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>
31 #include <asm/iommu.h>
33 #include <asm/mmu_context.h>
35 static unsigned long kvmppc_tce_pages(unsigned long iommu_pages)
37 return ALIGN(iommu_pages * sizeof(u64), PAGE_SIZE) / PAGE_SIZE;
40 static unsigned long kvmppc_stt_pages(unsigned long tce_pages)
42 unsigned long stt_bytes = sizeof(struct kvmppc_spapr_tce_table) +
43 (tce_pages * sizeof(struct page *));
45 return tce_pages + ALIGN(stt_bytes, PAGE_SIZE) / PAGE_SIZE;
48 static void kvm_spapr_tce_iommu_table_free(struct rcu_head *head)
50 struct kvmppc_spapr_tce_iommu_table *stit = container_of(head,
51 struct kvmppc_spapr_tce_iommu_table, rcu);
53 iommu_tce_table_put(stit->tbl);
58 static void kvm_spapr_tce_liobn_put(struct kref *kref)
60 struct kvmppc_spapr_tce_iommu_table *stit = container_of(kref,
61 struct kvmppc_spapr_tce_iommu_table, kref);
63 list_del_rcu(&stit->next);
65 call_rcu(&stit->rcu, kvm_spapr_tce_iommu_table_free);
68 extern void kvm_spapr_tce_release_iommu_group(struct kvm *kvm,
69 struct iommu_group *grp)
72 struct kvmppc_spapr_tce_table *stt;
73 struct kvmppc_spapr_tce_iommu_table *stit, *tmp;
74 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);
96 extern long kvm_spapr_tce_attach_iommu_group(struct kvm *kvm, int tablefd,
97 struct iommu_group *grp)
99 struct kvmppc_spapr_tce_table *stt = NULL;
101 struct iommu_table *tbl = NULL;
102 struct iommu_table_group *table_group;
104 struct kvmppc_spapr_tce_iommu_table *stit;
112 list_for_each_entry_rcu(stt, &kvm->arch.spapr_tce_tables, list) {
113 if (stt == f.file->private_data) {
125 table_group = iommu_group_get_iommudata(grp);
126 if (WARN_ON(!table_group))
129 for (i = 0; i < IOMMU_TABLE_GROUP_MAX_TABLES; ++i) {
130 struct iommu_table *tbltmp = table_group->tables[i];
134 /* Make sure hardware table parameters are compatible */
135 if ((tbltmp->it_page_shift <= stt->page_shift) &&
136 (tbltmp->it_offset << tbltmp->it_page_shift ==
137 stt->offset << stt->page_shift) &&
138 (tbltmp->it_size << tbltmp->it_page_shift >=
139 stt->size << stt->page_shift)) {
141 * Reference the table to avoid races with
142 * add/remove DMA windows.
144 tbl = iommu_tce_table_get(tbltmp);
152 list_for_each_entry_rcu(stit, &stt->iommu_tables, next) {
153 if (tbl != stit->tbl)
156 if (!kref_get_unless_zero(&stit->kref)) {
157 /* stit is being destroyed */
158 iommu_tce_table_put(tbl);
163 * The table is already known to this KVM, we just increased
164 * its KVM reference counter and can return.
