1 // SPDX-License-Identifier: GPL-2.0-only
3 * VFIO: IOMMU DMA mapping support for Type1 IOMMU
5 * Copyright (C) 2012 Red Hat, Inc. All rights reserved.
6 * Author: Alex Williamson <alex.williamson@redhat.com>
8 * Derived from original vfio:
9 * Copyright 2010 Cisco Systems, Inc. All rights reserved.
10 * Author: Tom Lyon, pugs@cisco.com
12 * We arbitrarily define a Type1 IOMMU as one matching the below code.
13 * It could be called the x86 IOMMU as it's designed for AMD-Vi & Intel
14 * VT-d, but that makes it harder to re-use as theoretically anyone
15 * implementing a similar IOMMU could make use of this. We expect the
16 * IOMMU to support the IOMMU API and have few to no restrictions around
17 * the IOVA range that can be mapped. The Type1 IOMMU is currently
18 * optimized for relatively static mappings of a userspace process with
19 * userpsace pages pinned into memory. We also assume devices and IOMMU
20 * domains are PCI based as the IOMMU API is still centered around a
21 * device/bus interface rather than a group interface.
24 #include <linux/compat.h>
25 #include <linux/device.h>
27 #include <linux/iommu.h>
28 #include <linux/module.h>
30 #include <linux/kthread.h>
31 #include <linux/rbtree.h>
32 #include <linux/sched/signal.h>
33 #include <linux/sched/mm.h>
34 #include <linux/slab.h>
35 #include <linux/uaccess.h>
36 #include <linux/vfio.h>
37 #include <linux/workqueue.h>
38 #include <linux/mdev.h>
39 #include <linux/notifier.h>
40 #include <linux/dma-iommu.h>
41 #include <linux/irqdomain.h>
43 #define DRIVER_VERSION "0.2"
44 #define DRIVER_AUTHOR "Alex Williamson <alex.williamson@redhat.com>"
45 #define DRIVER_DESC "Type1 IOMMU driver for VFIO"
47 static bool allow_unsafe_interrupts;
48 module_param_named(allow_unsafe_interrupts,
49 allow_unsafe_interrupts, bool, S_IRUGO | S_IWUSR);
50 MODULE_PARM_DESC(allow_unsafe_interrupts,
51 "Enable VFIO IOMMU support for on platforms without interrupt remapping support.");
53 static bool disable_hugepages;
54 module_param_named(disable_hugepages,
55 disable_hugepages, bool, S_IRUGO | S_IWUSR);
56 MODULE_PARM_DESC(disable_hugepages,
57 "Disable VFIO IOMMU support for IOMMU hugepages.");
59 static unsigned int dma_entry_limit __read_mostly = U16_MAX;
60 module_param_named(dma_entry_limit, dma_entry_limit, uint, 0644);
61 MODULE_PARM_DESC(dma_entry_limit,
62 "Maximum number of user DMA mappings per container (65535).");
65 struct list_head domain_list;
66 struct list_head iova_list;
67 struct vfio_domain *external_domain; /* domain for external user */
69 struct rb_root dma_list;
70 struct blocking_notifier_head notifier;
71 unsigned int dma_avail;
72 uint64_t pgsize_bitmap;
75 bool dirty_page_tracking;
76 bool pinned_page_dirty_scope;
80 struct iommu_domain *domain;
81 struct list_head next;
82 struct list_head group_list;
83 int prot; /* IOMMU_CACHE */
84 bool fgsp; /* Fine-grained super pages */
89 dma_addr_t iova; /* Device address */
90 unsigned long vaddr; /* Process virtual addr */
91 size_t size; /* Map size (bytes) */
92 int prot; /* IOMMU_READ/WRITE */
94 bool lock_cap; /* capable(CAP_IPC_LOCK) */
95 struct task_struct *task;
96 struct rb_root pfn_list; /* Ex-user pinned pfn list */
97 unsigned long *bitmap;
101 struct iommu_group *iommu_group;
102 struct list_head next;
103 bool mdev_group; /* An mdev group */
104 bool pinned_page_dirty_scope;
108 struct list_head list;
114 * Guest RAM pinning working set or DMA target
118 dma_addr_t iova; /* Device address */
119 unsigned long pfn; /* Host pfn */
120 unsigned int ref_count;
123 struct vfio_regions {
124 struct list_head list;
130 #define IS_IOMMU_CAP_DOMAIN_IN_CONTAINER(iommu) \
131 (!list_empty(&iommu->domain_list))
133 #define DIRTY_BITMAP_BYTES(n) (ALIGN(n, BITS_PER_TYPE(u64)) / BITS_PER_BYTE)
136 * Input argument of number of bits to bitmap_set() is unsigned integer, which
137 * further casts to signed integer for unaligned multi-bit operation,
139 * Then maximum bitmap size supported is 2^31 bits divided by 2^3 bits/byte,
140 * that is 2^28 (256 MB) which maps to 2^31 * 2^12 = 2^43 (8TB) on 4K page
143 #define DIRTY_BITMAP_PAGES_MAX ((u64)INT_MAX)
144 #define DIRTY_BITMAP_SIZE_MAX DIRTY_BITMAP_BYTES(DIRTY_BITMAP_PAGES_MAX)
146 static int put_pfn(unsigned long pfn, int prot);
148 static struct vfio_group *vfio_iommu_find_iommu_group(struct vfio_iommu *iommu,
149 struct iommu_group *iommu_group);
151 static void update_pinned_page_dirty_scope(struct vfio_iommu *iommu);
153 * This code handles mapping and unmapping of user data buffers
154 * into DMA'ble space using the IOMMU
157 static struct vfio_dma *vfio_find_dma(struct vfio_iommu *iommu,
158 dma_addr_t start, size_t size)
160 struct rb_node *node = iommu->dma_list.rb_node;
163 struct vfio_dma *dma = rb_entry(node, struct vfio_dma, node);
165 if (start + size <= dma->iova)
166 node = node->rb_left;
167 else if (start >= dma->iova + dma->size)
168 node = node->rb_right;
176 static struct rb_node *vfio_find_dma_first_node(struct vfio_iommu *iommu,
177 dma_addr_t start, size_t size)
179 struct rb_node *res = NULL;
180 struct rb_node *node = iommu->dma_list.rb_node;
181 struct vfio_dma *dma_res = NULL;
184 struct vfio_dma *dma = rb_entry(node, struct vfio_dma, node);
186 if (start < dma->iova + dma->size) {
189 if (start >= dma->iova)
191 node = node->rb_left;
193 node = node->rb_right;
196 if (res && size && dma_res->iova >= start + size)
201 static void vfio_link_dma(struct vfio_iommu *iommu, struct vfio_dma *new)
203 struct rb_node **link = &iommu->dma_list.rb_node, *parent = NULL;
204 struct vfio_dma *dma;
208 dma = rb_entry(parent, struct vfio_dma, node);
210 if (new->iova + new->size <= dma->iova)
211 link = &(*link)->rb_left;
213 link = &(*link)->rb_right;
216 rb_link_node(&new->node, parent, link);
217 rb_insert_color(&new->node, &iommu->dma_list);
220 static void vfio_unlink_dma(struct vfio_iommu *iommu, struct vfio_dma *old)
222 rb_erase(&old->node, &iommu->dma_list);
226 static int vfio_dma_bitmap_alloc(struct vfio_dma *dma, size_t pgsize)
228 uint64_t npages = dma->size / pgsize;
230 if (npages > DIRTY_BITMAP_PAGES_MAX)
234 * Allocate extra 64 bits that are used to calculate shift required for
235 * bitmap_shift_left() to manipulate and club unaligned number of pages
236 * in adjacent vfio_dma ranges.
238 dma->bitmap = kvzalloc(DIRTY_BITMAP_BYTES(npages) + sizeof(u64),
246 static void vfio_dma_bitmap_free(struct vfio_dma *dma)
252 static void vfio_dma_populate_bitmap(struct vfio_dma *dma, size_t pgsize)
255 unsigned long pgshift = __ffs(pgsize);
257 for (p = rb_first(&dma->pfn_list); p; p = rb_next(p)) {
258 struct vfio_pfn *vpfn = rb_entry(p, struct vfio_pfn, node);
260 bitmap_set(dma->bitmap, (vpfn->iova - dma->iova) >> pgshift, 1);
264 static int vfio_dma_bitmap_alloc_all(struct vfio_iommu *iommu, size_t pgsize)
268 for (n = rb_first(&iommu->dma_list); n; n = rb_next(n)) {
269 struct vfio_dma *dma = rb_entry(n, struct vfio_dma, node);
272 ret = vfio_dma_bitmap_alloc(dma, pgsize);
276 for (p = rb_prev(n); p; p = rb_prev(p)) {
277 struct vfio_dma *dma = rb_entry(n,
278 struct vfio_dma, node);
280 vfio_dma_bitmap_free(dma);
284 vfio_dma_populate_bitmap(dma, pgsize);
289 static void vfio_dma_bitmap_free_all(struct vfio_iommu *iommu)
293 for (n = rb_first(&iommu->dma_list); n; n = rb_next(n)) {
294 struct vfio_dma *dma = rb_entry(n, struct vfio_dma, node);
296 vfio_dma_bitmap_free(dma);
301 * Helper Functions for host iova-pfn list
303 static struct vfio_pfn *vfio_find_vpfn(struct vfio_dma *dma, dma_addr_t iova)
305 struct vfio_pfn *vpfn;
306 struct rb_node *node = dma->pfn_list.rb_node;
309 vpfn = rb_entry(node, struct vfio_pfn, node);
311 if (iova < vpfn->iova)
312 node = node->rb_left;
313 else if (iova > vpfn->iova)
314 node = node->rb_right;
321 static void vfio_link_pfn(struct vfio_dma *dma,
322 struct vfio_pfn *new)
324 struct rb_node **link, *parent = NULL;
325 struct vfio_pfn *vpfn;
327 link = &dma->pfn_list.rb_node;
330 vpfn = rb_entry(parent, struct vfio_pfn, node);
332 if (new->iova < vpfn->iova)
333 link = &(*link)->rb_left;
335 link = &(*link)->rb_right;
338 rb_link_node(&new->node, parent, link);
339 rb_insert_color(&new->node, &dma->pfn_list);
342 static void vfio_unlink_pfn(struct vfio_dma *dma, struct vfio_pfn *old)
344 rb_erase(&old->node, &dma->pfn_list);
347 static int vfio_add_to_pfn_list(struct vfio_dma *dma, dma_addr_t iova,
350 struct vfio_pfn *vpfn;
352 vpfn = kzalloc(sizeof(*vpfn), GFP_KERNEL);
359 vfio_link_pfn(dma, vpfn);
363 static void vfio_remove_from_pfn_list(struct vfio_dma *dma,
364 struct vfio_pfn *vpfn)
366 vfio_unlink_pfn(dma, vpfn);
370 static struct vfio_pfn *vfio_iova_get_vfio_pfn(struct vfio_dma *dma,
373 struct vfio_pfn *vpfn = vfio_find_vpfn(dma, iova);
380 static int vfio_iova_put_vfio_pfn(struct vfio_dma *dma, struct vfio_pfn *vpfn)
385 if (!vpfn->ref_count) {
386 ret = put_pfn(vpfn->pfn, dma->prot);
387 vfio_remove_from_pfn_list(dma, vpfn);
392 static int vfio_lock_acct(struct vfio_dma *dma, long npage, bool async)
394 struct mm_struct *mm;
400 mm = async ? get_task_mm(dma->task) : dma->task->mm;
402 return -ESRCH; /* process exited */
404 ret = mmap_write_lock_killable(mm);
406 ret = __account_locked_vm(mm, abs(npage), npage > 0, dma->task,
408 mmap_write_unlock(mm);
418 * Some mappings aren't backed by a struct page, for example an mmap'd
419 * MMIO range for our own or another device. These use a different
420 * pfn conversion and shouldn't be tracked as locked pages.
