 vfio_spapr_pci_eeh_release(vdev->pdev);
248 vfio_pci_disable(vdev);
251 mutex_unlock(&driver_lock);
253 module_put(THIS_MODULE);
256 static int vfio_pci_open(void *device_data)
258 struct vfio_pci_device *vdev = device_data;
261 if (!try_module_get(THIS_MODULE))
264 mutex_lock(&driver_lock);
267 ret = vfio_pci_enable(vdev);
271 vfio_spapr_pci_eeh_open(vdev->pdev);
275 mutex_unlock(&driver_lock);
277 module_put(THIS_MODULE);
281 static int vfio_pci_get_irq_count(struct vfio_pci_device *vdev, int irq_type)
283 if (irq_type == VFIO_PCI_INTX_IRQ_INDEX) {
285 pci_read_config_byte(vdev->pdev, PCI_INTERRUPT_PIN, &pin);
286 if (IS_ENABLED(CONFIG_VFIO_PCI_INTX) && pin)
289 } else if (irq_type == VFIO_PCI_MSI_IRQ_INDEX) {
293 pos = vdev->pdev->msi_cap;
295 pci_read_config_word(vdev->pdev,
296 pos + PCI_MSI_FLAGS, &flags);
297 return 1 << ((flags & PCI_MSI_FLAGS_QMASK) >> 1);
299 } else if (irq_type == VFIO_PCI_MSIX_IRQ_INDEX) {
303 pos = vdev->pdev->msix_cap;
305 pci_read_config_word(vdev->pdev,
306 pos + PCI_MSIX_FLAGS, &flags);
308 return (flags & PCI_MSIX_FLAGS_QSIZE) + 1;
310 } else if (irq_type == VFIO_PCI_ERR_IRQ_INDEX) {
311 if (pci_is_pcie(vdev->pdev))
313 } else if (irq_type == VFIO_PCI_REQ_IRQ_INDEX) {
320 static int vfio_pci_count_devs(struct pci_dev *pdev, void *data)
326 struct vfio_pci_fill_info {
329 struct vfio_pci_dependent_device *devices;
332 static int vfio_pci_fill_devs(struct pci_dev *pdev, void *data)
334 struct vfio_pci_fill_info *fill = data;
335 struct iommu_group *iommu_group;
337 if (fill->cur == fill->max)
338 return -EAGAIN; /* Something changed, try again */
340 iommu_group = iommu_group_get(&pdev->dev);
342 return -EPERM; /* Cannot reset non-isolated devices */
344 fill->devices[fill->cur].group_id = iommu_group_id(iommu_group);
345 fill->devices[fill->cur].segment = pci_domain_nr(pdev->bus);
346 fill->devices[fill->cur].bus = pdev->bus->number;
347 fill->devices[fill->cur].devfn = pdev->devfn;
349 iommu_group_put(iommu_group);
353 struct vfio_pci_group_entry {
354 struct vfio_group *group;
358 struct vfio_pci_group_info {
360 struct vfio_pci_group_entry *groups;
363 static int vfio_pci_validate_devs(struct pci_dev *pdev, void *data)
365 struct vfio_pci_group_info *info = data;
366 struct iommu_group *group;
369 group = iommu_group_get(&pdev->dev);
373 id = iommu_group_id(group);
375 for (i = 0; i < info->count; i++)
376 if (info->groups[i].id == id)
379 iommu_group_put(group);
381 return (i == info->count) ? -EINVAL : 0;
384 static bool vfio_pci_dev_below_slot(struct pci_dev *pdev, struct pci_slot *slot)
386 for (; pdev; pdev = pdev->bus->self)
387 if (pdev->bus == slot->bus)
388 return (pdev->slot == slot);
392 struct vfio_pci_walk_info {
