2 * This program is free software; you can redistribute it and/or modify
3 * it under the terms of the GNU General Public License, version 2, as
4 * published by the Free Software Foundation.
6 * This program is distributed in the hope that it will be useful,
7 * but WITHOUT ANY WARRANTY; without even the implied warranty of
8 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
9 * GNU General Public License for more details.
11 * You should have received a copy of the GNU General Public License
12 * along with this program; if not, write to the Free Software
13 * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
15 * Copyright 2010 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
16 * Copyright 2011 David Gibson, IBM Corporation <dwg@au1.ibm.com>
17 * Copyright 2016 Alexey Kardashevskiy, IBM Corporation <aik@au1.ibm.com>
20 #include <linux/types.h>
21 #include <linux/string.h>
22 #include <linux/kvm.h>
23 #include <linux/kvm_host.h>
24 #include <linux/highmem.h>
25 #include <linux/gfp.h>
26 #include <linux/slab.h>
27 #include <linux/sched/signal.h>
28 #include <linux/hugetlb.h>
29 #include <linux/list.h>
30 #include <linux/anon_inodes.h>
31 #include <linux/iommu.h>
32 #include <linux/file.h>
34 #include <asm/tlbflush.h>
35 #include <asm/kvm_ppc.h>
36 #include <asm/kvm_book3s.h>
37 #include <asm/book3s/64/mmu-hash.h>
38 #include <asm/hvcall.h>
39 #include <asm/synch.h>
40 #include <asm/ppc-opcode.h>
41 #include <asm/kvm_host.h>
43 #include <asm/iommu.h>
45 #include <asm/mmu_context.h>
47 static unsigned long kvmppc_tce_pages(unsigned long iommu_pages)
49 return ALIGN(iommu_pages * sizeof(u64), PAGE_SIZE) / PAGE_SIZE;
52 static unsigned long kvmppc_stt_pages(unsigned long tce_pages)
54 unsigned long stt_bytes = sizeof(struct kvmppc_spapr_tce_table) +
55 (tce_pages * sizeof(struct page *));
57 return tce_pages + ALIGN(stt_bytes, PAGE_SIZE) / PAGE_SIZE;
60 static long kvmppc_account_memlimit(unsigned long stt_pages, bool inc)
64 if (!current || !current->mm)
65 return ret; /* process exited */
67 down_write(¤t->mm->mmap_sem);
70 unsigned long locked, lock_limit;
72 locked = current->mm->locked_vm + stt_pages;
73 lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
74 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
77 current->mm->locked_vm += stt_pages;
79 if (WARN_ON_ONCE(stt_pages > current->mm->locked_vm))
80 stt_pages = current->mm->locked_vm;
82 current->mm->locked_vm -= stt_pages;
85 pr_debug("[%d] RLIMIT_MEMLOCK KVM %c%ld %ld/%ld%s\n", current->pid,
87 stt_pages << PAGE_SHIFT,
88 current->mm->locked_vm << PAGE_SHIFT,
89 rlimit(RLIMIT_MEMLOCK),
90 ret ? " - exceeded" : "");
92 up_write(¤t->mm->mmap_sem);
97 static void kvm_spapr_tce_iommu_table_free(struct rcu_head *head)
99 struct kvmppc_spapr_tce_iommu_table *stit = container_of(head,
100 struct kvmppc_spapr_tce_iommu_table, rcu);
102 iommu_tce_table_put(stit->tbl);
107 static void kvm_spapr_tce_liobn_put(struct kref *kref)
109 struct kvmppc_spapr_tce_iommu_table *stit = container_of(kref,
110 struct kvmppc_spapr_tce_iommu_table, kref);
112 list_del_rcu(&stit->next);
114 call_rcu(&stit->rcu, kvm_spapr_tce_iommu_table_free);
117 extern void kvm_spapr_tce_release_iommu_group(struct kvm *kvm,
118 struct iommu_group *grp)
121 struct kvmppc_spapr_tce_table *stt;
122 struct kvmppc_spapr_tce_iommu_table *stit, *tmp;
123 struct iommu_table_group *table_group = NULL;
125 list_for_each_entry_rcu(stt, &kvm->arch.spapr_tce_tables, list) {
127 table_group = iommu_group_get_iommudata(grp);
128 if (WARN_ON(!table_group))
131 list_for_each_entry_safe(stit, tmp, &stt->iommu_tables, next) {
132 for (i = 0; i < IOMMU_TABLE_GROUP_MAX_TABLES; ++i) {
133 if (table_group->tables[i] != stit->tbl)
136 kref_put(&stit->kref, kvm_spapr_tce_liobn_put);
143 extern long kvm_spapr_tce_attach_iommu_group(struct kvm *kvm, int tablefd,
144 struct iommu_group *grp)
146 struct kvmppc_spapr_tce_table *stt = NULL;
148 struct iommu_table *tbl = NULL;
149 struct iommu_table_group *table_group;
151 struct kvmppc_spapr_tce_iommu_table *stit;
158 list_for_each_entry_rcu(stt, &kvm->arch.spapr_tce_tables, list) {
159 if (stt == f.file->private_data) {
170 table_group = iommu_group_get_iommudata(grp);
171 if (WARN_ON(!table_group))
174 for (i = 0; i < IOMMU_TABLE_GROUP_MAX_TABLES; ++i) {
175 struct iommu_table *tbltmp = table_group->tables[i];
180 * Make sure hardware table parameters are exactly the same;
181 * this is used in the TCE handlers where boundary checks
182 * use only the first attached table.