421 * For compound pages, any driver that sets the reserved bit in head
422 * page needs to set the reserved bit in all subpages to be safe.
424 static bool is_invalid_reserved_pfn(unsigned long pfn)
427 return PageReserved(pfn_to_page(pfn));
432 static int put_pfn(unsigned long pfn, int prot)
434 if (!is_invalid_reserved_pfn(pfn)) {
435 struct page *page = pfn_to_page(pfn);
437 unpin_user_pages_dirty_lock(&page, 1, prot & IOMMU_WRITE);
443 static int follow_fault_pfn(struct vm_area_struct *vma, struct mm_struct *mm,
444 unsigned long vaddr, unsigned long *pfn,
449 ret = follow_pfn(vma, vaddr, pfn);
451 bool unlocked = false;
453 ret = fixup_user_fault(mm, vaddr,
455 (write_fault ? FAULT_FLAG_WRITE : 0),
463 ret = follow_pfn(vma, vaddr, pfn);
469 static int vaddr_get_pfn(struct mm_struct *mm, unsigned long vaddr,
470 int prot, unsigned long *pfn)
472 struct page *page[1];
473 struct vm_area_struct *vma;
474 unsigned int flags = 0;
477 if (prot & IOMMU_WRITE)
481 ret = pin_user_pages_remote(mm, vaddr, 1, flags | FOLL_LONGTERM,
484 *pfn = page_to_pfn(page[0]);
489 vaddr = untagged_addr(vaddr);
492 vma = find_vma_intersection(mm, vaddr, vaddr + 1);
494 if (vma && vma->vm_flags & VM_PFNMAP) {
495 ret = follow_fault_pfn(vma, mm, vaddr, pfn, prot & IOMMU_WRITE);
499 if (!ret && !is_invalid_reserved_pfn(*pfn))
503 mmap_read_unlock(mm);
508 * Attempt to pin pages. We really don't want to track all the pfns and
509 * the iommu can only map chunks of consecutive pfns anyway, so get the
510 * first page and all consecutive pages with the same locking.
512 static long vfio_pin_pages_remote(struct vfio_dma *dma, unsigned long vaddr,
513 long npage, unsigned long *pfn_base,
516 unsigned long pfn = 0;
517 long ret, pinned = 0, lock_acct = 0;
519 dma_addr_t iova = vaddr - dma->vaddr + dma->iova;
521 /* This code path is only user initiated */
525 ret = vaddr_get_pfn(current->mm, vaddr, dma->prot, pfn_base);
530 rsvd = is_invalid_reserved_pfn(*pfn_base);
533 * Reserved pages aren't counted against the user, externally pinned
534 * pages are already counted against the user.
536 if (!rsvd && !vfio_find_vpfn(dma, iova)) {
537 if (!dma->lock_cap && current->mm->locked_vm + 1 > limit) {
538 put_pfn(*pfn_base, dma->prot);
539 pr_warn("%s: RLIMIT_MEMLOCK (%ld) exceeded\n", __func__,
540 limit << PAGE_SHIFT);
546 if (unlikely(disable_hugepages))
549 /* Lock all the consecutive pages from pfn_base */
550 for (vaddr += PAGE_SIZE, iova += PAGE_SIZE; pinned < npage;
551 pinned++, vaddr += PAGE_SIZE, iova += PAGE_SIZE) {
552 ret = vaddr_get_pfn(current->mm, vaddr, dma->prot, &pfn);
556 if (pfn != *pfn_base + pinned ||
557 rsvd != is_invalid_reserved_pfn(pfn)) {
558 put_pfn(pfn, dma->prot);
562 if (!rsvd && !vfio_find_vpfn(dma, iova)) {
563 if (!dma->lock_cap &&
564 current->mm->locked_vm + lock_acct + 1 > limit) {
565 put_pfn(pfn, dma->prot);
566 pr_warn("%s: RLIMIT_MEMLOCK (%ld) exceeded\n",
567 __func__, limit << PAGE_SHIFT);
576 ret = vfio_lock_acct(dma, lock_acct, false);
581 for (pfn = *pfn_base ; pinned ; pfn++, pinned--)
582 put_pfn(pfn, dma->prot);
591 static long vfio_unpin_pages_remote(struct vfio_dma *dma, dma_addr_t iova,
592 unsigned long pfn, long npage,
595 long unlocked = 0, locked = 0;
598 for (i = 0; i < npage; i++, iova += PAGE_SIZE) {
599 if (put_pfn(pfn++, dma->prot)) {
601 if (vfio_find_vpfn(dma, iova))
607 vfio_lock_acct(dma, locked - unlocked, true);
612 static int vfio_pin_page_external(struct vfio_dma *dma, unsigned long vaddr,
613 unsigned long *pfn_base, bool do_accounting)
615 struct mm_struct *mm;
618 mm = get_task_mm(dma->task);
622 ret = vaddr_get_pfn(mm, vaddr, dma->prot, pfn_base);
623 if (!ret && do_accounting && !is_invalid_reserved_pfn(*pfn_base)) {
624 ret = vfio_lock_acct(dma, 1, true);
626 put_pfn(*pfn_base, dma->prot);
628 pr_warn("%s: Task %s (%d) RLIMIT_MEMLOCK "
629 "(%ld) exceeded\n", __func__,
630 dma->task->comm, task_pid_nr(dma->task),
631 task_rlimit(dma->task, RLIMIT_MEMLOCK));
639 static int vfio_unpin_page_external(struct vfio_dma *dma, dma_addr_t iova,
643 struct vfio_pfn *vpfn = vfio_find_vpfn(dma, iova);
648 unlocked = vfio_iova_put_vfio_pfn(dma, vpfn);
651 vfio_lock_acct(dma, -unlocked, true);
656 static int vfio_iommu_type1_pin_pages(void *iommu_data,
657 struct iommu_group *iommu_group,
658 unsigned long *user_pfn,
660 unsigned long *phys_pfn)
662 struct vfio_iommu *iommu = iommu_data;
663 struct vfio_group *group;
665 unsigned long remote_vaddr;
666 struct vfio_dma *dma;
669 if (!iommu || !user_pfn || !phys_pfn)
672 /* Supported for v2 version only */
676 mutex_lock(&iommu->lock);
678 /* Fail if notifier list is empty */
679 if (!iommu->notifier.head) {
685 * If iommu capable domain exist in the container then all pages are
686 * already pinned and accounted. Accouting should be done if there is no
687 * iommu capable domain in the container.
689 do_accounting = !IS_IOMMU_CAP_DOMAIN_IN_CONTAINER(iommu);
691 for (i = 0; i < npage; i++) {
693 struct vfio_pfn *vpfn;
695 iova = user_pfn[i] << PAGE_SHIFT;
696 dma = vfio_find_dma(iommu, iova, PAGE_SIZE);
702 if ((dma->prot & prot) != prot) {
707 vpfn = vfio_iova_get_vfio_pfn(dma, iova);
709 phys_pfn[i] = vpfn->pfn;
713 remote_vaddr = dma->vaddr + (iova - dma->iova);
714 ret = vfio_pin_page_external(dma, remote_vaddr, &phys_pfn[i],
719 ret = vfio_add_to_pfn_list(dma, iova, phys_pfn[i]);
721 if (put_pfn(phys_pfn[i], dma->prot) && do_accounting)
722 vfio_lock_acct(dma, -1, true);
726 if (iommu->dirty_page_tracking) {
727 unsigned long pgshift = __ffs(iommu->pgsize_bitmap);
730 * Bitmap populated with the smallest supported page
733 bitmap_set(dma->bitmap,
734 (iova - dma->iova) >> pgshift, 1);
739 group = vfio_iommu_find_iommu_group(iommu, iommu_group);
740 if (!group->pinned_page_dirty_scope) {
741 group->pinned_page_dirty_scope = true;
742 update_pinned_page_dirty_scope(iommu);
749 for (j = 0; j < i; j++) {
752 iova = user_pfn[j] << PAGE_SHIFT;
753 dma = vfio_find_dma(iommu, iova, PAGE_SIZE);
754 vfio_unpin_page_external(dma, iova, do_accounting);
758 mutex_unlock(&iommu->lock);
762 static int vfio_iommu_type1_unpin_pages(void *iommu_data,
763 unsigned long *user_pfn,
766 struct vfio_iommu *iommu = iommu_data;
770 if (!iommu || !user_pfn)
773 /* Supported for v2 version only */
777 mutex_lock(&iommu->lock);
779 do_accounting = !IS_IOMMU_CAP_DOMAIN_IN_CONTAINER(iommu);
780 for (i = 0; i < npage; i++) {
781 struct vfio_dma *dma;
784 iova = user_pfn[i] << PAGE_SHIFT;
785 dma = vfio_find_dma(iommu, iova, PAGE_SIZE);
788 vfio_unpin_page_external(dma, iova, do_accounting);
792 mutex_unlock(&iommu->lock);
793 return i > npage ? npage : (i > 0 ? i : -EINVAL);
796 static long vfio_sync_unpin(struct vfio_dma *dma, struct vfio_domain *domain,
797 struct list_head *regions,
798 struct iommu_iotlb_gather *iotlb_gather)
801 struct vfio_regions *entry, *next;
803 iommu_iotlb_sync(domain->domain, iotlb_gather);
805 list_for_each_entry_safe(entry, next, regions, list) {
806 unlocked += vfio_unpin_pages_remote(dma,
808 entry->phys >> PAGE_SHIFT,
809 entry->len >> PAGE_SHIFT,
811 list_del(&entry->list);
821 * Generally, VFIO needs to unpin remote pages after each IOTLB flush.