393 int (*fn)(struct pci_dev *, void *data);
395 struct pci_dev *pdev;
400 static int vfio_pci_walk_wrapper(struct pci_dev *pdev, void *data)
402 struct vfio_pci_walk_info *walk = data;
404 if (!walk->slot || vfio_pci_dev_below_slot(pdev, walk->pdev->slot))
405 walk->ret = walk->fn(pdev, walk->data);
410 static int vfio_pci_for_each_slot_or_bus(struct pci_dev *pdev,
411 int (*fn)(struct pci_dev *,
412 void *data), void *data,
415 struct vfio_pci_walk_info walk = {
416 .fn = fn, .data = data, .pdev = pdev, .slot = slot, .ret = 0,
419 pci_walk_bus(pdev->bus, vfio_pci_walk_wrapper, &walk);
424 static long vfio_pci_ioctl(void *device_data,
425 unsigned int cmd, unsigned long arg)
427 struct vfio_pci_device *vdev = device_data;
430 if (cmd == VFIO_DEVICE_GET_INFO) {
431 struct vfio_device_info info;
433 minsz = offsetofend(struct vfio_device_info, num_irqs);
435 if (copy_from_user(&info, (void __user *)arg, minsz))
438 if (info.argsz < minsz)
441 info.flags = VFIO_DEVICE_FLAGS_PCI;
443 if (vdev->reset_works)
444 info.flags |= VFIO_DEVICE_FLAGS_RESET;
446 info.num_regions = VFIO_PCI_NUM_REGIONS;
447 info.num_irqs = VFIO_PCI_NUM_IRQS;
449 return copy_to_user((void __user *)arg, &info, minsz) ?
452 } else if (cmd == VFIO_DEVICE_GET_REGION_INFO) {
453 struct pci_dev *pdev = vdev->pdev;
454 struct vfio_region_info info;
456 minsz = offsetofend(struct vfio_region_info, offset);
458 if (copy_from_user(&info, (void __user *)arg, minsz))
461 if (info.argsz < minsz)
464 switch (info.index) {
465 case VFIO_PCI_CONFIG_REGION_INDEX:
466 info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index);
467 info.size = pdev->cfg_size;
468 info.flags = VFIO_REGION_INFO_FLAG_READ |
469 VFIO_REGION_INFO_FLAG_WRITE;
471 case VFIO_PCI_BAR0_REGION_INDEX ... VFIO_PCI_BAR5_REGION_INDEX:
472 info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index);
473 info.size = pci_resource_len(pdev, info.index);
479 info.flags = VFIO_REGION_INFO_FLAG_READ |
480 VFIO_REGION_INFO_FLAG_WRITE;
481 if (IS_ENABLED(CONFIG_VFIO_PCI_MMAP) &&
482 pci_resource_flags(pdev, info.index) &
483 IORESOURCE_MEM && info.size >= PAGE_SIZE)
484 info.flags |= VFIO_REGION_INFO_FLAG_MMAP;
486 case VFIO_PCI_ROM_REGION_INDEX:
491 info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index);
494 /* Report the BAR size, not the ROM size */
495 info.size = pci_resource_len(pdev, info.index);
499 /* Is it really there? */
500 io = pci_map_rom(pdev, &size);
505 pci_unmap_rom(pdev, io);
507 info.flags = VFIO_REGION_INFO_FLAG_READ;
510 case VFIO_PCI_VGA_REGION_INDEX:
514 info.offset = VFIO_PCI_INDEX_TO_OFFSET(info.index);
516 info.flags = VFIO_REGION_INFO_FLAG_READ |
517 VFIO_REGION_INFO_FLAG_WRITE;
524 return copy_to_user((void __user *)arg, &info, minsz) ?