184 if ((tbltmp->it_page_shift == stt->page_shift) &&
185 (tbltmp->it_offset == stt->offset) &&
186 (tbltmp->it_size == stt->size)) {
188 * Reference the table to avoid races with
189 * add/remove DMA windows.
191 tbl = iommu_tce_table_get(tbltmp);
198 list_for_each_entry_rcu(stit, &stt->iommu_tables, next) {
199 if (tbl != stit->tbl)
202 if (!kref_get_unless_zero(&stit->kref)) {
203 /* stit is being destroyed */
204 iommu_tce_table_put(tbl);
208 * The table is already known to this KVM, we just increased
209 * its KVM reference counter and can return.
214 stit = kzalloc(sizeof(*stit), GFP_KERNEL);
216 iommu_tce_table_put(tbl);
221 kref_init(&stit->kref);
223 list_add_rcu(&stit->next, &stt->iommu_tables);
228 static void release_spapr_tce_table(struct rcu_head *head)
230 struct kvmppc_spapr_tce_table *stt = container_of(head,
231 struct kvmppc_spapr_tce_table, rcu);
232 unsigned long i, npages = kvmppc_tce_pages(stt->size);
234 for (i = 0; i < npages; i++)
235 __free_page(stt->pages[i]);
240 static int kvm_spapr_tce_fault(struct vm_fault *vmf)
242 struct kvmppc_spapr_tce_table *stt = vmf->vma->vm_file->private_data;
245 if (vmf->pgoff >= kvmppc_tce_pages(stt->size))
246 return VM_FAULT_SIGBUS;
248 page = stt->pages[vmf->pgoff];
254 static const struct vm_operations_struct kvm_spapr_tce_vm_ops = {
255 .fault = kvm_spapr_tce_fault,
258 static int kvm_spapr_tce_mmap(struct file *file, struct vm_area_struct *vma)
260 vma->vm_ops = &kvm_spapr_tce_vm_ops;
264 static int kvm_spapr_tce_release(struct inode *inode, struct file *filp)
266 struct kvmppc_spapr_tce_table *stt = filp->private_data;
267 struct kvmppc_spapr_tce_iommu_table *stit, *tmp;
269 list_del_rcu(&stt->list);
271 list_for_each_entry_safe(stit, tmp, &stt->iommu_tables, next) {
272 WARN_ON(!kref_read(&stit->kref));
274 if (kref_put(&stit->kref, kvm_spapr_tce_liobn_put))
279 kvm_put_kvm(stt->kvm);
281 kvmppc_account_memlimit(
282 kvmppc_stt_pages(kvmppc_tce_pages(stt->size)), false);
283 call_rcu(&stt->rcu, release_spapr_tce_table);
288 static const struct file_operations kvm_spapr_tce_fops = {
289 .mmap = kvm_spapr_tce_mmap,
290 .release = kvm_spapr_tce_release,
293 long kvm_vm_ioctl_create_spapr_tce(struct kvm *kvm,
294 struct kvm_create_spapr_tce_64 *args)
296 struct kvmppc_spapr_tce_table *stt = NULL;
297 unsigned long npages, size;
304 /* Check this LIOBN hasn't been previously allocated */
305 list_for_each_entry(stt, &kvm->arch.spapr_tce_tables, list) {
306 if (stt->liobn == args->liobn)
310 size = _ALIGN_UP(args->size, PAGE_SIZE >> 3);
311 npages = kvmppc_tce_pages(size);
312 ret = kvmppc_account_memlimit(kvmppc_stt_pages(npages), true);
319 stt = kzalloc(sizeof(*stt) + npages * sizeof(struct page *),
324 stt->liobn = args->liobn;
325 stt->page_shift = args->page_shift;
326 stt->offset = args->offset;