822 * Therefore, when using IOTLB flush sync interface, VFIO need to keep track
823 * of these regions (currently using a list).
825 * This value specifies maximum number of regions for each IOTLB flush sync.
827 #define VFIO_IOMMU_TLB_SYNC_MAX 512
829 static size_t unmap_unpin_fast(struct vfio_domain *domain,
830 struct vfio_dma *dma, dma_addr_t *iova,
831 size_t len, phys_addr_t phys, long *unlocked,
832 struct list_head *unmapped_list,
834 struct iommu_iotlb_gather *iotlb_gather)
837 struct vfio_regions *entry = kzalloc(sizeof(*entry), GFP_KERNEL);
840 unmapped = iommu_unmap_fast(domain->domain, *iova, len,
848 entry->len = unmapped;
849 list_add_tail(&entry->list, unmapped_list);
857 * Sync if the number of fast-unmap regions hits the limit
858 * or in case of errors.
860 if (*unmapped_cnt >= VFIO_IOMMU_TLB_SYNC_MAX || !unmapped) {
861 *unlocked += vfio_sync_unpin(dma, domain, unmapped_list,
869 static size_t unmap_unpin_slow(struct vfio_domain *domain,
870 struct vfio_dma *dma, dma_addr_t *iova,
871 size_t len, phys_addr_t phys,
874 size_t unmapped = iommu_unmap(domain->domain, *iova, len);
877 *unlocked += vfio_unpin_pages_remote(dma, *iova,
879 unmapped >> PAGE_SHIFT,
887 static long vfio_unmap_unpin(struct vfio_iommu *iommu, struct vfio_dma *dma,
890 dma_addr_t iova = dma->iova, end = dma->iova + dma->size;
891 struct vfio_domain *domain, *d;
892 LIST_HEAD(unmapped_region_list);
893 struct iommu_iotlb_gather iotlb_gather;
894 int unmapped_region_cnt = 0;
900 if (!IS_IOMMU_CAP_DOMAIN_IN_CONTAINER(iommu))
904 * We use the IOMMU to track the physical addresses, otherwise we'd
905 * need a much more complicated tracking system. Unfortunately that
906 * means we need to use one of the iommu domains to figure out the
907 * pfns to unpin. The rest need to be unmapped in advance so we have
908 * no iommu translations remaining when the pages are unpinned.
910 domain = d = list_first_entry(&iommu->domain_list,
911 struct vfio_domain, next);
913 list_for_each_entry_continue(d, &iommu->domain_list, next) {
914 iommu_unmap(d->domain, dma->iova, dma->size);
918 iommu_iotlb_gather_init(&iotlb_gather);
920 size_t unmapped, len;
921 phys_addr_t phys, next;
923 phys = iommu_iova_to_phys(domain->domain, iova);
924 if (WARN_ON(!phys)) {
930 * To optimize for fewer iommu_unmap() calls, each of which
931 * may require hardware cache flushing, try to find the
932 * largest contiguous physical memory chunk to unmap.
934 for (len = PAGE_SIZE;
935 !domain->fgsp && iova + len < end; len += PAGE_SIZE) {
936 next = iommu_iova_to_phys(domain->domain, iova + len);
937 if (next != phys + len)
942 * First, try to use fast unmap/unpin. In case of failure,
943 * switch to slow unmap/unpin path.
945 unmapped = unmap_unpin_fast(domain, dma, &iova, len, phys,
946 &unlocked, &unmapped_region_list,
947 &unmapped_region_cnt,
950 unmapped = unmap_unpin_slow(domain, dma, &iova, len,
952 if (WARN_ON(!unmapped))
957 dma->iommu_mapped = false;
959 if (unmapped_region_cnt) {
960 unlocked += vfio_sync_unpin(dma, domain, &unmapped_region_list,
965 vfio_lock_acct(dma, -unlocked, true);
971 static void vfio_remove_dma(struct vfio_iommu *iommu, struct vfio_dma *dma)
973 vfio_unmap_unpin(iommu, dma, true);
974 vfio_unlink_dma(iommu, dma);
975 put_task_struct(dma->task);
976 vfio_dma_bitmap_free(dma);
981 static void vfio_update_pgsize_bitmap(struct vfio_iommu *iommu)
983 struct vfio_domain *domain;
985 iommu->pgsize_bitmap = ULONG_MAX;
987 list_for_each_entry(domain, &iommu->domain_list, next)
988 iommu->pgsize_bitmap &= domain->domain->pgsize_bitmap;
991 * In case the IOMMU supports page sizes smaller than PAGE_SIZE
992 * we pretend PAGE_SIZE is supported and hide sub-PAGE_SIZE sizes.
993 * That way the user will be able to map/unmap buffers whose size/
994 * start address is aligned with PAGE_SIZE. Pinning code uses that
995 * granularity while iommu driver can use the sub-PAGE_SIZE size
998 if (iommu->pgsize_bitmap & ~PAGE_MASK) {
999 iommu->pgsize_bitmap &= PAGE_MASK;
1000 iommu->pgsize_bitmap |= PAGE_SIZE;
1004 static int update_user_bitmap(u64 __user *bitmap, struct vfio_iommu *iommu,
1005 struct vfio_dma *dma, dma_addr_t base_iova,
1008 unsigned long pgshift = __ffs(pgsize);
1009 unsigned long nbits = dma->size >> pgshift;
1010 unsigned long bit_offset = (dma->iova - base_iova) >> pgshift;
1011 unsigned long copy_offset = bit_offset / BITS_PER_LONG;
1012 unsigned long shift = bit_offset % BITS_PER_LONG;
1013 unsigned long leftover;
1016 * mark all pages dirty if any IOMMU capable device is not able
1017 * to report dirty pages and all pages are pinned and mapped.
1019 if (!iommu->pinned_page_dirty_scope && dma->iommu_mapped)
1020 bitmap_set(dma->bitmap, 0, nbits);
1023 bitmap_shift_left(dma->bitmap, dma->bitmap, shift,
1026 if (copy_from_user(&leftover,
1027 (void __user *)(bitmap + copy_offset),
1031 bitmap_or(dma->bitmap, dma->bitmap, &leftover, shift);
1034 if (copy_to_user((void __user *)(bitmap + copy_offset), dma->bitmap,
1035 DIRTY_BITMAP_BYTES(nbits + shift)))
1041 static int vfio_iova_dirty_bitmap(u64 __user *bitmap, struct vfio_iommu *iommu,
1042 dma_addr_t iova, size_t size, size_t pgsize)
1044 struct vfio_dma *dma;
1046 unsigned long pgshift = __ffs(pgsize);
1050 * GET_BITMAP request must fully cover vfio_dma mappings. Multiple
1051 * vfio_dma mappings may be clubbed by specifying large ranges, but
1052 * there must not be any previous mappings bisected by the range.
1053 * An error will be returned if these conditions are not met.
1055 dma = vfio_find_dma(iommu, iova, 1);
1056 if (dma && dma->iova != iova)
1059 dma = vfio_find_dma(iommu, iova + size - 1, 0);
1060 if (dma && dma->iova + dma->size != iova + size)
1063 for (n = rb_first(&iommu->dma_list); n; n = rb_next(n)) {
1064 struct vfio_dma *dma = rb_entry(n, struct vfio_dma, node);
1066 if (dma->iova < iova)
1069 if (dma->iova > iova + size - 1)
1072 ret = update_user_bitmap(bitmap, iommu, dma, iova, pgsize);
1077 * Re-populate bitmap to include all pinned pages which are
1078 * considered as dirty but exclude pages which are unpinned and
1079 * pages which are marked dirty by vfio_dma_rw()
1081 bitmap_clear(dma->bitmap, 0, dma->size >> pgshift);
1082 vfio_dma_populate_bitmap(dma, pgsize);
1087 static int verify_bitmap_size(uint64_t npages, uint64_t bitmap_size)
1089 if (!npages || !bitmap_size || (bitmap_size > DIRTY_BITMAP_SIZE_MAX) ||
1090 (bitmap_size < DIRTY_BITMAP_BYTES(npages)))
1096 static int vfio_dma_do_unmap(struct vfio_iommu *iommu,
1097 struct vfio_iommu_type1_dma_unmap *unmap,
1098 struct vfio_bitmap *bitmap)
1100 struct vfio_dma *dma, *dma_last = NULL;
1101 size_t unmapped = 0, pgsize;
1102 int ret = -EINVAL, retries = 0;
1103 unsigned long pgshift;
1104 dma_addr_t iova = unmap->iova;
1105 unsigned long size = unmap->size;
1106 bool unmap_all = unmap->flags & VFIO_DMA_UNMAP_FLAG_ALL;
1109 mutex_lock(&iommu->lock);
1111 pgshift = __ffs(iommu->pgsize_bitmap);
1112 pgsize = (size_t)1 << pgshift;
1114 if (iova & (pgsize - 1))
1121 } else if (!size || size & (pgsize - 1)) {
1125 if (iova + size - 1 < iova || size > SIZE_MAX)
1128 /* When dirty tracking is enabled, allow only min supported pgsize */
1129 if ((unmap->flags & VFIO_DMA_UNMAP_FLAG_GET_DIRTY_BITMAP) &&
1130 (!iommu->dirty_page_tracking || (bitmap->pgsize != pgsize))) {
1134 WARN_ON((pgsize - 1) & PAGE_MASK);
1137 * vfio-iommu-type1 (v1) - User mappings were coalesced together to
1138 * avoid tracking individual mappings. This means that the granularity
1139 * of the original mapping was lost and the user was allowed to attempt
1140 * to unmap any range. Depending on the contiguousness of physical
1141 * memory and page sizes supported by the IOMMU, arbitrary unmaps may
1142 * or may not have worked. We only guaranteed unmap granularity
1143 * matching the original mapping; even though it was untracked here,
1144 * the original mappings are reflected in IOMMU mappings. This
1145 * resulted in a couple unusual behaviors. First, if a range is not
1146 * able to be unmapped, ex. a set of 4k pages that was mapped as a
1147 * 2M hugepage into the IOMMU, the unmap ioctl returns success but with
1148 * a zero sized unmap. Also, if an unmap request overlaps the first
1149 * address of a hugepage, the IOMMU will unmap the entire hugepage.