527 } else if (cmd == VFIO_DEVICE_GET_IRQ_INFO) {
528 struct vfio_irq_info info;
530 minsz = offsetofend(struct vfio_irq_info, count);
532 if (copy_from_user(&info, (void __user *)arg, minsz))
535 if (info.argsz < minsz || info.index >= VFIO_PCI_NUM_IRQS)
538 switch (info.index) {
539 case VFIO_PCI_INTX_IRQ_INDEX ... VFIO_PCI_MSIX_IRQ_INDEX:
540 case VFIO_PCI_REQ_IRQ_INDEX:
542 case VFIO_PCI_ERR_IRQ_INDEX:
543 if (pci_is_pcie(vdev->pdev))
545 /* pass thru to return error */
550 info.flags = VFIO_IRQ_INFO_EVENTFD;
552 info.count = vfio_pci_get_irq_count(vdev, info.index);
554 if (info.index == VFIO_PCI_INTX_IRQ_INDEX)
555 info.flags |= (VFIO_IRQ_INFO_MASKABLE |
556 VFIO_IRQ_INFO_AUTOMASKED);
558 info.flags |= VFIO_IRQ_INFO_NORESIZE;
560 return copy_to_user((void __user *)arg, &info, minsz) ?
563 } else if (cmd == VFIO_DEVICE_SET_IRQS) {
564 struct vfio_irq_set hdr;
568 minsz = offsetofend(struct vfio_irq_set, count);
570 if (copy_from_user(&hdr, (void __user *)arg, minsz))
573 if (hdr.argsz < minsz || hdr.index >= VFIO_PCI_NUM_IRQS ||
574 hdr.flags & ~(VFIO_IRQ_SET_DATA_TYPE_MASK |
575 VFIO_IRQ_SET_ACTION_TYPE_MASK))
578 if (!(hdr.flags & VFIO_IRQ_SET_DATA_NONE)) {
580 int max = vfio_pci_get_irq_count(vdev, hdr.index);
582 if (hdr.flags & VFIO_IRQ_SET_DATA_BOOL)
583 size = sizeof(uint8_t);
584 else if (hdr.flags & VFIO_IRQ_SET_DATA_EVENTFD)
585 size = sizeof(int32_t);
589 if (hdr.argsz - minsz < hdr.count * size ||
590 hdr.start >= max || hdr.start + hdr.count > max)
593 data = memdup_user((void __user *)(arg + minsz),
596 return PTR_ERR(data);
599 mutex_lock(&vdev->igate);
601 ret = vfio_pci_set_irqs_ioctl(vdev, hdr.flags, hdr.index,
602 hdr.start, hdr.count, data);
604 mutex_unlock(&vdev->igate);
609 } else if (cmd == VFIO_DEVICE_RESET) {
610 return vdev->reset_works ?
611 pci_try_reset_function(vdev->pdev) : -EINVAL;
613 } else if (cmd == VFIO_DEVICE_GET_PCI_HOT_RESET_INFO) {
614 struct vfio_pci_hot_reset_info hdr;
615 struct vfio_pci_fill_info fill = { 0 };
616 struct vfio_pci_dependent_device *devices = NULL;
620 minsz = offsetofend(struct vfio_pci_hot_reset_info, count);
622 if (copy_from_user(&hdr, (void __user *)arg, minsz))
625 if (hdr.argsz < minsz)
630 /* Can we do a slot or bus reset or neither? */
631 if (!pci_probe_reset_slot(vdev->pdev->slot))
633 else if (pci_probe_reset_bus(vdev->pdev->bus))
636 /* How many devices are affected? */
637 ret = vfio_pci_for_each_slot_or_bus(vdev->pdev,
643 WARN_ON(!fill.max); /* Should always be at least one */
646 * If there's enough space, fill it now, otherwise return
647 * -ENOSPC and the number of devices affected.
649 if (hdr.argsz < sizeof(hdr) + (fill.max * sizeof(*devices))) {
651 hdr.count = fill.max;
652 goto reset_info_exit;
655 devices = kcalloc(fill.max, sizeof(*devices), GFP_KERNEL);
659 fill.devices = devices;
661 ret = vfio_pci_for_each_slot_or_bus(vdev->pdev,
666 * If a device was removed between counting and filling,
667 * we may come up short of fill.max. If a device was
668 * added, we'll have a return of -EAGAIN above.
