1 // SPDX-License-Identifier: GPL-2.0-only
4 * Copyright 2010 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
5 * Copyright 2011 David Gibson, IBM Corporation <dwg@au1.ibm.com>
6 * Copyright 2016 Alexey Kardashevskiy, IBM Corporation <aik@au1.ibm.com>
9 #include <linux/types.h>
10 #include <linux/string.h>
11 #include <linux/kvm.h>
12 #include <linux/kvm_host.h>
13 #include <linux/highmem.h>
14 #include <linux/gfp.h>
15 #include <linux/slab.h>
16 #include <linux/sched/signal.h>
17 #include <linux/hugetlb.h>
18 #include <linux/list.h>
19 #include <linux/anon_inodes.h>
20 #include <linux/iommu.h>
21 #include <linux/file.h>
24 #include <asm/kvm_ppc.h>
25 #include <asm/kvm_book3s.h>
26 #include <asm/book3s/64/mmu-hash.h>
27 #include <asm/hvcall.h>
28 #include <asm/synch.h>
29 #include <asm/ppc-opcode.h>
30 #include <asm/kvm_host.h>
32 #include <asm/iommu.h>
34 #include <asm/mmu_context.h>
36 static unsigned long kvmppc_tce_pages(unsigned long iommu_pages)
38 return ALIGN(iommu_pages * sizeof(u64), PAGE_SIZE) / PAGE_SIZE;
41 static unsigned long kvmppc_stt_pages(unsigned long tce_pages)
43 unsigned long stt_bytes = sizeof(struct kvmppc_spapr_tce_table) +
44 (tce_pages * sizeof(struct page *));
46 return tce_pages + ALIGN(stt_bytes, PAGE_SIZE) / PAGE_SIZE;
49 static void kvm_spapr_tce_iommu_table_free(struct rcu_head *head)
51 struct kvmppc_spapr_tce_iommu_table *stit = container_of(head,
52 struct kvmppc_spapr_tce_iommu_table, rcu);
54 iommu_tce_table_put(stit->tbl);
59 static void kvm_spapr_tce_liobn_put(struct kref *kref)
61 struct kvmppc_spapr_tce_iommu_table *stit = container_of(kref,
62 struct kvmppc_spapr_tce_iommu_table, kref);
64 list_del_rcu(&stit->next);
66 call_rcu(&stit->rcu, kvm_spapr_tce_iommu_table_free);
69 extern void kvm_spapr_tce_release_iommu_group(struct kvm *kvm,
70 struct iommu_group *grp)
73 struct kvmppc_spapr_tce_table *stt;
74 struct kvmppc_spapr_tce_iommu_table *stit, *tmp;
75 struct iommu_table_group *table_group = NULL;
77 list_for_each_entry_rcu(stt, &kvm->arch.spapr_tce_tables, list) {
79 table_group = iommu_group_get_iommudata(grp);
80 if (WARN_ON(!table_group))
83 list_for_each_entry_safe(stit, tmp, &stt->iommu_tables, next) {
84 for (i = 0; i < IOMMU_TABLE_GROUP_MAX_TABLES; ++i) {
85 if (table_group->tables[i] != stit->tbl)
88 kref_put(&stit->kref, kvm_spapr_tce_liobn_put);
94 extern long kvm_spapr_tce_attach_iommu_group(struct kvm *kvm, int tablefd,
95 struct iommu_group *grp)
97 struct kvmppc_spapr_tce_table *stt = NULL;
99 struct iommu_table *tbl = NULL;
100 struct iommu_table_group *table_group;
102 struct kvmppc_spapr_tce_iommu_table *stit;
109 list_for_each_entry_rcu(stt, &kvm->arch.spapr_tce_tables, list) {
110 if (stt == f.file->private_data) {
121 table_group = iommu_group_get_iommudata(grp);
122 if (WARN_ON(!table_group))
125 for (i = 0; i < IOMMU_TABLE_GROUP_MAX_TABLES; ++i) {
126 struct iommu_table *tbltmp = table_group->tables[i];
130 /* Make sure hardware table parameters are compatible */
131 if ((tbltmp->it_page_shift <= stt->page_shift) &&
132 (tbltmp->it_offset << tbltmp->it_page_shift ==
133 stt->offset << stt->page_shift) &&
134 (tbltmp->it_size << tbltmp->it_page_shift >=
135 stt->size << stt->page_shift)) {
137 * Reference the table to avoid races with
138 * add/remove DMA windows.