1150 * This also returns success and the returned unmap size reflects the
1151 * actual size unmapped.
1153 * We attempt to maintain compatibility with this "v1" interface, but
1154 * we take control out of the hands of the IOMMU. Therefore, an unmap
1155 * request offset from the beginning of the original mapping will
1156 * return success with zero sized unmap. And an unmap request covering
1157 * the first iova of mapping will unmap the entire range.
1159 * The v2 version of this interface intends to be more deterministic.
1160 * Unmap requests must fully cover previous mappings. Multiple
1161 * mappings may still be unmaped by specifying large ranges, but there
1162 * must not be any previous mappings bisected by the range. An error
1163 * will be returned if these conditions are not met. The v2 interface
1164 * will only return success and a size of zero if there were no
1165 * mappings within the range.
1167 if (iommu->v2 && !unmap_all) {
1168 dma = vfio_find_dma(iommu, iova, 1);
1169 if (dma && dma->iova != iova)
1172 dma = vfio_find_dma(iommu, iova + size - 1, 0);
1173 if (dma && dma->iova + dma->size != iova + size)
1178 n = vfio_find_dma_first_node(iommu, iova, size);
1181 dma = rb_entry(n, struct vfio_dma, node);
1182 if (dma->iova >= iova + size)
1185 if (!iommu->v2 && iova > dma->iova)
1188 * Task with same address space who mapped this iova range is
1189 * allowed to unmap the iova range.
1191 if (dma->task->mm != current->mm)
1194 if (!RB_EMPTY_ROOT(&dma->pfn_list)) {
1195 struct vfio_iommu_type1_dma_unmap nb_unmap;
1197 if (dma_last == dma) {
1198 BUG_ON(++retries > 10);
1204 nb_unmap.iova = dma->iova;
1205 nb_unmap.size = dma->size;
1208 * Notify anyone (mdev vendor drivers) to invalidate and
1209 * unmap iovas within the range we're about to unmap.
1210 * Vendor drivers MUST unpin pages in response to an
1213 mutex_unlock(&iommu->lock);
1214 blocking_notifier_call_chain(&iommu->notifier,
1215 VFIO_IOMMU_NOTIFY_DMA_UNMAP,
1217 mutex_lock(&iommu->lock);
1221 if (unmap->flags & VFIO_DMA_UNMAP_FLAG_GET_DIRTY_BITMAP) {
1222 ret = update_user_bitmap(bitmap->data, iommu, dma,
1228 unmapped += dma->size;
1230 vfio_remove_dma(iommu, dma);
1234 mutex_unlock(&iommu->lock);
1236 /* Report how much was unmapped */
1237 unmap->size = unmapped;
1242 static int vfio_iommu_map(struct vfio_iommu *iommu, dma_addr_t iova,
1243 unsigned long pfn, long npage, int prot)
1245 struct vfio_domain *d;
1248 list_for_each_entry(d, &iommu->domain_list, next) {
1249 ret = iommu_map(d->domain, iova, (phys_addr_t)pfn << PAGE_SHIFT,
1250 npage << PAGE_SHIFT, prot | d->prot);
1260 list_for_each_entry_continue_reverse(d, &iommu->domain_list, next) {
1261 iommu_unmap(d->domain, iova, npage << PAGE_SHIFT);
1268 static int vfio_pin_map_dma(struct vfio_iommu *iommu, struct vfio_dma *dma,
1271 dma_addr_t iova = dma->iova;
1272 unsigned long vaddr = dma->vaddr;
1273 size_t size = map_size;
1275 unsigned long pfn, limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1279 /* Pin a contiguous chunk of memory */
1280 npage = vfio_pin_pages_remote(dma, vaddr + dma->size,
1281 size >> PAGE_SHIFT, &pfn, limit);
1289 ret = vfio_iommu_map(iommu, iova + dma->size, pfn, npage,
1292 vfio_unpin_pages_remote(dma, iova + dma->size, pfn,
1297 size -= npage << PAGE_SHIFT;
1298 dma->size += npage << PAGE_SHIFT;
1301 dma->iommu_mapped = true;
1304 vfio_remove_dma(iommu, dma);
1310 * Check dma map request is within a valid iova range
1312 static bool vfio_iommu_iova_dma_valid(struct vfio_iommu *iommu,
1313 dma_addr_t start, dma_addr_t end)
1315 struct list_head *iova = &iommu->iova_list;
1316 struct vfio_iova *node;
1318 list_for_each_entry(node, iova, list) {
1319 if (start >= node->start && end <= node->end)
1324 * Check for list_empty() as well since a container with
1325 * a single mdev device will have an empty list.
1327 return list_empty(iova);
1330 static int vfio_dma_do_map(struct vfio_iommu *iommu,
1331 struct vfio_iommu_type1_dma_map *map)
1333 dma_addr_t iova = map->iova;
1334 unsigned long vaddr = map->vaddr;
1335 size_t size = map->size;
1336 int ret = 0, prot = 0;
1338 struct vfio_dma *dma;
1340 /* Verify that none of our __u64 fields overflow */
1341 if (map->size != size || map->vaddr != vaddr || map->iova != iova)
1344 /* READ/WRITE from device perspective */
1345 if (map->flags & VFIO_DMA_MAP_FLAG_WRITE)
1346 prot |= IOMMU_WRITE;
1347 if (map->flags & VFIO_DMA_MAP_FLAG_READ)
1350 mutex_lock(&iommu->lock);
1352 pgsize = (size_t)1 << __ffs(iommu->pgsize_bitmap);
1354 WARN_ON((pgsize - 1) & PAGE_MASK);
1356 if (!prot || !size || (size | iova | vaddr) & (pgsize - 1)) {
1361 /* Don't allow IOVA or virtual address wrap */
1362 if (iova + size - 1 < iova || vaddr + size - 1 < vaddr) {
1367 if (vfio_find_dma(iommu, iova, size)) {
1372 if (!iommu->dma_avail) {
1377 if (!vfio_iommu_iova_dma_valid(iommu, iova, iova + size - 1)) {
1382 dma = kzalloc(sizeof(*dma), GFP_KERNEL);
1394 * We need to be able to both add to a task's locked memory and test
1395 * against the locked memory limit and we need to be able to do both
1396 * outside of this call path as pinning can be asynchronous via the
1397 * external interfaces for mdev devices. RLIMIT_MEMLOCK requires a
1398 * task_struct and VM locked pages requires an mm_struct, however
1399 * holding an indefinite mm reference is not recommended, therefore we
1400 * only hold a reference to a task. We could hold a reference to
1401 * current, however QEMU uses this call path through vCPU threads,
1402 * which can be killed resulting in a NULL mm and failure in the unmap
1403 * path when called via a different thread. Avoid this problem by
1404 * using the group_leader as threads within the same group require
1405 * both CLONE_THREAD and CLONE_VM and will therefore use the same
1408 * Previously we also used the task for testing CAP_IPC_LOCK at the
1409 * time of pinning and accounting, however has_capability() makes use
1410 * of real_cred, a copy-on-write field, so we can't guarantee that it
1411 * matches group_leader, or in fact that it might not change by the
1412 * time it's evaluated. If a process were to call MAP_DMA with
1413 * CAP_IPC_LOCK but later drop it, it doesn't make sense that they
1414 * possibly see different results for an iommu_mapped vfio_dma vs
1415 * externally mapped. Therefore track CAP_IPC_LOCK in vfio_dma at the
1416 * time of calling MAP_DMA.
1418 get_task_struct(current->group_leader);
1419 dma->task = current->group_leader;
1420 dma->lock_cap = capable(CAP_IPC_LOCK);
1422 dma->pfn_list = RB_ROOT;
1424 /* Insert zero-sized and grow as we map chunks of it */
1425 vfio_link_dma(iommu, dma);
1427 /* Don't pin and map if container doesn't contain IOMMU capable domain*/
1428 if (!IS_IOMMU_CAP_DOMAIN_IN_CONTAINER(iommu))
1431 ret = vfio_pin_map_dma(iommu, dma, size);
1433 if (!ret && iommu->dirty_page_tracking) {
1434 ret = vfio_dma_bitmap_alloc(dma, pgsize);
1436 vfio_remove_dma(iommu, dma);
1440 mutex_unlock(&iommu->lock);
1444 static int vfio_bus_type(struct device *dev, void *data)
1446 struct bus_type **bus = data;
1448 if (*bus && *bus != dev->bus)
1456 static int vfio_iommu_replay(struct vfio_iommu *iommu,
1457 struct vfio_domain *domain)
1459 struct vfio_domain *d = NULL;
1461 unsigned long limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1464 /* Arbitrarily pick the first domain in the list for lookups */
1465 if (!list_empty(&iommu->domain_list))
1466 d = list_first_entry(&iommu->domain_list,
1467 struct vfio_domain, next);
1469 n = rb_first(&iommu->dma_list);
1471 for (; n; n = rb_next(n)) {
1472 struct vfio_dma *dma;
1475 dma = rb_entry(n, struct vfio_dma, node);
1478 while (iova < dma->iova + dma->size) {
1482 if (dma->iommu_mapped) {
1486 if (WARN_ON(!d)) { /* mapped w/o a domain?! */
1491 phys = iommu_iova_to_phys(d->domain, iova);
1493 if (WARN_ON(!phys)) {
1501 while (i < dma->iova + dma->size &&
1502 p == iommu_iova_to_phys(d->domain, i)) {
1509 unsigned long vaddr = dma->vaddr +
1511 size_t n = dma->iova + dma->size - iova;
1514 npage = vfio_pin_pages_remote(dma, vaddr,
1523 phys = pfn << PAGE_SHIFT;
1524 size = npage << PAGE_SHIFT;
1527 ret = iommu_map(domain->domain, iova, phys,
1528 size, dma->prot | domain->prot);
1530 if (!dma->iommu_mapped)
1531 vfio_unpin_pages_remote(dma, iova,
1542 /* All dmas are now mapped, defer to second tree walk for unwind */
1543 for (n = rb_first(&iommu->dma_list); n; n = rb_next(n)) {
1544 struct vfio_dma *dma = rb_entry(n, struct vfio_dma, node);
1546 dma->iommu_mapped = true;
1552 for (; n; n = rb_prev(n)) {
1553 struct vfio_dma *dma = rb_entry(n, struct vfio_dma, node);
1556 if (dma->iommu_mapped) {
1557 iommu_unmap(domain->domain, dma->iova, dma->size);
1562 while (iova < dma->iova + dma->size) {
1563 phys_addr_t phys, p;
1567 phys = iommu_iova_to_phys(domain->domain, iova);
1576 while (i < dma->iova + dma->size &&
1577 p == iommu_iova_to_phys(domain->domain, i)) {
1583 iommu_unmap(domain->domain, iova, size);
1584 vfio_unpin_pages_remote(dma, iova, phys >> PAGE_SHIFT,
1585 size >> PAGE_SHIFT, true);
1593 * We change our unmap behavior slightly depending on whether the IOMMU
1594 * supports fine-grained superpages. IOMMUs like AMD-Vi will use a superpage
1595 * for practically any contiguous power-of-two mapping we give it. This means
1596 * we don't need to look for contiguous chunks ourselves to make unmapping
1597 * more efficient. On IOMMUs with coarse-grained super pages, like Intel VT-d
1598 * with discrete 2M/1G/512G/1T superpages, identifying contiguous chunks
1599 * significantly boosts non-hugetlbfs mappings and doesn't seem to hurt when
1600 * hugetlbfs is in use.