671 hdr.count = fill.cur;
674 if (copy_to_user((void __user *)arg, &hdr, minsz))
678 if (copy_to_user((void __user *)(arg + minsz), devices,
679 hdr.count * sizeof(*devices)))
686 } else if (cmd == VFIO_DEVICE_PCI_HOT_RESET) {
687 struct vfio_pci_hot_reset hdr;
689 struct vfio_pci_group_entry *groups;
690 struct vfio_pci_group_info info;
692 int i, count = 0, ret = 0;
694 minsz = offsetofend(struct vfio_pci_hot_reset, count);
696 if (copy_from_user(&hdr, (void __user *)arg, minsz))
699 if (hdr.argsz < minsz || hdr.flags)
702 /* Can we do a slot or bus reset or neither? */
703 if (!pci_probe_reset_slot(vdev->pdev->slot))
705 else if (pci_probe_reset_bus(vdev->pdev->bus))
709 * We can't let userspace give us an arbitrarily large
710 * buffer to copy, so verify how many we think there
711 * could be. Note groups can have multiple devices so
712 * one group per device is the max.
714 ret = vfio_pci_for_each_slot_or_bus(vdev->pdev,
720 /* Somewhere between 1 and count is OK */
721 if (!hdr.count || hdr.count > count)
724 group_fds = kcalloc(hdr.count, sizeof(*group_fds), GFP_KERNEL);
725 groups = kcalloc(hdr.count, sizeof(*groups), GFP_KERNEL);
726 if (!group_fds || !groups) {
732 if (copy_from_user(group_fds, (void __user *)(arg + minsz),
733 hdr.count * sizeof(*group_fds))) {
740 * For each group_fd, get the group through the vfio external
741 * user interface and store the group and iommu ID. This
742 * ensures the group is held across the reset.
744 for (i = 0; i < hdr.count; i++) {
745 struct vfio_group *group;
746 struct fd f = fdget(group_fds[i]);
752 group = vfio_group_get_external_user(f.file);
755 ret = PTR_ERR(group);
759 groups[i].group = group;
760 groups[i].id = vfio_external_user_iommu_id(group);
765 /* release reference to groups on error */
767 goto hot_reset_release;
769 info.count = hdr.count;
770 info.groups = groups;
773 * Test whether all the affected devices are contained
774 * by the set of groups provided by the user.
776 ret = vfio_pci_for_each_slot_or_bus(vdev->pdev,
777 vfio_pci_validate_devs,
780 /* User has access, do the reset */
781 ret = slot ? pci_try_reset_slot(vdev->pdev->slot) :
782 pci_try_reset_bus(vdev->pdev->bus);
785 for (i--; i >= 0; i--)
786 vfio_group_put_external_user(groups[i].group);
795 static ssize_t vfio_pci_rw(void *device_data, char __user *buf,
796 size_t count, loff_t *ppos, bool iswrite)
798 unsigned int index = VFIO_PCI_OFFSET_TO_INDEX(*ppos);
799 struct vfio_pci_device *vdev = device_data;
801 if (index >= VFIO_PCI_NUM_REGIONS)
805 case VFIO_PCI_CONFIG_REGION_INDEX:
806 return vfio_pci_config_rw(vdev, buf, count, ppos, iswrite);
808 case VFIO_PCI_ROM_REGION_INDEX:
811 return vfio_pci_bar_rw(vdev, buf, count, ppos, false);
813 case VFIO_PCI_BAR0_REGION_INDEX ... VFIO_PCI_BAR5_REGION_INDEX:
814 return vfio_pci_bar_rw(vdev, buf, count, ppos, iswrite);
816 case VFIO_PCI_VGA_REGION_INDEX:
817 return vfio_pci_vga_rw(vdev, buf, count, ppos, iswrite);
823 static ssize_t vfio_pci_read(void *device_data, char __user *buf,
824 size_t count, loff_t *ppos)
829 return vfio_pci_rw(device_data, buf, count, ppos, false);
832 static ssize_t vfio_pci_write(void *device_data, const char __user *buf,
833 size_t count, loff_t *ppos)
838 return vfio_pci_rw(device_data, (char __user *)buf, count, ppos, true);
841 static int vfio_pci_mmap(void *device_data, struct vm_area_struct *vma)
843 struct vfio_pci_device *vdev = device_data;