140 tbl = iommu_tce_table_get(tbltmp);
147 list_for_each_entry_rcu(stit, &stt->iommu_tables, next) {
148 if (tbl != stit->tbl)
151 if (!kref_get_unless_zero(&stit->kref)) {
152 /* stit is being destroyed */
153 iommu_tce_table_put(tbl);
157 * The table is already known to this KVM, we just increased
158 * its KVM reference counter and can return.
163 stit = kzalloc(sizeof(*stit), GFP_KERNEL);
165 iommu_tce_table_put(tbl);
170 kref_init(&stit->kref);
172 list_add_rcu(&stit->next, &stt->iommu_tables);
177 static void release_spapr_tce_table(struct rcu_head *head)
179 struct kvmppc_spapr_tce_table *stt = container_of(head,
180 struct kvmppc_spapr_tce_table, rcu);
181 unsigned long i, npages = kvmppc_tce_pages(stt->size);
183 for (i = 0; i < npages; i++)
185 __free_page(stt->pages[i]);
190 static struct page *kvm_spapr_get_tce_page(struct kvmppc_spapr_tce_table *stt,
191 unsigned long sttpage)
193 struct page *page = stt->pages[sttpage];
198 mutex_lock(&stt->alloc_lock);
199 page = stt->pages[sttpage];
201 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
204 stt->pages[sttpage] = page;
206 mutex_unlock(&stt->alloc_lock);
211 static vm_fault_t kvm_spapr_tce_fault(struct vm_fault *vmf)
213 struct kvmppc_spapr_tce_table *stt = vmf->vma->vm_file->private_data;
216 if (vmf->pgoff >= kvmppc_tce_pages(stt->size))
217 return VM_FAULT_SIGBUS;
219 page = kvm_spapr_get_tce_page(stt, vmf->pgoff);
228 static const struct vm_operations_struct kvm_spapr_tce_vm_ops = {
229 .fault = kvm_spapr_tce_fault,
232 static int kvm_spapr_tce_mmap(struct file *file, struct vm_area_struct *vma)
234 vma->vm_ops = &kvm_spapr_tce_vm_ops;
238 static int kvm_spapr_tce_release(struct inode *inode, struct file *filp)
240 struct kvmppc_spapr_tce_table *stt = filp->private_data;
241 struct kvmppc_spapr_tce_iommu_table *stit, *tmp;
242 struct kvm *kvm = stt->kvm;
244 mutex_lock(&kvm->lock);
245 list_del_rcu(&stt->list);
246 mutex_unlock(&kvm->lock);
248 list_for_each_entry_safe(stit, tmp, &stt->iommu_tables, next) {
249 WARN_ON(!kref_read(&stit->kref));
251 if (kref_put(&stit->kref, kvm_spapr_tce_liobn_put))
256 account_locked_vm(kvm->mm,
257 kvmppc_stt_pages(kvmppc_tce_pages(stt->size)), false);
259 kvm_put_kvm(stt->kvm);
261 call_rcu(&stt->rcu, release_spapr_tce_table);
266 static const struct file_operations kvm_spapr_tce_fops = {
267 .mmap = kvm_spapr_tce_mmap,
268 .release = kvm_spapr_tce_release,
271 long kvm_vm_ioctl_create_spapr_tce(struct kvm *kvm,
272 struct kvm_create_spapr_tce_64 *args)
274 struct kvmppc_spapr_tce_table *stt = NULL;
275 struct kvmppc_spapr_tce_table *siter;
276 struct mm_struct *mm = kvm->mm;
277 unsigned long npages, size = args->size;
280 if (!args->size || args->page_shift < 12 || args->page_shift > 34 ||
281 (args->offset + args->size > (ULLONG_MAX >> args->page_shift)))
284 npages = kvmppc_tce_pages(size);
285 ret = account_locked_vm(mm, kvmppc_stt_pages(npages), true);
290 stt = kzalloc(sizeof(*stt) + npages * sizeof(struct page *),
295 stt->liobn = args->liobn;
296 stt->page_shift = args->page_shift;