1602 static void vfio_test_domain_fgsp(struct vfio_domain *domain)
1605 int ret, order = get_order(PAGE_SIZE * 2);
1607 pages = alloc_pages(GFP_KERNEL | __GFP_ZERO, order);
1611 ret = iommu_map(domain->domain, 0, page_to_phys(pages), PAGE_SIZE * 2,
1612 IOMMU_READ | IOMMU_WRITE | domain->prot);
1614 size_t unmapped = iommu_unmap(domain->domain, 0, PAGE_SIZE);
1616 if (unmapped == PAGE_SIZE)
1617 iommu_unmap(domain->domain, PAGE_SIZE, PAGE_SIZE);
1619 domain->fgsp = true;
1622 __free_pages(pages, order);
1625 static struct vfio_group *find_iommu_group(struct vfio_domain *domain,
1626 struct iommu_group *iommu_group)
1628 struct vfio_group *g;
1630 list_for_each_entry(g, &domain->group_list, next) {
1631 if (g->iommu_group == iommu_group)
1638 static struct vfio_group *vfio_iommu_find_iommu_group(struct vfio_iommu *iommu,
1639 struct iommu_group *iommu_group)
1641 struct vfio_domain *domain;
1642 struct vfio_group *group = NULL;
1644 list_for_each_entry(domain, &iommu->domain_list, next) {
1645 group = find_iommu_group(domain, iommu_group);
1650 if (iommu->external_domain)
1651 group = find_iommu_group(iommu->external_domain, iommu_group);
1656 static void update_pinned_page_dirty_scope(struct vfio_iommu *iommu)
1658 struct vfio_domain *domain;
1659 struct vfio_group *group;
1661 list_for_each_entry(domain, &iommu->domain_list, next) {
1662 list_for_each_entry(group, &domain->group_list, next) {
1663 if (!group->pinned_page_dirty_scope) {
1664 iommu->pinned_page_dirty_scope = false;
1670 if (iommu->external_domain) {
1671 domain = iommu->external_domain;
1672 list_for_each_entry(group, &domain->group_list, next) {
1673 if (!group->pinned_page_dirty_scope) {
1674 iommu->pinned_page_dirty_scope = false;
1680 iommu->pinned_page_dirty_scope = true;
1683 static bool vfio_iommu_has_sw_msi(struct list_head *group_resv_regions,
1686 struct iommu_resv_region *region;
1689 list_for_each_entry(region, group_resv_regions, list) {
1691 * The presence of any 'real' MSI regions should take
1692 * precedence over the software-managed one if the
1693 * IOMMU driver happens to advertise both types.
1695 if (region->type == IOMMU_RESV_MSI) {
1700 if (region->type == IOMMU_RESV_SW_MSI) {
1701 *base = region->start;
1709 static struct device *vfio_mdev_get_iommu_device(struct device *dev)
1711 struct device *(*fn)(struct device *dev);
1712 struct device *iommu_device;
1714 fn = symbol_get(mdev_get_iommu_device);
1716 iommu_device = fn(dev);
1717 symbol_put(mdev_get_iommu_device);
1719 return iommu_device;
1725 static int vfio_mdev_attach_domain(struct device *dev, void *data)
1727 struct iommu_domain *domain = data;
1728 struct device *iommu_device;
1730 iommu_device = vfio_mdev_get_iommu_device(dev);
1732 if (iommu_dev_feature_enabled(iommu_device, IOMMU_DEV_FEAT_AUX))
1733 return iommu_aux_attach_device(domain, iommu_device);
1735 return iommu_attach_device(domain, iommu_device);
1741 static int vfio_mdev_detach_domain(struct device *dev, void *data)
1743 struct iommu_domain *domain = data;
1744 struct device *iommu_device;
1746 iommu_device = vfio_mdev_get_iommu_device(dev);
1748 if (iommu_dev_feature_enabled(iommu_device, IOMMU_DEV_FEAT_AUX))
1749 iommu_aux_detach_device(domain, iommu_device);
1751 iommu_detach_device(domain, iommu_device);
1757 static int vfio_iommu_attach_group(struct vfio_domain *domain,
1758 struct vfio_group *group)
1760 if (group->mdev_group)
1761 return iommu_group_for_each_dev(group->iommu_group,
1763 vfio_mdev_attach_domain);
1765 return iommu_attach_group(domain->domain, group->iommu_group);
1768 static void vfio_iommu_detach_group(struct vfio_domain *domain,
1769 struct vfio_group *group)
1771 if (group->mdev_group)
1772 iommu_group_for_each_dev(group->iommu_group, domain->domain,
1773 vfio_mdev_detach_domain);
1775 iommu_detach_group(domain->domain, group->iommu_group);
1778 static bool vfio_bus_is_mdev(struct bus_type *bus)
1780 struct bus_type *mdev_bus;
1783 mdev_bus = symbol_get(mdev_bus_type);
1785 ret = (bus == mdev_bus);
1786 symbol_put(mdev_bus_type);
1792 static int vfio_mdev_iommu_device(struct device *dev, void *data)
1794 struct device **old = data, *new;
1796 new = vfio_mdev_get_iommu_device(dev);
1797 if (!new || (*old && *old != new))
1806 * This is a helper function to insert an address range to iova list.
1807 * The list is initially created with a single entry corresponding to
1808 * the IOMMU domain geometry to which the device group is attached.
1809 * The list aperture gets modified when a new domain is added to the
1810 * container if the new aperture doesn't conflict with the current one
1811 * or with any existing dma mappings. The list is also modified to
1812 * exclude any reserved regions associated with the device group.
1814 static int vfio_iommu_iova_insert(struct list_head *head,
1815 dma_addr_t start, dma_addr_t end)
1817 struct vfio_iova *region;
1819 region = kmalloc(sizeof(*region), GFP_KERNEL);
1823 INIT_LIST_HEAD(®ion->list);
1824 region->start = start;
1827 list_add_tail(®ion->list, head);
1832 * Check the new iommu aperture conflicts with existing aper or with any
1833 * existing dma mappings.
1835 static bool vfio_iommu_aper_conflict(struct vfio_iommu *iommu,
1836 dma_addr_t start, dma_addr_t end)
1838 struct vfio_iova *first, *last;
1839 struct list_head *iova = &iommu->iova_list;
1841 if (list_empty(iova))
1844 /* Disjoint sets, return conflict */
1845 first = list_first_entry(iova, struct vfio_iova, list);
1846 last = list_last_entry(iova, struct vfio_iova, list);
1847 if (start > last->end || end < first->start)
1850 /* Check for any existing dma mappings below the new start */
1851 if (start > first->start) {
1852 if (vfio_find_dma(iommu, first->start, start - first->start))
1856 /* Check for any existing dma mappings beyond the new end */
1857 if (end < last->end) {
1858 if (vfio_find_dma(iommu, end + 1, last->end - end))
1866 * Resize iommu iova aperture window. This is called only if the new
1867 * aperture has no conflict with existing aperture and dma mappings.
1869 static int vfio_iommu_aper_resize(struct list_head *iova,
1870 dma_addr_t start, dma_addr_t end)
1872 struct vfio_iova *node, *next;
1874 if (list_empty(iova))
1875 return vfio_iommu_iova_insert(iova, start, end);
1877 /* Adjust iova list start */
1878 list_for_each_entry_safe(node, next, iova, list) {
1879 if (start < node->start)
1881 if (start >= node->start && start < node->end) {
1882 node->start = start;
1885 /* Delete nodes before new start */
1886 list_del(&node->list);
1890 /* Adjust iova list end */
1891 list_for_each_entry_safe(node, next, iova, list) {
1892 if (end > node->end)
1894 if (end > node->start && end <= node->end) {
1898 /* Delete nodes after new end */
1899 list_del(&node->list);
1907 * Check reserved region conflicts with existing dma mappings
1909 static bool vfio_iommu_resv_conflict(struct vfio_iommu *iommu,
1910 struct list_head *resv_regions)
1912 struct iommu_resv_region *region;
1914 /* Check for conflict with existing dma mappings */
1915 list_for_each_entry(region, resv_regions, list) {
1916 if (region->type == IOMMU_RESV_DIRECT_RELAXABLE)
1919 if (vfio_find_dma(iommu, region->start, region->length))
1927 * Check iova region overlap with reserved regions and
1928 * exclude them from the iommu iova range
1930 static int vfio_iommu_resv_exclude(struct list_head *iova,
1931 struct list_head *resv_regions)
1933 struct iommu_resv_region *resv;
1934 struct vfio_iova *n, *next;
1936 list_for_each_entry(resv, resv_regions, list) {
1937 phys_addr_t start, end;
1939 if (resv->type == IOMMU_RESV_DIRECT_RELAXABLE)
1942 start = resv->start;
1943 end = resv->start + resv->length - 1;
1945 list_for_each_entry_safe(n, next, iova, list) {
1949 if (start > n->end || end < n->start)
1952 * Insert a new node if current node overlaps with the
1953 * reserve region to exlude that from valid iova range.
1954 * Note that, new node is inserted before the current
1955 * node and finally the current node is deleted keeping
1956 * the list updated and sorted.