844 struct pci_dev *pdev = vdev->pdev;
846 u64 phys_len, req_len, pgoff, req_start;
849 index = vma->vm_pgoff >> (VFIO_PCI_OFFSET_SHIFT - PAGE_SHIFT);
851 if (vma->vm_end < vma->vm_start)
853 if ((vma->vm_flags & VM_SHARED) == 0)
855 if (index >= VFIO_PCI_ROM_REGION_INDEX)
857 if (!(pci_resource_flags(pdev, index) & IORESOURCE_MEM))
860 phys_len = pci_resource_len(pdev, index);
861 req_len = vma->vm_end - vma->vm_start;
862 pgoff = vma->vm_pgoff &
863 ((1U << (VFIO_PCI_OFFSET_SHIFT - PAGE_SHIFT)) - 1);
864 req_start = pgoff << PAGE_SHIFT;
866 if (phys_len < PAGE_SIZE || req_start + req_len > phys_len)
869 if (index == vdev->msix_bar) {
871 * Disallow mmaps overlapping the MSI-X table; users don't
872 * get to touch this directly. We could find somewhere
873 * else to map the overlap, but page granularity is only
874 * a recommendation, not a requirement, so the user needs
875 * to know which bits are real. Requiring them to mmap
876 * around the table makes that clear.
879 /* If neither entirely above nor below, then it overlaps */
880 if (!(req_start >= vdev->msix_offset + vdev->msix_size ||
881 req_start + req_len <= vdev->msix_offset))
886 * Even though we don't make use of the barmap for the mmap,
887 * we need to request the region and the barmap tracks that.
889 if (!vdev->barmap[index]) {
890 ret = pci_request_selected_regions(pdev,
891 1 << index, "vfio-pci");
895 vdev->barmap[index] = pci_iomap(pdev, index, 0);
898 vma->vm_private_data = vdev;
899 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
900 vma->vm_pgoff = (pci_resource_start(pdev, index) >> PAGE_SHIFT) + pgoff;
902 return remap_pfn_range(vma, vma->vm_start, vma->vm_pgoff,
903 req_len, vma->vm_page_prot);
906 static void vfio_pci_request(void *device_data, unsigned int count)
908 struct vfio_pci_device *vdev = device_data;
910 mutex_lock(&vdev->igate);
912 if (vdev->req_trigger) {
914 dev_notice_ratelimited(&vdev->pdev->dev,
915 "Relaying device request to user (#%u)\n",
917 eventfd_signal(vdev->req_trigger, 1);
918 } else if (count == 0) {
919 dev_warn(&vdev->pdev->dev,
920 "No device request channel registered, blocked until released by user\n");
923 mutex_unlock(&vdev->igate);
926 static const struct vfio_device_ops vfio_pci_ops = {
928 .open = vfio_pci_open,
929 .release = vfio_pci_release,
930 .ioctl = vfio_pci_ioctl,
931 .read = vfio_pci_read,
932 .write = vfio_pci_write,
933 .mmap = vfio_pci_mmap,
934 .request = vfio_pci_request,
937 static int vfio_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)
939 struct vfio_pci_device *vdev;
940 struct iommu_group *group;
943 if (pdev->hdr_type != PCI_HEADER_TYPE_NORMAL)
946 group = iommu_group_get(&pdev->dev);
950 vdev = kzalloc(sizeof(*vdev), GFP_KERNEL);
952 iommu_group_put(group);
957 vdev->irq_type = VFIO_PCI_NUM_IRQS;
958 mutex_init(&vdev->igate);
959 spin_lock_init(&vdev->irqlock);
961 ret = vfio_add_group_dev(&pdev->dev, &vfio_pci_ops, vdev);
963 iommu_group_put(group);
968 if (vfio_pci_is_vga(pdev)) {
969 vga_client_register(pdev, vdev, NULL, vfio_pci_set_vga_decode);
970 vga_set_legacy_decoding(pdev,
971 vfio_pci_set_vga_decode(vdev, false));
974 if (!disable_idle_d3) {
976 * pci-core sets the device power state to an unknown value at
977 * bootup and after being removed from a driver. The only
978 * transition it allows from this unknown state is to D0, which
979 * typically happens when a driver calls pci_enable_device().