297 stt->offset = args->offset;
300 mutex_init(&stt->alloc_lock);
301 INIT_LIST_HEAD_RCU(&stt->iommu_tables);
303 mutex_lock(&kvm->lock);
305 /* Check this LIOBN hasn't been previously allocated */
307 list_for_each_entry(siter, &kvm->arch.spapr_tce_tables, list) {
308 if (siter->liobn == args->liobn) {
316 ret = anon_inode_getfd("kvm-spapr-tce", &kvm_spapr_tce_fops,
317 stt, O_RDWR | O_CLOEXEC);
320 list_add_rcu(&stt->list, &kvm->arch.spapr_tce_tables);
322 kvm_put_kvm_no_destroy(kvm);
324 mutex_unlock(&kvm->lock);
331 account_locked_vm(mm, kvmppc_stt_pages(npages), false);
335 static long kvmppc_tce_to_ua(struct kvm *kvm, unsigned long tce,
338 unsigned long gfn = tce >> PAGE_SHIFT;
339 struct kvm_memory_slot *memslot;
341 memslot = search_memslots(kvm_memslots(kvm), gfn);
345 *ua = __gfn_to_hva_memslot(memslot, gfn) |
346 (tce & ~(PAGE_MASK | TCE_PCI_READ | TCE_PCI_WRITE));
351 static long kvmppc_tce_validate(struct kvmppc_spapr_tce_table *stt,
354 unsigned long gpa = tce & ~(TCE_PCI_READ | TCE_PCI_WRITE);
355 enum dma_data_direction dir = iommu_tce_direction(tce);
356 struct kvmppc_spapr_tce_iommu_table *stit;
357 unsigned long ua = 0;
359 /* Allow userspace to poison TCE table */
363 if (iommu_tce_check_gpa(stt->page_shift, gpa))
366 if (kvmppc_tce_to_ua(stt->kvm, tce, &ua))
369 list_for_each_entry_rcu(stit, &stt->iommu_tables, next) {
370 unsigned long hpa = 0;
371 struct mm_iommu_table_group_mem_t *mem;
372 long shift = stit->tbl->it_page_shift;
374 mem = mm_iommu_lookup(stt->kvm->mm, ua, 1ULL << shift);
378 if (mm_iommu_ua_to_hpa(mem, ua, shift, &hpa))
386 * Handles TCE requests for emulated devices.
387 * Puts guest TCE values to the table and expects user space to convert them.
388 * Cannot fail so kvmppc_tce_validate must be called before it.
390 static void kvmppc_tce_put(struct kvmppc_spapr_tce_table *stt,
391 unsigned long idx, unsigned long tce)
395 unsigned long sttpage;
398 sttpage = idx / TCES_PER_PAGE;
399 page = stt->pages[sttpage];
402 /* We allow any TCE, not just with read|write permissions */
406 page = kvm_spapr_get_tce_page(stt, sttpage);
410 tbl = page_to_virt(page);
412 tbl[idx % TCES_PER_PAGE] = tce;
415 static void kvmppc_clear_tce(struct mm_struct *mm, struct iommu_table *tbl,
418 unsigned long hpa = 0;
419 enum dma_data_direction dir = DMA_NONE;
421 iommu_tce_xchg_no_kill(mm, tbl, entry, &hpa, &dir);
424 static long kvmppc_tce_iommu_mapped_dec(struct kvm *kvm,
425 struct iommu_table *tbl, unsigned long entry)
427 struct mm_iommu_table_group_mem_t *mem = NULL;
428 const unsigned long pgsize = 1ULL << tbl->it_page_shift;
429 __be64 *pua = IOMMU_TABLE_USERSPACE_ENTRY_RO(tbl, entry);
434 mem = mm_iommu_lookup(kvm->mm, be64_to_cpu(*pua), pgsize);
438 mm_iommu_mapped_dec(mem);
440 *pua = cpu_to_be64(0);
445 static long kvmppc_tce_iommu_do_unmap(struct kvm *kvm,
446 struct iommu_table *tbl, unsigned long entry)
448 enum dma_data_direction dir = DMA_NONE;
449 unsigned long hpa = 0;
452 if (WARN_ON_ONCE(iommu_tce_xchg_no_kill(kvm->mm, tbl, entry, &hpa,
459 ret = kvmppc_tce_iommu_mapped_dec(kvm, tbl, entry);