1958 if (start > n->start)
1959 ret = vfio_iommu_iova_insert(&n->list, n->start,
1961 if (!ret && end < n->end)
1962 ret = vfio_iommu_iova_insert(&n->list, end + 1,
1972 if (list_empty(iova))
1978 static void vfio_iommu_resv_free(struct list_head *resv_regions)
1980 struct iommu_resv_region *n, *next;
1982 list_for_each_entry_safe(n, next, resv_regions, list) {
1988 static void vfio_iommu_iova_free(struct list_head *iova)
1990 struct vfio_iova *n, *next;
1992 list_for_each_entry_safe(n, next, iova, list) {
1998 static int vfio_iommu_iova_get_copy(struct vfio_iommu *iommu,
1999 struct list_head *iova_copy)
2001 struct list_head *iova = &iommu->iova_list;
2002 struct vfio_iova *n;
2005 list_for_each_entry(n, iova, list) {
2006 ret = vfio_iommu_iova_insert(iova_copy, n->start, n->end);
2014 vfio_iommu_iova_free(iova_copy);
2018 static void vfio_iommu_iova_insert_copy(struct vfio_iommu *iommu,
2019 struct list_head *iova_copy)
2021 struct list_head *iova = &iommu->iova_list;
2023 vfio_iommu_iova_free(iova);
2025 list_splice_tail(iova_copy, iova);
2028 static int vfio_iommu_type1_attach_group(void *iommu_data,
2029 struct iommu_group *iommu_group)
2031 struct vfio_iommu *iommu = iommu_data;
2032 struct vfio_group *group;
2033 struct vfio_domain *domain, *d;
2034 struct bus_type *bus = NULL;
2036 bool resv_msi, msi_remap;
2037 phys_addr_t resv_msi_base = 0;
2038 struct iommu_domain_geometry geo;
2039 LIST_HEAD(iova_copy);
2040 LIST_HEAD(group_resv_regions);
2042 mutex_lock(&iommu->lock);
2044 /* Check for duplicates */
2045 if (vfio_iommu_find_iommu_group(iommu, iommu_group)) {
2046 mutex_unlock(&iommu->lock);
2050 group = kzalloc(sizeof(*group), GFP_KERNEL);
2051 domain = kzalloc(sizeof(*domain), GFP_KERNEL);
2052 if (!group || !domain) {
2057 group->iommu_group = iommu_group;
2059 /* Determine bus_type in order to allocate a domain */
2060 ret = iommu_group_for_each_dev(iommu_group, &bus, vfio_bus_type);
2064 if (vfio_bus_is_mdev(bus)) {
2065 struct device *iommu_device = NULL;
2067 group->mdev_group = true;
2069 /* Determine the isolation type */
2070 ret = iommu_group_for_each_dev(iommu_group, &iommu_device,
2071 vfio_mdev_iommu_device);
2072 if (ret || !iommu_device) {
2073 if (!iommu->external_domain) {
2074 INIT_LIST_HEAD(&domain->group_list);
2075 iommu->external_domain = domain;
2076 vfio_update_pgsize_bitmap(iommu);
2081 list_add(&group->next,
2082 &iommu->external_domain->group_list);
2084 * Non-iommu backed group cannot dirty memory directly,
2085 * it can only use interfaces that provide dirty
2087 * The iommu scope can only be promoted with the
2088 * addition of a dirty tracking group.
2090 group->pinned_page_dirty_scope = true;
2091 if (!iommu->pinned_page_dirty_scope)
2092 update_pinned_page_dirty_scope(iommu);
2093 mutex_unlock(&iommu->lock);
2098 bus = iommu_device->bus;
2101 domain->domain = iommu_domain_alloc(bus);
2102 if (!domain->domain) {
2107 if (iommu->nesting) {
2110 ret = iommu_domain_set_attr(domain->domain, DOMAIN_ATTR_NESTING,
2116 ret = vfio_iommu_attach_group(domain, group);
2120 /* Get aperture info */
2121 iommu_domain_get_attr(domain->domain, DOMAIN_ATTR_GEOMETRY, &geo);
2123 if (vfio_iommu_aper_conflict(iommu, geo.aperture_start,
2124 geo.aperture_end)) {
2129 ret = iommu_get_group_resv_regions(iommu_group, &group_resv_regions);
2133 if (vfio_iommu_resv_conflict(iommu, &group_resv_regions)) {
2139 * We don't want to work on the original iova list as the list
2140 * gets modified and in case of failure we have to retain the
2141 * original list. Get a copy here.
2143 ret = vfio_iommu_iova_get_copy(iommu, &iova_copy);
2147 ret = vfio_iommu_aper_resize(&iova_copy, geo.aperture_start,
2152 ret = vfio_iommu_resv_exclude(&iova_copy, &group_resv_regions);
2156 resv_msi = vfio_iommu_has_sw_msi(&group_resv_regions, &resv_msi_base);
2158 INIT_LIST_HEAD(&domain->group_list);
2159 list_add(&group->next, &domain->group_list);
2161 msi_remap = irq_domain_check_msi_remap() ||
2162 iommu_capable(bus, IOMMU_CAP_INTR_REMAP);
2164 if (!allow_unsafe_interrupts && !msi_remap) {
2165 pr_warn("%s: No interrupt remapping support. Use the module param \"allow_unsafe_interrupts\" to enable VFIO IOMMU support on this platform\n",
2171 if (iommu_capable(bus, IOMMU_CAP_CACHE_COHERENCY))
2172 domain->prot |= IOMMU_CACHE;
2175 * Try to match an existing compatible domain. We don't want to
2176 * preclude an IOMMU driver supporting multiple bus_types and being
2177 * able to include different bus_types in the same IOMMU domain, so
2178 * we test whether the domains use the same iommu_ops rather than
2179 * testing if they're on the same bus_type.
2181 list_for_each_entry(d, &iommu->domain_list, next) {
2182 if (d->domain->ops == domain->domain->ops &&
2183 d->prot == domain->prot) {
2184 vfio_iommu_detach_group(domain, group);
2185 if (!vfio_iommu_attach_group(d, group)) {
2186 list_add(&group->next, &d->group_list);
2187 iommu_domain_free(domain->domain);
2192 ret = vfio_iommu_attach_group(domain, group);
2198 vfio_test_domain_fgsp(domain);
2200 /* replay mappings on new domains */
2201 ret = vfio_iommu_replay(iommu, domain);
2206 ret = iommu_get_msi_cookie(domain->domain, resv_msi_base);
2207 if (ret && ret != -ENODEV)
2211 list_add(&domain->next, &iommu->domain_list);
2212 vfio_update_pgsize_bitmap(iommu);
2214 /* Delete the old one and insert new iova list */
2215 vfio_iommu_iova_insert_copy(iommu, &iova_copy);
2218 * An iommu backed group can dirty memory directly and therefore
2219 * demotes the iommu scope until it declares itself dirty tracking
2220 * capable via the page pinning interface.
2222 iommu->pinned_page_dirty_scope = false;
2223 mutex_unlock(&iommu->lock);
2224 vfio_iommu_resv_free(&group_resv_regions);
2229 vfio_iommu_detach_group(domain, group);
2231 iommu_domain_free(domain->domain);
2232 vfio_iommu_iova_free(&iova_copy);
2233 vfio_iommu_resv_free(&group_resv_regions);
2237 mutex_unlock(&iommu->lock);
2241 static void vfio_iommu_unmap_unpin_all(struct vfio_iommu *iommu)
2243 struct rb_node *node;
2245 while ((node = rb_first(&iommu->dma_list)))
2246 vfio_remove_dma(iommu, rb_entry(node, struct vfio_dma, node));
2249 static void vfio_iommu_unmap_unpin_reaccount(struct vfio_iommu *iommu)
2251 struct rb_node *n, *p;
2253 n = rb_first(&iommu->dma_list);
2254 for (; n; n = rb_next(n)) {
2255 struct vfio_dma *dma;
2256 long locked = 0, unlocked = 0;
2258 dma = rb_entry(n, struct vfio_dma, node);
2259 unlocked += vfio_unmap_unpin(iommu, dma, false);
2260 p = rb_first(&dma->pfn_list);
2261 for (; p; p = rb_next(p)) {
2262 struct vfio_pfn *vpfn = rb_entry(p, struct vfio_pfn,
2265 if (!is_invalid_reserved_pfn(vpfn->pfn))
2268 vfio_lock_acct(dma, locked - unlocked, true);
2272 static void vfio_sanity_check_pfn_list(struct vfio_iommu *iommu)
2276 n = rb_first(&iommu->dma_list);
2277 for (; n; n = rb_next(n)) {
2278 struct vfio_dma *dma;
2280 dma = rb_entry(n, struct vfio_dma, node);
2282 if (WARN_ON(!RB_EMPTY_ROOT(&dma->pfn_list)))
2285 /* mdev vendor driver must unregister notifier */
2286 WARN_ON(iommu->notifier.head);
2290 * Called when a domain is removed in detach. It is possible that
2291 * the removed domain decided the iova aperture window. Modify the
2292 * iova aperture with the smallest window among existing domains.
2294 static void vfio_iommu_aper_expand(struct vfio_iommu *iommu,
2295 struct list_head *iova_copy)
2297 struct vfio_domain *domain;
2298 struct iommu_domain_geometry geo;
2299 struct vfio_iova *node;
2300 dma_addr_t start = 0;
2301 dma_addr_t end = (dma_addr_t)~0;
2303 if (list_empty(iova_copy))
2306 list_for_each_entry(domain, &iommu->domain_list, next) {
2307 iommu_domain_get_attr(domain->domain, DOMAIN_ATTR_GEOMETRY,
2309 if (geo.aperture_start > start)
2310 start = geo.aperture_start;
2311 if (geo.aperture_end < end)
2312 end = geo.aperture_end;
2315 /* Modify aperture limits. The new aper is either same or bigger */
2316 node = list_first_entry(iova_copy, struct vfio_iova, list);
2317 node->start = start;
2318 node = list_last_entry(iova_copy, struct vfio_iova, list);
2323 * Called when a group is detached. The reserved regions for that
2324 * group can be part of valid iova now. But since reserved regions
2325 * may be duplicated among groups, populate the iova valid regions
2328 static int vfio_iommu_resv_refresh(struct vfio_iommu *iommu,
2329 struct list_head *iova_copy)
2331 struct vfio_domain *d;
2332 struct vfio_group *g;
2333 struct vfio_iova *node;
2334 dma_addr_t start, end;
2335 LIST_HEAD(resv_regions);
2338 if (list_empty(iova_copy))
2341 list_for_each_entry(d, &iommu->domain_list, next) {
2342 list_for_each_entry(g, &d->group_list, next) {
2343 ret = iommu_get_group_resv_regions(g->iommu_group,
2350 node = list_first_entry(iova_copy, struct vfio_iova, list);
2351 start = node->start;
2352 node = list_last_entry(iova_copy, struct vfio_iova, list);
2355 /* purge the iova list and create new one */
2356 vfio_iommu_iova_free(iova_copy);
2358 ret = vfio_iommu_aper_resize(iova_copy, start, end);
2362 /* Exclude current reserved regions from iova ranges */
2363 ret = vfio_iommu_resv_exclude(iova_copy, &resv_regions);
2365 vfio_iommu_resv_free(&resv_regions);
2369 static void vfio_iommu_type1_detach_group(void *iommu_data,
2370 struct iommu_group *iommu_group)
2372 struct vfio_iommu *iommu = iommu_data;
2373 struct vfio_domain *domain;
2374 struct vfio_group *group;
2375 bool update_dirty_scope = false;
2376 LIST_HEAD(iova_copy);
2378 mutex_lock(&iommu->lock);
2380 if (iommu->external_domain) {
2381 group = find_iommu_group(iommu->external_domain, iommu_group);
2383 update_dirty_scope = !group->pinned_page_dirty_scope;
2384 list_del(&group->next);
2387 if (list_empty(&iommu->external_domain->group_list)) {
2388 vfio_sanity_check_pfn_list(iommu);
2390 if (!IS_IOMMU_CAP_DOMAIN_IN_CONTAINER(iommu))
2391 vfio_iommu_unmap_unpin_all(iommu);
2393 kfree(iommu->external_domain);
2394 iommu->external_domain = NULL;
2396 goto detach_group_done;
2401 * Get a copy of iova list. This will be used to update
2402 * and to replace the current one later. Please note that
2403 * we will leave the original list as it is if update fails.