171 stit = kzalloc(sizeof(*stit), GFP_KERNEL);
173 iommu_tce_table_put(tbl);
178 kref_init(&stit->kref);
180 list_add_rcu(&stit->next, &stt->iommu_tables);
185 static void release_spapr_tce_table(struct rcu_head *head)
187 struct kvmppc_spapr_tce_table *stt = container_of(head,
188 struct kvmppc_spapr_tce_table, rcu);
189 unsigned long i, npages = kvmppc_tce_pages(stt->size);
191 for (i = 0; i < npages; i++)
193 __free_page(stt->pages[i]);
198 static struct page *kvm_spapr_get_tce_page(struct kvmppc_spapr_tce_table *stt,
199 unsigned long sttpage)
201 struct page *page = stt->pages[sttpage];
206 mutex_lock(&stt->alloc_lock);
207 page = stt->pages[sttpage];
209 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
212 stt->pages[sttpage] = page;
214 mutex_unlock(&stt->alloc_lock);
219 static vm_fault_t kvm_spapr_tce_fault(struct vm_fault *vmf)
221 struct kvmppc_spapr_tce_table *stt = vmf->vma->vm_file->private_data;
224 if (vmf->pgoff >= kvmppc_tce_pages(stt->size))
225 return VM_FAULT_SIGBUS;
227 page = kvm_spapr_get_tce_page(stt, vmf->pgoff);
236 static const struct vm_operations_struct kvm_spapr_tce_vm_ops = {
237 .fault = kvm_spapr_tce_fault,
240 static int kvm_spapr_tce_mmap(struct file *file, struct vm_area_struct *vma)
242 vma->vm_ops = &kvm_spapr_tce_vm_ops;
246 static int kvm_spapr_tce_release(struct inode *inode, struct file *filp)
248 struct kvmppc_spapr_tce_table *stt = filp->private_data;
249 struct kvmppc_spapr_tce_iommu_table *stit, *tmp;
250 struct kvm *kvm = stt->kvm;
252 mutex_lock(&kvm->lock);
253 list_del_rcu(&stt->list);
254 mutex_unlock(&kvm->lock);
256 list_for_each_entry_safe(stit, tmp, &stt->iommu_tables, next) {
257 WARN_ON(!kref_read(&stit->kref));
259 if (kref_put(&stit->kref, kvm_spapr_tce_liobn_put))
264 account_locked_vm(kvm->mm,
265 kvmppc_stt_pages(kvmppc_tce_pages(stt->size)), false);
267 kvm_put_kvm(stt->kvm);
269 call_rcu(&stt->rcu, release_spapr_tce_table);
274 static const struct file_operations kvm_spapr_tce_fops = {
275 .mmap = kvm_spapr_tce_mmap,
276 .release = kvm_spapr_tce_release,
279 long kvm_vm_ioctl_create_spapr_tce(struct kvm *kvm,
280 struct kvm_create_spapr_tce_64 *args)
282 struct kvmppc_spapr_tce_table *stt = NULL;
283 struct kvmppc_spapr_tce_table *siter;
284 struct mm_struct *mm = kvm->mm;
285 unsigned long npages, size = args->size;
288 if (!args->size || args->page_shift < 12 || args->page_shift > 34 ||
289 (args->offset + args->size > (ULLONG_MAX >> args->page_shift)))
292 npages = kvmppc_tce_pages(size);
293 ret = account_locked_vm(mm, kvmppc_stt_pages(npages), true);
298 stt = kzalloc(sizeof(*stt) + npages * sizeof(struct page *),
303 stt->liobn = args->liobn;
304 stt->page_shift = args->page_shift;
305 stt->offset = args->offset;
308 mutex_init(&stt->alloc_lock);
309 INIT_LIST_HEAD_RCU(&stt->iommu_tables);
311 mutex_lock(&kvm->lock);
313 /* Check this LIOBN hasn't been previously allocated */
315 list_for_each_entry(siter, &kvm->arch.spapr_tce_tables, list) {