460 if (ret != H_SUCCESS)
461 iommu_tce_xchg_no_kill(kvm->mm, tbl, entry, &hpa, &dir);
466 static long kvmppc_tce_iommu_unmap(struct kvm *kvm,
467 struct kvmppc_spapr_tce_table *stt, struct iommu_table *tbl,
470 unsigned long i, ret = H_SUCCESS;
471 unsigned long subpages = 1ULL << (stt->page_shift - tbl->it_page_shift);
472 unsigned long io_entry = entry * subpages;
474 for (i = 0; i < subpages; ++i) {
475 ret = kvmppc_tce_iommu_do_unmap(kvm, tbl, io_entry + i);
476 if (ret != H_SUCCESS)
483 long kvmppc_tce_iommu_do_map(struct kvm *kvm, struct iommu_table *tbl,
484 unsigned long entry, unsigned long ua,
485 enum dma_data_direction dir)
489 __be64 *pua = IOMMU_TABLE_USERSPACE_ENTRY(tbl, entry);
490 struct mm_iommu_table_group_mem_t *mem;
493 /* it_userspace allocation might be delayed */
496 mem = mm_iommu_lookup(kvm->mm, ua, 1ULL << tbl->it_page_shift);
498 /* This only handles v2 IOMMU type, v1 is handled via ioctl() */
501 if (WARN_ON_ONCE(mm_iommu_ua_to_hpa(mem, ua, tbl->it_page_shift, &hpa)))
504 if (mm_iommu_mapped_inc(mem))
507 ret = iommu_tce_xchg_no_kill(kvm->mm, tbl, entry, &hpa, &dir);
508 if (WARN_ON_ONCE(ret)) {
509 mm_iommu_mapped_dec(mem);
514 kvmppc_tce_iommu_mapped_dec(kvm, tbl, entry);
516 *pua = cpu_to_be64(ua);
521 static long kvmppc_tce_iommu_map(struct kvm *kvm,
522 struct kvmppc_spapr_tce_table *stt, struct iommu_table *tbl,
523 unsigned long entry, unsigned long ua,
524 enum dma_data_direction dir)
526 unsigned long i, pgoff, ret = H_SUCCESS;
527 unsigned long subpages = 1ULL << (stt->page_shift - tbl->it_page_shift);
528 unsigned long io_entry = entry * subpages;
530 for (i = 0, pgoff = 0; i < subpages;
531 ++i, pgoff += IOMMU_PAGE_SIZE(tbl)) {
533 ret = kvmppc_tce_iommu_do_map(kvm, tbl,
534 io_entry + i, ua + pgoff, dir);
535 if (ret != H_SUCCESS)
542 long kvmppc_h_put_tce(struct kvm_vcpu *vcpu, unsigned long liobn,
543 unsigned long ioba, unsigned long tce)
545 struct kvmppc_spapr_tce_table *stt;
547 struct kvmppc_spapr_tce_iommu_table *stit;
548 unsigned long entry, ua = 0;
549 enum dma_data_direction dir;
551 /* udbg_printf("H_PUT_TCE(): liobn=0x%lx ioba=0x%lx, tce=0x%lx\n", */
552 /* liobn, ioba, tce); */
554 stt = kvmppc_find_table(vcpu->kvm, liobn);
558 ret = kvmppc_ioba_validate(stt, ioba, 1);
559 if (ret != H_SUCCESS)
562 idx = srcu_read_lock(&vcpu->kvm->srcu);
564 ret = kvmppc_tce_validate(stt, tce);
565 if (ret != H_SUCCESS)
568 dir = iommu_tce_direction(tce);
570 if ((dir != DMA_NONE) && kvmppc_tce_to_ua(vcpu->kvm, tce, &ua)) {
575 entry = ioba >> stt->page_shift;
577 list_for_each_entry_lockless(stit, &stt->iommu_tables, next) {
579 ret = kvmppc_tce_iommu_unmap(vcpu->kvm, stt,
582 ret = kvmppc_tce_iommu_map(vcpu->kvm, stt, stit->tbl,
585 iommu_tce_kill(stit->tbl, entry, 1);
587 if (ret != H_SUCCESS) {
588 kvmppc_clear_tce(vcpu->kvm->mm, stit->tbl, entry);
593 kvmppc_tce_put(stt, entry, tce);
596 srcu_read_unlock(&vcpu->kvm->srcu, idx);
600 EXPORT_SYMBOL_GPL(kvmppc_h_put_tce);
602 long kvmppc_h_put_tce_indirect(struct kvm_vcpu *vcpu,