2405 vfio_iommu_iova_get_copy(iommu, &iova_copy);
2407 list_for_each_entry(domain, &iommu->domain_list, next) {
2408 group = find_iommu_group(domain, iommu_group);
2412 vfio_iommu_detach_group(domain, group);
2413 update_dirty_scope = !group->pinned_page_dirty_scope;
2414 list_del(&group->next);
2417 * Group ownership provides privilege, if the group list is
2418 * empty, the domain goes away. If it's the last domain with
2419 * iommu and external domain doesn't exist, then all the
2420 * mappings go away too. If it's the last domain with iommu and
2421 * external domain exist, update accounting
2423 if (list_empty(&domain->group_list)) {
2424 if (list_is_singular(&iommu->domain_list)) {
2425 if (!iommu->external_domain)
2426 vfio_iommu_unmap_unpin_all(iommu);
2428 vfio_iommu_unmap_unpin_reaccount(iommu);
2430 iommu_domain_free(domain->domain);
2431 list_del(&domain->next);
2433 vfio_iommu_aper_expand(iommu, &iova_copy);
2434 vfio_update_pgsize_bitmap(iommu);
2439 if (!vfio_iommu_resv_refresh(iommu, &iova_copy))
2440 vfio_iommu_iova_insert_copy(iommu, &iova_copy);
2442 vfio_iommu_iova_free(&iova_copy);
2446 * Removal of a group without dirty tracking may allow the iommu scope
2449 if (update_dirty_scope)
2450 update_pinned_page_dirty_scope(iommu);
2451 mutex_unlock(&iommu->lock);
2454 static void *vfio_iommu_type1_open(unsigned long arg)
2456 struct vfio_iommu *iommu;
2458 iommu = kzalloc(sizeof(*iommu), GFP_KERNEL);
2460 return ERR_PTR(-ENOMEM);
2463 case VFIO_TYPE1_IOMMU:
2465 case VFIO_TYPE1_NESTING_IOMMU:
2466 iommu->nesting = true;
2468 case VFIO_TYPE1v2_IOMMU:
2473 return ERR_PTR(-EINVAL);
2476 INIT_LIST_HEAD(&iommu->domain_list);
2477 INIT_LIST_HEAD(&iommu->iova_list);
2478 iommu->dma_list = RB_ROOT;
2479 iommu->dma_avail = dma_entry_limit;
2480 mutex_init(&iommu->lock);
2481 BLOCKING_INIT_NOTIFIER_HEAD(&iommu->notifier);
2486 static void vfio_release_domain(struct vfio_domain *domain, bool external)
2488 struct vfio_group *group, *group_tmp;
2490 list_for_each_entry_safe(group, group_tmp,
2491 &domain->group_list, next) {
2493 vfio_iommu_detach_group(domain, group);
2494 list_del(&group->next);
2499 iommu_domain_free(domain->domain);
2502 static void vfio_iommu_type1_release(void *iommu_data)
2504 struct vfio_iommu *iommu = iommu_data;
2505 struct vfio_domain *domain, *domain_tmp;
2507 if (iommu->external_domain) {
2508 vfio_release_domain(iommu->external_domain, true);
2509 vfio_sanity_check_pfn_list(iommu);
2510 kfree(iommu->external_domain);
2513 vfio_iommu_unmap_unpin_all(iommu);
2515 list_for_each_entry_safe(domain, domain_tmp,
2516 &iommu->domain_list, next) {
2517 vfio_release_domain(domain, false);
2518 list_del(&domain->next);
2522 vfio_iommu_iova_free(&iommu->iova_list);
2527 static int vfio_domains_have_iommu_cache(struct vfio_iommu *iommu)
2529 struct vfio_domain *domain;
2532 mutex_lock(&iommu->lock);
2533 list_for_each_entry(domain, &iommu->domain_list, next) {
2534 if (!(domain->prot & IOMMU_CACHE)) {
2539 mutex_unlock(&iommu->lock);
2544 static int vfio_iommu_type1_check_extension(struct vfio_iommu *iommu,
2548 case VFIO_TYPE1_IOMMU:
2549 case VFIO_TYPE1v2_IOMMU:
2550 case VFIO_TYPE1_NESTING_IOMMU:
2551 case VFIO_UNMAP_ALL:
2553 case VFIO_DMA_CC_IOMMU:
2556 return vfio_domains_have_iommu_cache(iommu);
2562 static int vfio_iommu_iova_add_cap(struct vfio_info_cap *caps,
2563 struct vfio_iommu_type1_info_cap_iova_range *cap_iovas,
2566 struct vfio_info_cap_header *header;
2567 struct vfio_iommu_type1_info_cap_iova_range *iova_cap;
2569 header = vfio_info_cap_add(caps, size,
2570 VFIO_IOMMU_TYPE1_INFO_CAP_IOVA_RANGE, 1);
2572 return PTR_ERR(header);
2574 iova_cap = container_of(header,
2575 struct vfio_iommu_type1_info_cap_iova_range,
2577 iova_cap->nr_iovas = cap_iovas->nr_iovas;
2578 memcpy(iova_cap->iova_ranges, cap_iovas->iova_ranges,
2579 cap_iovas->nr_iovas * sizeof(*cap_iovas->iova_ranges));
2583 static int vfio_iommu_iova_build_caps(struct vfio_iommu *iommu,
2584 struct vfio_info_cap *caps)
2586 struct vfio_iommu_type1_info_cap_iova_range *cap_iovas;
2587 struct vfio_iova *iova;
2589 int iovas = 0, i = 0, ret;
2591 list_for_each_entry(iova, &iommu->iova_list, list)
2596 * Return 0 as a container with a single mdev device
2597 * will have an empty list
2602 size = sizeof(*cap_iovas) + (iovas * sizeof(*cap_iovas->iova_ranges));
2604 cap_iovas = kzalloc(size, GFP_KERNEL);
2608 cap_iovas->nr_iovas = iovas;
2610 list_for_each_entry(iova, &iommu->iova_list, list) {
2611 cap_iovas->iova_ranges[i].start = iova->start;
2612 cap_iovas->iova_ranges[i].end = iova->end;
2616 ret = vfio_iommu_iova_add_cap(caps, cap_iovas, size);
2622 static int vfio_iommu_migration_build_caps(struct vfio_iommu *iommu,
2623 struct vfio_info_cap *caps)
2625 struct vfio_iommu_type1_info_cap_migration cap_mig;
2627 cap_mig.header.id = VFIO_IOMMU_TYPE1_INFO_CAP_MIGRATION;
2628 cap_mig.header.version = 1;
2631 /* support minimum pgsize */
2632 cap_mig.pgsize_bitmap = (size_t)1 << __ffs(iommu->pgsize_bitmap);
2633 cap_mig.max_dirty_bitmap_size = DIRTY_BITMAP_SIZE_MAX;
2635 return vfio_info_add_capability(caps, &cap_mig.header, sizeof(cap_mig));
2638 static int vfio_iommu_dma_avail_build_caps(struct vfio_iommu *iommu,
2639 struct vfio_info_cap *caps)
2641 struct vfio_iommu_type1_info_dma_avail cap_dma_avail;
2643 cap_dma_avail.header.id = VFIO_IOMMU_TYPE1_INFO_DMA_AVAIL;
2644 cap_dma_avail.header.version = 1;
2646 cap_dma_avail.avail = iommu->dma_avail;
2648 return vfio_info_add_capability(caps, &cap_dma_avail.header,
2649 sizeof(cap_dma_avail));
2652 static int vfio_iommu_type1_get_info(struct vfio_iommu *iommu,
2655 struct vfio_iommu_type1_info info;
2656 unsigned long minsz;
2657 struct vfio_info_cap caps = { .buf = NULL, .size = 0 };
2658 unsigned long capsz;
2661 minsz = offsetofend(struct vfio_iommu_type1_info, iova_pgsizes);
2663 /* For backward compatibility, cannot require this */
2664 capsz = offsetofend(struct vfio_iommu_type1_info, cap_offset);
2666 if (copy_from_user(&info, (void __user *)arg, minsz))
2669 if (info.argsz < minsz)
2672 if (info.argsz >= capsz) {
2674 info.cap_offset = 0; /* output, no-recopy necessary */
2677 mutex_lock(&iommu->lock);
2678 info.flags = VFIO_IOMMU_INFO_PGSIZES;
2680 info.iova_pgsizes = iommu->pgsize_bitmap;
2682 ret = vfio_iommu_migration_build_caps(iommu, &caps);
2685 ret = vfio_iommu_dma_avail_build_caps(iommu, &caps);
2688 ret = vfio_iommu_iova_build_caps(iommu, &caps);
2690 mutex_unlock(&iommu->lock);
2696 info.flags |= VFIO_IOMMU_INFO_CAPS;
2698 if (info.argsz < sizeof(info) + caps.size) {
2699 info.argsz = sizeof(info) + caps.size;
2701 vfio_info_cap_shift(&caps, sizeof(info));
2702 if (copy_to_user((void __user *)arg +
2703 sizeof(info), caps.buf,
2708 info.cap_offset = sizeof(info);
2714 return copy_to_user((void __user *)arg, &info, minsz) ?