316 if (siter->liobn == args->liobn) {
324 ret = anon_inode_getfd("kvm-spapr-tce", &kvm_spapr_tce_fops,
325 stt, O_RDWR | O_CLOEXEC);
328 list_add_rcu(&stt->list, &kvm->arch.spapr_tce_tables);
330 kvm_put_kvm_no_destroy(kvm);
332 mutex_unlock(&kvm->lock);
339 account_locked_vm(mm, kvmppc_stt_pages(npages), false);
343 static long kvmppc_tce_to_ua(struct kvm *kvm, unsigned long tce,
346 unsigned long gfn = tce >> PAGE_SHIFT;
347 struct kvm_memory_slot *memslot;
349 memslot = search_memslots(kvm_memslots(kvm), gfn);
353 *ua = __gfn_to_hva_memslot(memslot, gfn) |
354 (tce & ~(PAGE_MASK | TCE_PCI_READ | TCE_PCI_WRITE));
359 static long kvmppc_tce_validate(struct kvmppc_spapr_tce_table *stt,
362 unsigned long gpa = tce & ~(TCE_PCI_READ | TCE_PCI_WRITE);
363 enum dma_data_direction dir = iommu_tce_direction(tce);
364 struct kvmppc_spapr_tce_iommu_table *stit;
365 unsigned long ua = 0;
367 /* Allow userspace to poison TCE table */
371 if (iommu_tce_check_gpa(stt->page_shift, gpa))
374 if (kvmppc_tce_to_ua(stt->kvm, tce, &ua))
378 list_for_each_entry_rcu(stit, &stt->iommu_tables, next) {
379 unsigned long hpa = 0;
380 struct mm_iommu_table_group_mem_t *mem;
381 long shift = stit->tbl->it_page_shift;
383 mem = mm_iommu_lookup(stt->kvm->mm, ua, 1ULL << shift);
384 if (!mem || mm_iommu_ua_to_hpa(mem, ua, shift, &hpa)) {
395 * Handles TCE requests for emulated devices.
396 * Puts guest TCE values to the table and expects user space to convert them.
397 * Cannot fail so kvmppc_tce_validate must be called before it.
399 static void kvmppc_tce_put(struct kvmppc_spapr_tce_table *stt,
400 unsigned long idx, unsigned long tce)
404 unsigned long sttpage;
407 sttpage = idx / TCES_PER_PAGE;
408 page = stt->pages[sttpage];
411 /* We allow any TCE, not just with read|write permissions */
415 page = kvm_spapr_get_tce_page(stt, sttpage);
419 tbl = page_to_virt(page);
421 tbl[idx % TCES_PER_PAGE] = tce;
424 static void kvmppc_clear_tce(struct mm_struct *mm, struct iommu_table *tbl,
427 unsigned long hpa = 0;
428 enum dma_data_direction dir = DMA_NONE;
430 iommu_tce_xchg_no_kill(mm, tbl, entry, &hpa, &dir);
433 static long kvmppc_tce_iommu_mapped_dec(struct kvm *kvm,
434 struct iommu_table *tbl, unsigned long entry)
436 struct mm_iommu_table_group_mem_t *mem = NULL;
437 const unsigned long pgsize = 1ULL << tbl->it_page_shift;
438 __be64 *pua = IOMMU_TABLE_USERSPACE_ENTRY_RO(tbl, entry);
443 mem = mm_iommu_lookup(kvm->mm, be64_to_cpu(*pua), pgsize);
447 mm_iommu_mapped_dec(mem);
449 *pua = cpu_to_be64(0);
454 static long kvmppc_tce_iommu_do_unmap(struct kvm *kvm,
455 struct iommu_table *tbl, unsigned long entry)
457 enum dma_data_direction dir = DMA_NONE;
458 unsigned long hpa = 0;
461 if (WARN_ON_ONCE(iommu_tce_xchg_no_kill(kvm->mm, tbl, entry, &hpa,
468 ret = kvmppc_tce_iommu_mapped_dec(kvm, tbl, entry);
469 if (ret != H_SUCCESS)
470 iommu_tce_xchg_no_kill(kvm->mm, tbl, entry, &hpa, &dir);
475 static long kvmppc_tce_iommu_unmap(struct kvm *kvm,