329 INIT_LIST_HEAD_RCU(&stt->iommu_tables);
331 for (i = 0; i < npages; i++) {
332 stt->pages[i] = alloc_page(GFP_KERNEL | __GFP_ZERO);
339 mutex_lock(&kvm->lock);
340 list_add_rcu(&stt->list, &kvm->arch.spapr_tce_tables);
342 mutex_unlock(&kvm->lock);
344 return anon_inode_getfd("kvm-spapr-tce", &kvm_spapr_tce_fops,
345 stt, O_RDWR | O_CLOEXEC);
349 for (i = 0; i < npages; i++)
351 __free_page(stt->pages[i]);
358 static void kvmppc_clear_tce(struct iommu_table *tbl, unsigned long entry)
360 unsigned long hpa = 0;
361 enum dma_data_direction dir = DMA_NONE;
363 iommu_tce_xchg(tbl, entry, &hpa, &dir);
366 static long kvmppc_tce_iommu_mapped_dec(struct kvm *kvm,
367 struct iommu_table *tbl, unsigned long entry)
369 struct mm_iommu_table_group_mem_t *mem = NULL;
370 const unsigned long pgsize = 1ULL << tbl->it_page_shift;
371 unsigned long *pua = IOMMU_TABLE_USERSPACE_ENTRY(tbl, entry);
374 /* it_userspace allocation might be delayed */
377 mem = mm_iommu_lookup(kvm->mm, *pua, pgsize);
381 mm_iommu_mapped_dec(mem);
388 static long kvmppc_tce_iommu_unmap(struct kvm *kvm,
389 struct iommu_table *tbl, unsigned long entry)
391 enum dma_data_direction dir = DMA_NONE;
392 unsigned long hpa = 0;
395 if (WARN_ON_ONCE(iommu_tce_xchg(tbl, entry, &hpa, &dir)))
401 ret = kvmppc_tce_iommu_mapped_dec(kvm, tbl, entry);
402 if (ret != H_SUCCESS)
403 iommu_tce_xchg(tbl, entry, &hpa, &dir);
408 long kvmppc_tce_iommu_map(struct kvm *kvm, struct iommu_table *tbl,
409 unsigned long entry, unsigned long ua,
410 enum dma_data_direction dir)
413 unsigned long hpa, *pua = IOMMU_TABLE_USERSPACE_ENTRY(tbl, entry);
414 struct mm_iommu_table_group_mem_t *mem;
417 /* it_userspace allocation might be delayed */
420 mem = mm_iommu_lookup(kvm->mm, ua, 1ULL << tbl->it_page_shift);
422 /* This only handles v2 IOMMU type, v1 is handled via ioctl() */
425 if (WARN_ON_ONCE(mm_iommu_ua_to_hpa(mem, ua, &hpa)))
428 if (mm_iommu_mapped_inc(mem))
431 ret = iommu_tce_xchg(tbl, entry, &hpa, &dir);
432 if (WARN_ON_ONCE(ret)) {
433 mm_iommu_mapped_dec(mem);
438 kvmppc_tce_iommu_mapped_dec(kvm, tbl, entry);
445 long kvmppc_h_put_tce(struct kvm_vcpu *vcpu, unsigned long liobn,
446 unsigned long ioba, unsigned long tce)
448 struct kvmppc_spapr_tce_table *stt;
450 struct kvmppc_spapr_tce_iommu_table *stit;
451 unsigned long entry, ua = 0;
452 enum dma_data_direction dir;
454 /* udbg_printf("H_PUT_TCE(): liobn=0x%lx ioba=0x%lx, tce=0x%lx\n", */
455 /* liobn, ioba, tce); */
457 stt = kvmppc_find_table(vcpu->kvm, liobn);
461 ret = kvmppc_ioba_validate(stt, ioba, 1);
462 if (ret != H_SUCCESS)
465 ret = kvmppc_tce_validate(stt, tce);
466 if (ret != H_SUCCESS)
469 dir = iommu_tce_direction(tce);
470 if ((dir != DMA_NONE) && kvmppc_gpa_to_ua(vcpu->kvm,
471 tce & ~(TCE_PCI_READ | TCE_PCI_WRITE), &ua, NULL))
474 entry = ioba >> stt->page_shift;