2718 static int vfio_iommu_type1_map_dma(struct vfio_iommu *iommu,
2721 struct vfio_iommu_type1_dma_map map;
2722 unsigned long minsz;
2723 uint32_t mask = VFIO_DMA_MAP_FLAG_READ | VFIO_DMA_MAP_FLAG_WRITE;
2725 minsz = offsetofend(struct vfio_iommu_type1_dma_map, size);
2727 if (copy_from_user(&map, (void __user *)arg, minsz))
2730 if (map.argsz < minsz || map.flags & ~mask)
2733 return vfio_dma_do_map(iommu, &map);
2736 static int vfio_iommu_type1_unmap_dma(struct vfio_iommu *iommu,
2739 struct vfio_iommu_type1_dma_unmap unmap;
2740 struct vfio_bitmap bitmap = { 0 };
2741 uint32_t mask = VFIO_DMA_UNMAP_FLAG_GET_DIRTY_BITMAP |
2742 VFIO_DMA_UNMAP_FLAG_ALL;
2743 unsigned long minsz;
2746 minsz = offsetofend(struct vfio_iommu_type1_dma_unmap, size);
2748 if (copy_from_user(&unmap, (void __user *)arg, minsz))
2751 if (unmap.argsz < minsz || unmap.flags & ~mask)
2754 if ((unmap.flags & VFIO_DMA_UNMAP_FLAG_GET_DIRTY_BITMAP) &&
2755 (unmap.flags & VFIO_DMA_UNMAP_FLAG_ALL))
2758 if (unmap.flags & VFIO_DMA_UNMAP_FLAG_GET_DIRTY_BITMAP) {
2759 unsigned long pgshift;
2761 if (unmap.argsz < (minsz + sizeof(bitmap)))
2764 if (copy_from_user(&bitmap,
2765 (void __user *)(arg + minsz),
2769 if (!access_ok((void __user *)bitmap.data, bitmap.size))
2772 pgshift = __ffs(bitmap.pgsize);
2773 ret = verify_bitmap_size(unmap.size >> pgshift,
2779 ret = vfio_dma_do_unmap(iommu, &unmap, &bitmap);
2783 return copy_to_user((void __user *)arg, &unmap, minsz) ?
2787 static int vfio_iommu_type1_dirty_pages(struct vfio_iommu *iommu,
2790 struct vfio_iommu_type1_dirty_bitmap dirty;
2791 uint32_t mask = VFIO_IOMMU_DIRTY_PAGES_FLAG_START |
2792 VFIO_IOMMU_DIRTY_PAGES_FLAG_STOP |
2793 VFIO_IOMMU_DIRTY_PAGES_FLAG_GET_BITMAP;
2794 unsigned long minsz;
2800 minsz = offsetofend(struct vfio_iommu_type1_dirty_bitmap, flags);
2802 if (copy_from_user(&dirty, (void __user *)arg, minsz))
2805 if (dirty.argsz < minsz || dirty.flags & ~mask)
2808 /* only one flag should be set at a time */
2809 if (__ffs(dirty.flags) != __fls(dirty.flags))
2812 if (dirty.flags & VFIO_IOMMU_DIRTY_PAGES_FLAG_START) {
2815 mutex_lock(&iommu->lock);
2816 pgsize = 1 << __ffs(iommu->pgsize_bitmap);
2817 if (!iommu->dirty_page_tracking) {
2818 ret = vfio_dma_bitmap_alloc_all(iommu, pgsize);
2820 iommu->dirty_page_tracking = true;
2822 mutex_unlock(&iommu->lock);
2824 } else if (dirty.flags & VFIO_IOMMU_DIRTY_PAGES_FLAG_STOP) {
2825 mutex_lock(&iommu->lock);
2826 if (iommu->dirty_page_tracking) {
2827 iommu->dirty_page_tracking = false;
2828 vfio_dma_bitmap_free_all(iommu);
2830 mutex_unlock(&iommu->lock);
2832 } else if (dirty.flags & VFIO_IOMMU_DIRTY_PAGES_FLAG_GET_BITMAP) {
2833 struct vfio_iommu_type1_dirty_bitmap_get range;
2834 unsigned long pgshift;
2835 size_t data_size = dirty.argsz - minsz;
2836 size_t iommu_pgsize;
2838 if (!data_size || data_size < sizeof(range))
2841 if (copy_from_user(&range, (void __user *)(arg + minsz),
2845 if (range.iova + range.size < range.iova)
2847 if (!access_ok((void __user *)range.bitmap.data,
2851 pgshift = __ffs(range.bitmap.pgsize);
2852 ret = verify_bitmap_size(range.size >> pgshift,
2857 mutex_lock(&iommu->lock);
2859 iommu_pgsize = (size_t)1 << __ffs(iommu->pgsize_bitmap);
2861 /* allow only smallest supported pgsize */
2862 if (range.bitmap.pgsize != iommu_pgsize) {
2866 if (range.iova & (iommu_pgsize - 1)) {
2870 if (!range.size || range.size & (iommu_pgsize - 1)) {
2875 if (iommu->dirty_page_tracking)
2876 ret = vfio_iova_dirty_bitmap(range.bitmap.data,
2879 range.bitmap.pgsize);
2883 mutex_unlock(&iommu->lock);
2891 static long vfio_iommu_type1_ioctl(void *iommu_data,
2892 unsigned int cmd, unsigned long arg)
2894 struct vfio_iommu *iommu = iommu_data;
2897 case VFIO_CHECK_EXTENSION:
2898 return vfio_iommu_type1_check_extension(iommu, arg);
2899 case VFIO_IOMMU_GET_INFO:
2900 return vfio_iommu_type1_get_info(iommu, arg);
2901 case VFIO_IOMMU_MAP_DMA:
2902 return vfio_iommu_type1_map_dma(iommu, arg);
2903 case VFIO_IOMMU_UNMAP_DMA:
2904 return vfio_iommu_type1_unmap_dma(iommu, arg);
2905 case VFIO_IOMMU_DIRTY_PAGES:
2906 return vfio_iommu_type1_dirty_pages(iommu, arg);
2912 static int vfio_iommu_type1_register_notifier(void *iommu_data,
2913 unsigned long *events,
2914 struct notifier_block *nb)
2916 struct vfio_iommu *iommu = iommu_data;
2918 /* clear known events */
2919 *events &= ~VFIO_IOMMU_NOTIFY_DMA_UNMAP;
2921 /* refuse to register if still events remaining */
2925 return blocking_notifier_chain_register(&iommu->notifier, nb);
2928 static int vfio_iommu_type1_unregister_notifier(void *iommu_data,
2929 struct notifier_block *nb)
2931 struct vfio_iommu *iommu = iommu_data;
2933 return blocking_notifier_chain_unregister(&iommu->notifier, nb);
2936 static int vfio_iommu_type1_dma_rw_chunk(struct vfio_iommu *iommu,
2937 dma_addr_t user_iova, void *data,
2938 size_t count, bool write,
2941 struct mm_struct *mm;
2942 unsigned long vaddr;
2943 struct vfio_dma *dma;
2944 bool kthread = current->mm == NULL;
2949 dma = vfio_find_dma(iommu, user_iova, 1);
2953 if ((write && !(dma->prot & IOMMU_WRITE)) ||
2954 !(dma->prot & IOMMU_READ))
2957 mm = get_task_mm(dma->task);
2964 else if (current->mm != mm)
2967 offset = user_iova - dma->iova;
2969 if (count > dma->size - offset)
2970 count = dma->size - offset;
2972 vaddr = dma->vaddr + offset;
2975 *copied = copy_to_user((void __user *)vaddr, data,
2977 if (*copied && iommu->dirty_page_tracking) {
2978 unsigned long pgshift = __ffs(iommu->pgsize_bitmap);
2980 * Bitmap populated with the smallest supported page
2983 bitmap_set(dma->bitmap, offset >> pgshift,
2984 ((offset + *copied - 1) >> pgshift) -
2985 (offset >> pgshift) + 1);
2988 *copied = copy_from_user(data, (void __user *)vaddr,
2991 kthread_unuse_mm(mm);
2994 return *copied ? 0 : -EFAULT;
2997 static int vfio_iommu_type1_dma_rw(void *iommu_data, dma_addr_t user_iova,
2998 void *data, size_t count, bool write)
3000 struct vfio_iommu *iommu = iommu_data;
3004 mutex_lock(&iommu->lock);
3006 ret = vfio_iommu_type1_dma_rw_chunk(iommu, user_iova, data,
3007 count, write, &done);
3016 mutex_unlock(&iommu->lock);
3020 static struct iommu_domain *
3021 vfio_iommu_type1_group_iommu_domain(void *iommu_data,
3022 struct iommu_group *iommu_group)
3024 struct iommu_domain *domain = ERR_PTR(-ENODEV);
3025 struct vfio_iommu *iommu = iommu_data;
3026 struct vfio_domain *d;
3028 if (!iommu || !iommu_group)
3029 return ERR_PTR(-EINVAL);
3031 mutex_lock(&iommu->lock);
3032 list_for_each_entry(d, &iommu->domain_list, next) {
3033 if (find_iommu_group(d, iommu_group)) {
3038 mutex_unlock(&iommu->lock);
3043 static const struct vfio_iommu_driver_ops vfio_iommu_driver_ops_type1 = {
3044 .name = "vfio-iommu-type1",
3045 .owner = THIS_MODULE,
3046 .open = vfio_iommu_type1_open,
3047 .release = vfio_iommu_type1_release,
3048 .ioctl = vfio_iommu_type1_ioctl,
3049 .attach_group = vfio_iommu_type1_attach_group,
3050 .detach_group = vfio_iommu_type1_detach_group,
3051 .pin_pages = vfio_iommu_type1_pin_pages,
3052 .unpin_pages = vfio_iommu_type1_unpin_pages,
3053 .register_notifier = vfio_iommu_type1_register_notifier,
3054 .unregister_notifier = vfio_iommu_type1_unregister_notifier,
3055 .dma_rw = vfio_iommu_type1_dma_rw,
3056 .group_iommu_domain = vfio_iommu_type1_group_iommu_domain,
3059 static int __init vfio_iommu_type1_init(void)
3061 return vfio_register_iommu_driver(&vfio_iommu_driver_ops_type1);
3064 static void __exit vfio_iommu_type1_cleanup(void)
3066 vfio_unregister_iommu_driver(&vfio_iommu_driver_ops_type1);
3069 module_init(vfio_iommu_type1_init);
3070 module_exit(vfio_iommu_type1_cleanup);
3072 MODULE_VERSION(DRIVER_VERSION);
3073 MODULE_LICENSE("GPL v2");
3074 MODULE_AUTHOR(DRIVER_AUTHOR);
3075 MODULE_DESCRIPTION(DRIVER_DESC);