980 * We're not ready to enable the device yet, but we do want to
981 * be able to get to D3. Therefore first do a D0 transition
982 * before going to D3.
984 pci_set_power_state(pdev, PCI_D0);
985 pci_set_power_state(pdev, PCI_D3hot);
991 static void vfio_pci_remove(struct pci_dev *pdev)
993 struct vfio_pci_device *vdev;
995 vdev = vfio_del_group_dev(&pdev->dev);
999 iommu_group_put(pdev->dev.iommu_group);
1002 if (vfio_pci_is_vga(pdev)) {
1003 vga_client_register(pdev, NULL, NULL, NULL);
1004 vga_set_legacy_decoding(pdev,
1005 VGA_RSRC_NORMAL_IO | VGA_RSRC_NORMAL_MEM |
1006 VGA_RSRC_LEGACY_IO | VGA_RSRC_LEGACY_MEM);
1009 if (!disable_idle_d3)
1010 pci_set_power_state(pdev, PCI_D0);
1013 static pci_ers_result_t vfio_pci_aer_err_detected(struct pci_dev *pdev,
1014 pci_channel_state_t state)
1016 struct vfio_pci_device *vdev;
1017 struct vfio_device *device;
1019 device = vfio_device_get_from_dev(&pdev->dev);
1021 return PCI_ERS_RESULT_DISCONNECT;
1023 vdev = vfio_device_data(device);
1025 vfio_device_put(device);
1026 return PCI_ERS_RESULT_DISCONNECT;
1029 mutex_lock(&vdev->igate);
1031 if (vdev->err_trigger)
1032 eventfd_signal(vdev->err_trigger, 1);
1034 mutex_unlock(&vdev->igate);
1036 vfio_device_put(device);
1038 return PCI_ERS_RESULT_CAN_RECOVER;
1041 static const struct pci_error_handlers vfio_err_handlers = {
1042 .error_detected = vfio_pci_aer_err_detected,
1045 static struct pci_driver vfio_pci_driver = {
1047 .id_table = NULL, /* only dynamic ids */
1048 .probe = vfio_pci_probe,
1049 .remove = vfio_pci_remove,
1050 .err_handler = &vfio_err_handlers,
1053 struct vfio_devices {
1054 struct vfio_device **devices;
1059 static int vfio_pci_get_devs(struct pci_dev *pdev, void *data)
1061 struct vfio_devices *devs = data;
1062 struct vfio_device *device;
1064 if (devs->cur_index == devs->max_index)
1067 device = vfio_device_get_from_dev(&pdev->dev);
1071 if (pci_dev_driver(pdev) != &vfio_pci_driver) {
1072 vfio_device_put(device);
1076 devs->devices[devs->cur_index++] = device;
1081 * Attempt to do a bus/slot reset if there are devices affected by a reset for
1082 * this device that are needs_reset and all of the affected devices are unused
1083 * (!refcnt). Callers are required to hold driver_lock when calling this to
1084 * prevent device opens and concurrent bus reset attempts. We prevent device
1085 * unbinds by acquiring and holding a reference to the vfio_device.