476 struct kvmppc_spapr_tce_table *stt, struct iommu_table *tbl,
479 unsigned long i, ret = H_SUCCESS;
480 unsigned long subpages = 1ULL << (stt->page_shift - tbl->it_page_shift);
481 unsigned long io_entry = entry * subpages;
483 for (i = 0; i < subpages; ++i) {
484 ret = kvmppc_tce_iommu_do_unmap(kvm, tbl, io_entry + i);
485 if (ret != H_SUCCESS)
492 static long kvmppc_tce_iommu_do_map(struct kvm *kvm, struct iommu_table *tbl,
493 unsigned long entry, unsigned long ua,
494 enum dma_data_direction dir)
498 __be64 *pua = IOMMU_TABLE_USERSPACE_ENTRY(tbl, entry);
499 struct mm_iommu_table_group_mem_t *mem;
502 /* it_userspace allocation might be delayed */
505 mem = mm_iommu_lookup(kvm->mm, ua, 1ULL << tbl->it_page_shift);
507 /* This only handles v2 IOMMU type, v1 is handled via ioctl() */
510 if (WARN_ON_ONCE(mm_iommu_ua_to_hpa(mem, ua, tbl->it_page_shift, &hpa)))
513 if (mm_iommu_mapped_inc(mem))
516 ret = iommu_tce_xchg_no_kill(kvm->mm, tbl, entry, &hpa, &dir);
517 if (WARN_ON_ONCE(ret)) {
518 mm_iommu_mapped_dec(mem);
523 kvmppc_tce_iommu_mapped_dec(kvm, tbl, entry);
525 *pua = cpu_to_be64(ua);
530 static long kvmppc_tce_iommu_map(struct kvm *kvm,
531 struct kvmppc_spapr_tce_table *stt, struct iommu_table *tbl,
532 unsigned long entry, unsigned long ua,
533 enum dma_data_direction dir)
535 unsigned long i, pgoff, ret = H_SUCCESS;
536 unsigned long subpages = 1ULL << (stt->page_shift - tbl->it_page_shift);
537 unsigned long io_entry = entry * subpages;
539 for (i = 0, pgoff = 0; i < subpages;
540 ++i, pgoff += IOMMU_PAGE_SIZE(tbl)) {
542 ret = kvmppc_tce_iommu_do_map(kvm, tbl,
543 io_entry + i, ua + pgoff, dir);
544 if (ret != H_SUCCESS)
551 long kvmppc_h_put_tce(struct kvm_vcpu *vcpu, unsigned long liobn,
552 unsigned long ioba, unsigned long tce)
554 struct kvmppc_spapr_tce_table *stt;
556 struct kvmppc_spapr_tce_iommu_table *stit;
557 unsigned long entry, ua = 0;
558 enum dma_data_direction dir;
560 /* udbg_printf("H_PUT_TCE(): liobn=0x%lx ioba=0x%lx, tce=0x%lx\n", */
561 /* liobn, ioba, tce); */
563 stt = kvmppc_find_table(vcpu->kvm, liobn);
567 ret = kvmppc_ioba_validate(stt, ioba, 1);
568 if (ret != H_SUCCESS)
571 idx = srcu_read_lock(&vcpu->kvm->srcu);
573 ret = kvmppc_tce_validate(stt, tce);
574 if (ret != H_SUCCESS)
577 dir = iommu_tce_direction(tce);
579 if ((dir != DMA_NONE) && kvmppc_tce_to_ua(vcpu->kvm, tce, &ua)) {
584 entry = ioba >> stt->page_shift;
586 list_for_each_entry_lockless(stit, &stt->iommu_tables, next) {
588 ret = kvmppc_tce_iommu_unmap(vcpu->kvm, stt,
591 ret = kvmppc_tce_iommu_map(vcpu->kvm, stt, stit->tbl,
594 iommu_tce_kill(stit->tbl, entry, 1);
596 if (ret != H_SUCCESS) {
597 kvmppc_clear_tce(vcpu->kvm->mm, stit->tbl, entry);
602 kvmppc_tce_put(stt, entry, tce);
605 srcu_read_unlock(&vcpu->kvm->srcu, idx);
609 EXPORT_SYMBOL_GPL(kvmppc_h_put_tce);
611 long kvmppc_h_put_tce_indirect(struct kvm_vcpu *vcpu,
612 unsigned long liobn, unsigned long ioba,
613 unsigned long tce_list, unsigned long npages)
