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);
142 extern long kvm_spapr_tce_attach_iommu_group(struct kvm *kvm, int tablefd,
143 struct iommu_group *grp)
145 struct kvmppc_spapr_tce_table *stt = NULL;
147 struct iommu_table *tbl = NULL;
148 struct iommu_table_group *table_group;
150 struct kvmppc_spapr_tce_iommu_table *stit;
157 list_for_each_entry_rcu(stt, &kvm->arch.spapr_tce_tables, list) {
158 if (stt == f.file->private_data) {
169 table_group = iommu_group_get_iommudata(grp);
170 if (WARN_ON(!table_group))
173 for (i = 0; i < IOMMU_TABLE_GROUP_MAX_TABLES; ++i) {
174 struct iommu_table *tbltmp = table_group->tables[i];
179 * Make sure hardware table parameters are exactly the same;
180 * this is used in the TCE handlers where boundary checks
181 * use only the first attached table.
183 if ((tbltmp->it_page_shift == stt->page_shift) &&
184 (tbltmp->it_offset == stt->offset) &&
185 (tbltmp->it_size == stt->size)) {
187 * Reference the table to avoid races with
188 * add/remove DMA windows.
190 tbl = iommu_tce_table_get(tbltmp);
197 list_for_each_entry_rcu(stit, &stt->iommu_tables, next) {
198 if (tbl != stit->tbl)
201 if (!kref_get_unless_zero(&stit->kref)) {
202 /* stit is being destroyed */
203 iommu_tce_table_put(tbl);
207 * The table is already known to this KVM, we just increased
208 * its KVM reference counter and can return.
213 stit = kzalloc(sizeof(*stit), GFP_KERNEL);
215 iommu_tce_table_put(tbl);
220 kref_init(&stit->kref);
222 list_add_rcu(&stit->next, &stt->iommu_tables);
227 static void release_spapr_tce_table(struct rcu_head *head)
229 struct kvmppc_spapr_tce_table *stt = container_of(head,
230 struct kvmppc_spapr_tce_table, rcu);
231 unsigned long i, npages = kvmppc_tce_pages(stt->size);
233 for (i = 0; i < npages; i++)
234 __free_page(stt->pages[i]);
239 static int kvm_spapr_tce_fault(struct vm_fault *vmf)
241 struct kvmppc_spapr_tce_table *stt = vmf->vma->vm_file->private_data;
244 if (vmf->pgoff >= kvmppc_tce_pages(stt->size))
245 return VM_FAULT_SIGBUS;
247 page = stt->pages[vmf->pgoff];
253 static const struct vm_operations_struct kvm_spapr_tce_vm_ops = {
254 .fault = kvm_spapr_tce_fault,
257 static int kvm_spapr_tce_mmap(struct file *file, struct vm_area_struct *vma)
259 vma->vm_ops = &kvm_spapr_tce_vm_ops;
263 static int kvm_spapr_tce_release(struct inode *inode, struct file *filp)
265 struct kvmppc_spapr_tce_table *stt = filp->private_data;
266 struct kvmppc_spapr_tce_iommu_table *stit, *tmp;
267 struct kvm *kvm = stt->kvm;
269 mutex_lock(&kvm->lock);
270 list_del_rcu(&stt->list);
271 mutex_unlock(&kvm->lock);
273 list_for_each_entry_safe(stit, tmp, &stt->iommu_tables, next) {
274 WARN_ON(!kref_read(&stit->kref));
276 if (kref_put(&stit->kref, kvm_spapr_tce_liobn_put))
281 kvm_put_kvm(stt->kvm);
283 kvmppc_account_memlimit(
284 kvmppc_stt_pages(kvmppc_tce_pages(stt->size)), false);
285 call_rcu(&stt->rcu, release_spapr_tce_table);
290 static const struct file_operations kvm_spapr_tce_fops = {
291 .mmap = kvm_spapr_tce_mmap,
292 .release = kvm_spapr_tce_release,
295 long kvm_vm_ioctl_create_spapr_tce(struct kvm *kvm,
296 struct kvm_create_spapr_tce_64 *args)
298 struct kvmppc_spapr_tce_table *stt = NULL;
299 struct kvmppc_spapr_tce_table *siter;
300 unsigned long npages, size;
307 size = _ALIGN_UP(args->size, PAGE_SIZE >> 3);
308 npages = kvmppc_tce_pages(size);
309 ret = kvmppc_account_memlimit(kvmppc_stt_pages(npages), true);
314 stt = kzalloc(sizeof(*stt) + npages * sizeof(struct page *),
319 stt->liobn = args->liobn;
320 stt->page_shift = args->page_shift;
321 stt->offset = args->offset;
324 INIT_LIST_HEAD_RCU(&stt->iommu_tables);
326 for (i = 0; i < npages; i++) {