603 unsigned long liobn, unsigned long ioba,
604 unsigned long tce_list, unsigned long npages)
606 struct kvmppc_spapr_tce_table *stt;
607 long i, ret = H_SUCCESS, idx;
608 unsigned long entry, ua = 0;
611 struct kvmppc_spapr_tce_iommu_table *stit;
613 stt = kvmppc_find_table(vcpu->kvm, liobn);
617 entry = ioba >> stt->page_shift;
619 * SPAPR spec says that the maximum size of the list is 512 TCEs
620 * so the whole table fits in 4K page
625 if (tce_list & (SZ_4K - 1))
628 ret = kvmppc_ioba_validate(stt, ioba, npages);
629 if (ret != H_SUCCESS)
632 idx = srcu_read_lock(&vcpu->kvm->srcu);
633 if (kvmppc_tce_to_ua(vcpu->kvm, tce_list, &ua)) {
637 tces = (u64 __user *) ua;
639 for (i = 0; i < npages; ++i) {
640 if (get_user(tce, tces + i)) {
644 tce = be64_to_cpu(tce);
646 ret = kvmppc_tce_validate(stt, tce);
647 if (ret != H_SUCCESS)
651 for (i = 0; i < npages; ++i) {
653 * This looks unsafe, because we validate, then regrab
654 * the TCE from userspace which could have been changed by
657 * But it actually is safe, because the relevant checks will be
658 * re-executed in the following code. If userspace tries to
659 * change this dodgily it will result in a messier failure mode
660 * but won't threaten the host.
662 if (get_user(tce, tces + i)) {
664 goto invalidate_exit;
666 tce = be64_to_cpu(tce);
668 if (kvmppc_tce_to_ua(vcpu->kvm, tce, &ua)) {
670 goto invalidate_exit;
673 list_for_each_entry_lockless(stit, &stt->iommu_tables, next) {
674 ret = kvmppc_tce_iommu_map(vcpu->kvm, stt,
675 stit->tbl, entry + i, ua,
676 iommu_tce_direction(tce));
678 if (ret != H_SUCCESS) {
679 kvmppc_clear_tce(vcpu->kvm->mm, stit->tbl,
681 goto invalidate_exit;
685 kvmppc_tce_put(stt, entry + i, tce);
689 list_for_each_entry_lockless(stit, &stt->iommu_tables, next)
690 iommu_tce_kill(stit->tbl, entry, npages);
693 srcu_read_unlock(&vcpu->kvm->srcu, idx);
697 EXPORT_SYMBOL_GPL(kvmppc_h_put_tce_indirect);
699 long kvmppc_h_stuff_tce(struct kvm_vcpu *vcpu,
700 unsigned long liobn, unsigned long ioba,
701 unsigned long tce_value, unsigned long npages)
703 struct kvmppc_spapr_tce_table *stt;
705 struct kvmppc_spapr_tce_iommu_table *stit;
707 stt = kvmppc_find_table(vcpu->kvm, liobn);
711 ret = kvmppc_ioba_validate(stt, ioba, npages);
712 if (ret != H_SUCCESS)
715 /* Check permission bits only to allow userspace poison TCE for debug */
716 if (tce_value & (TCE_PCI_WRITE | TCE_PCI_READ))
719 list_for_each_entry_lockless(stit, &stt->iommu_tables, next) {
720 unsigned long entry = ioba >> stt->page_shift;
722 for (i = 0; i < npages; ++i) {
723 ret = kvmppc_tce_iommu_unmap(vcpu->kvm, stt,
724 stit->tbl, entry + i);
726 if (ret == H_SUCCESS)
729 if (ret == H_TOO_HARD)
730 goto invalidate_exit;
733 kvmppc_clear_tce(vcpu->kvm->mm, stit->tbl, entry);
737 for (i = 0; i < npages; ++i, ioba += (1ULL << stt->page_shift))
738 kvmppc_tce_put(stt, ioba >> stt->page_shift, tce_value);
741 list_for_each_entry_lockless(stit, &stt->iommu_tables, next)
742 iommu_tce_kill(stit->tbl, ioba >> stt->page_shift, npages);
746 EXPORT_SYMBOL_GPL(kvmppc_h_stuff_tce);