1087 * NB: vfio-core considers a group to be viable even if some devices are
1088 * bound to drivers like pci-stub or pcieport. Here we require all devices
1089 * to be bound to vfio_pci since that's the only way we can be sure they
1092 static void vfio_pci_try_bus_reset(struct vfio_pci_device *vdev)
1094 struct vfio_devices devs = { .cur_index = 0 };
1095 int i = 0, ret = -EINVAL;
1096 bool needs_reset = false, slot = false;
1097 struct vfio_pci_device *tmp;
1099 if (!pci_probe_reset_slot(vdev->pdev->slot))
1101 else if (pci_probe_reset_bus(vdev->pdev->bus))
1104 if (vfio_pci_for_each_slot_or_bus(vdev->pdev, vfio_pci_count_devs,
1109 devs.devices = kcalloc(i, sizeof(struct vfio_device *), GFP_KERNEL);
1113 if (vfio_pci_for_each_slot_or_bus(vdev->pdev,
1114 vfio_pci_get_devs, &devs, slot))
1117 for (i = 0; i < devs.cur_index; i++) {
1118 tmp = vfio_device_data(devs.devices[i]);
1119 if (tmp->needs_reset)
1126 ret = slot ? pci_try_reset_slot(vdev->pdev->slot) :
1127 pci_try_reset_bus(vdev->pdev->bus);
1130 for (i = 0; i < devs.cur_index; i++) {
1131 tmp = vfio_device_data(devs.devices[i]);
1133 tmp->needs_reset = false;
1135 if (!tmp->refcnt && !disable_idle_d3)
1136 pci_set_power_state(tmp->pdev, PCI_D3hot);
1138 vfio_device_put(devs.devices[i]);
1141 kfree(devs.devices);
1144 static void __exit vfio_pci_cleanup(void)
1146 pci_unregister_driver(&vfio_pci_driver);
1147 vfio_pci_uninit_perm_bits();
1150 static void __init vfio_pci_fill_ids(void)
1155 /* no ids passed actually */
1159 /* add ids specified in the module parameter */
1161 while ((id = strsep(&p, ","))) {
1162 unsigned int vendor, device, subvendor = PCI_ANY_ID,
1163 subdevice = PCI_ANY_ID, class = 0, class_mask = 0;
1169 fields = sscanf(id, "%x:%x:%x:%x:%x:%x",
1170 &vendor, &device, &subvendor, &subdevice,
1171 &class, &class_mask);
1174 pr_warn("invalid id string \"%s\"\n", id);
1178 rc = pci_add_dynid(&vfio_pci_driver, vendor, device,
1179 subvendor, subdevice, class, class_mask, 0);
1181 pr_warn("failed to add dynamic id [%04hx:%04hx[%04hx:%04hx]] class %#08x/%08x (%d)\n",
1182 vendor, device, subvendor, subdevice,
1183 class, class_mask, rc);
1185 pr_info("add [%04hx:%04hx[%04hx:%04hx]] class %#08x/%08x\n",
1186 vendor, device, subvendor, subdevice,
1191 static int __init vfio_pci_init(void)
1195 /* Allocate shared config space permision data used by all devices */
1196 ret = vfio_pci_init_perm_bits();
1200 /* Register and scan for devices */
1201 ret = pci_register_driver(&vfio_pci_driver);
1205 vfio_pci_fill_ids();
1210 vfio_pci_uninit_perm_bits();
1214 module_init(vfio_pci_init);
1215 module_exit(vfio_pci_cleanup);
1217 MODULE_VERSION(DRIVER_VERSION);
1218 MODULE_LICENSE("GPL v2");
1219 MODULE_AUTHOR(DRIVER_AUTHOR);
1220 MODULE_DESCRIPTION(DRIVER_DESC);