476 list_for_each_entry_lockless(stit, &stt->iommu_tables, next) {
477 if (dir == DMA_NONE) {
478 ret = kvmppc_tce_iommu_unmap(vcpu->kvm,
481 idx = srcu_read_lock(&vcpu->kvm->srcu);
482 ret = kvmppc_tce_iommu_map(vcpu->kvm, stit->tbl,
484 srcu_read_unlock(&vcpu->kvm->srcu, idx);
487 if (ret == H_SUCCESS)
490 if (ret == H_TOO_HARD)
494 kvmppc_clear_tce(stit->tbl, entry);
497 kvmppc_tce_put(stt, entry, tce);
501 EXPORT_SYMBOL_GPL(kvmppc_h_put_tce);
503 long kvmppc_h_put_tce_indirect(struct kvm_vcpu *vcpu,
504 unsigned long liobn, unsigned long ioba,
505 unsigned long tce_list, unsigned long npages)
507 struct kvmppc_spapr_tce_table *stt;
508 long i, ret = H_SUCCESS, idx;
509 unsigned long entry, ua = 0;
512 struct kvmppc_spapr_tce_iommu_table *stit;
514 stt = kvmppc_find_table(vcpu->kvm, liobn);
518 entry = ioba >> stt->page_shift;
520 * SPAPR spec says that the maximum size of the list is 512 TCEs
521 * so the whole table fits in 4K page
526 if (tce_list & (SZ_4K - 1))
529 ret = kvmppc_ioba_validate(stt, ioba, npages);
530 if (ret != H_SUCCESS)
533 idx = srcu_read_lock(&vcpu->kvm->srcu);
534 if (kvmppc_gpa_to_ua(vcpu->kvm, tce_list, &ua, NULL)) {
538 tces = (u64 __user *) ua;
540 for (i = 0; i < npages; ++i) {
541 if (get_user(tce, tces + i)) {
545 tce = be64_to_cpu(tce);
547 ret = kvmppc_tce_validate(stt, tce);
548 if (ret != H_SUCCESS)
551 if (kvmppc_gpa_to_ua(vcpu->kvm,
552 tce & ~(TCE_PCI_READ | TCE_PCI_WRITE),
556 list_for_each_entry_lockless(stit, &stt->iommu_tables, next) {
557 ret = kvmppc_tce_iommu_map(vcpu->kvm,
558 stit->tbl, entry + i, ua,
559 iommu_tce_direction(tce));
561 if (ret == H_SUCCESS)
564 if (ret == H_TOO_HARD)
568 kvmppc_clear_tce(stit->tbl, entry);
571 kvmppc_tce_put(stt, entry + i, tce);
575 srcu_read_unlock(&vcpu->kvm->srcu, idx);
579 EXPORT_SYMBOL_GPL(kvmppc_h_put_tce_indirect);
581 long kvmppc_h_stuff_tce(struct kvm_vcpu *vcpu,
582 unsigned long liobn, unsigned long ioba,
583 unsigned long tce_value, unsigned long npages)
585 struct kvmppc_spapr_tce_table *stt;
587 struct kvmppc_spapr_tce_iommu_table *stit;
589 stt = kvmppc_find_table(vcpu->kvm, liobn);
593 ret = kvmppc_ioba_validate(stt, ioba, npages);
594 if (ret != H_SUCCESS)
597 /* Check permission bits only to allow userspace poison TCE for debug */
598 if (tce_value & (TCE_PCI_WRITE | TCE_PCI_READ))
601 list_for_each_entry_lockless(stit, &stt->iommu_tables, next) {
602 unsigned long entry = ioba >> stit->tbl->it_page_shift;
604 for (i = 0; i < npages; ++i) {
605 ret = kvmppc_tce_iommu_unmap(vcpu->kvm,
606 stit->tbl, entry + i);
608 if (ret == H_SUCCESS)
611 if (ret == H_TOO_HARD)
615 kvmppc_clear_tce(stit->tbl, entry);
619 for (i = 0; i < npages; ++i, ioba += (1ULL << stt->page_shift))
620 kvmppc_tce_put(stt, ioba >> stt->page_shift, tce_value);
624 EXPORT_SYMBOL_GPL(kvmppc_h_stuff_tce);