615 struct kvmppc_spapr_tce_table *stt;
616 long i, ret = H_SUCCESS, idx;
617 unsigned long entry, ua = 0;
620 struct kvmppc_spapr_tce_iommu_table *stit;
622 stt = kvmppc_find_table(vcpu->kvm, liobn);
626 entry = ioba >> stt->page_shift;
628 * SPAPR spec says that the maximum size of the list is 512 TCEs
629 * so the whole table fits in 4K page
634 if (tce_list & (SZ_4K - 1))
637 ret = kvmppc_ioba_validate(stt, ioba, npages);
638 if (ret != H_SUCCESS)
641 idx = srcu_read_lock(&vcpu->kvm->srcu);
642 if (kvmppc_tce_to_ua(vcpu->kvm, tce_list, &ua)) {
646 tces = (u64 __user *) ua;
648 for (i = 0; i < npages; ++i) {
649 if (get_user(tce, tces + i)) {
653 tce = be64_to_cpu(tce);
655 ret = kvmppc_tce_validate(stt, tce);
656 if (ret != H_SUCCESS)
660 for (i = 0; i < npages; ++i) {
662 * This looks unsafe, because we validate, then regrab
663 * the TCE from userspace which could have been changed by
666 * But it actually is safe, because the relevant checks will be
667 * re-executed in the following code. If userspace tries to
668 * change this dodgily it will result in a messier failure mode
669 * but won't threaten the host.
671 if (get_user(tce, tces + i)) {
673 goto invalidate_exit;
675 tce = be64_to_cpu(tce);
677 if (kvmppc_tce_to_ua(vcpu->kvm, tce, &ua)) {
679 goto invalidate_exit;
682 list_for_each_entry_lockless(stit, &stt->iommu_tables, next) {
683 ret = kvmppc_tce_iommu_map(vcpu->kvm, stt,
684 stit->tbl, entry + i, ua,
685 iommu_tce_direction(tce));
687 if (ret != H_SUCCESS) {
688 kvmppc_clear_tce(vcpu->kvm->mm, stit->tbl,
690 goto invalidate_exit;
694 kvmppc_tce_put(stt, entry + i, tce);
698 list_for_each_entry_lockless(stit, &stt->iommu_tables, next)
699 iommu_tce_kill(stit->tbl, entry, npages);
702 srcu_read_unlock(&vcpu->kvm->srcu, idx);
706 EXPORT_SYMBOL_GPL(kvmppc_h_put_tce_indirect);
708 long kvmppc_h_stuff_tce(struct kvm_vcpu *vcpu,
709 unsigned long liobn, unsigned long ioba,
710 unsigned long tce_value, unsigned long npages)
712 struct kvmppc_spapr_tce_table *stt;
714 struct kvmppc_spapr_tce_iommu_table *stit;
716 stt = kvmppc_find_table(vcpu->kvm, liobn);
720 ret = kvmppc_ioba_validate(stt, ioba, npages);
721 if (ret != H_SUCCESS)
724 /* Check permission bits only to allow userspace poison TCE for debug */
725 if (tce_value & (TCE_PCI_WRITE | TCE_PCI_READ))
728 list_for_each_entry_lockless(stit, &stt->iommu_tables, next) {
729 unsigned long entry = ioba >> stt->page_shift;
731 for (i = 0; i < npages; ++i) {
732 ret = kvmppc_tce_iommu_unmap(vcpu->kvm, stt,
733 stit->tbl, entry + i);
735 if (ret == H_SUCCESS)
738 if (ret == H_TOO_HARD)
739 goto invalidate_exit;
742 kvmppc_clear_tce(vcpu->kvm->mm, stit->tbl, entry);
746 for (i = 0; i < npages; ++i, ioba += (1ULL << stt->page_shift))
747 kvmppc_tce_put(stt, ioba >> stt->page_shift, tce_value);
750 list_for_each_entry_lockless(stit, &stt->iommu_tables, next)
751 iommu_tce_kill(stit->tbl, ioba >> stt->page_shift, npages);
755 EXPORT_SYMBOL_GPL(kvmppc_h_stuff_tce);