327 stt->pages[i] = alloc_page(GFP_KERNEL | __GFP_ZERO);
332 mutex_lock(&kvm->lock);
334 /* Check this LIOBN hasn't been previously allocated */
336 list_for_each_entry(siter, &kvm->arch.spapr_tce_tables, list) {
337 if (siter->liobn == args->liobn) {
344 ret = anon_inode_getfd("kvm-spapr-tce", &kvm_spapr_tce_fops,
345 stt, O_RDWR | O_CLOEXEC);
348 list_add_rcu(&stt->list, &kvm->arch.spapr_tce_tables);
352 mutex_unlock(&kvm->lock);
358 for (i = 0; i < npages; i++)
360 __free_page(stt->pages[i]);
364 kvmppc_account_memlimit(kvmppc_stt_pages(npages), false);
368 static void kvmppc_clear_tce(struct iommu_table *tbl, unsigned long entry)
370 unsigned long hpa = 0;
371 enum dma_data_direction dir = DMA_NONE;
373 iommu_tce_xchg(tbl, entry, &hpa, &dir);
376 static long kvmppc_tce_iommu_mapped_dec(struct kvm *kvm,
377 struct iommu_table *tbl, unsigned long entry)
379 struct mm_iommu_table_group_mem_t *mem = NULL;
380 const unsigned long pgsize = 1ULL << tbl->it_page_shift;
381 unsigned long *pua = IOMMU_TABLE_USERSPACE_ENTRY(tbl, entry);
384 /* it_userspace allocation might be delayed */
387 mem = mm_iommu_lookup(kvm->mm, *pua, pgsize);
391 mm_iommu_mapped_dec(mem);
398 static long kvmppc_tce_iommu_unmap(struct kvm *kvm,
399 struct iommu_table *tbl, unsigned long entry)
401 enum dma_data_direction dir = DMA_NONE;
402 unsigned long hpa = 0;
405 if (WARN_ON_ONCE(iommu_tce_xchg(tbl, entry, &hpa, &dir)))
411 ret = kvmppc_tce_iommu_mapped_dec(kvm, tbl, entry);
412 if (ret != H_SUCCESS)
413 iommu_tce_xchg(tbl, entry, &hpa, &dir);
418 long kvmppc_tce_iommu_map(struct kvm *kvm, struct iommu_table *tbl,
419 unsigned long entry, unsigned long ua,
420 enum dma_data_direction dir)
423 unsigned long hpa, *pua = IOMMU_TABLE_USERSPACE_ENTRY(tbl, entry);
424 struct mm_iommu_table_group_mem_t *mem;
427 /* it_userspace allocation might be delayed */
430 mem = mm_iommu_lookup(kvm->mm, ua, 1ULL << tbl->it_page_shift);
432 /* This only handles v2 IOMMU type, v1 is handled via ioctl() */
435 if (WARN_ON_ONCE(mm_iommu_ua_to_hpa(mem, ua, tbl->it_page_shift, &hpa)))
438 if (mm_iommu_mapped_inc(mem))
441 ret = iommu_tce_xchg(tbl, entry, &hpa, &dir);
442 if (WARN_ON_ONCE(ret)) {
443 mm_iommu_mapped_dec(mem);
448 kvmppc_tce_iommu_mapped_dec(kvm, tbl, entry);
455 long kvmppc_h_put_tce(struct kvm_vcpu *vcpu, unsigned long liobn,
456 unsigned long ioba, unsigned long tce)
458 struct kvmppc_spapr_tce_table *stt;
460 struct kvmppc_spapr_tce_iommu_table *stit;
461 unsigned long entry, ua = 0;
462 enum dma_data_direction dir;
464 /* udbg_printf("H_PUT_TCE(): liobn=0x%lx ioba=0x%lx, tce=0x%lx\n", */
465 /* liobn, ioba, tce); */
467 stt = kvmppc_find_table(vcpu->kvm, liobn);
471 ret = kvmppc_ioba_validate(stt, ioba, 1);
472 if (ret != H_SUCCESS)
475 ret = kvmppc_tce_validate(stt, tce);
476 if (ret != H_SUCCESS)
479 dir = iommu_tce_direction(tce);
481 idx = srcu_read_lock(&vcpu->kvm->srcu);
483 if ((dir != DMA_NONE) && kvmppc_gpa_to_ua(vcpu->kvm,
484 tce & ~(TCE_PCI_READ | TCE_PCI_WRITE), &ua, NULL)) {
489 entry = ioba >> stt->page_shift;
491 list_for_each_entry_lockless(stit, &stt->iommu_tables, next) {
493 ret = kvmppc_tce_iommu_unmap(vcpu->kvm,
496 ret = kvmppc_tce_iommu_map(vcpu->kvm, stit->tbl,
499 if (ret == H_SUCCESS)
502 if (ret == H_TOO_HARD)
506 kvmppc_clear_tce(stit->tbl, entry);
509 kvmppc_tce_put(stt, entry, tce);
512 srcu_read_unlock(&vcpu->kvm->srcu, idx);
516 EXPORT_SYMBOL_GPL(kvmppc_h_put_tce);
518 long kvmppc_h_put_tce_indirect(struct kvm_vcpu *vcpu,
519 unsigned long liobn, unsigned long ioba,
520 unsigned long tce_list, unsigned long npages)
522 struct kvmppc_spapr_tce_table *stt;
523 long i, ret = H_SUCCESS, idx;
524 unsigned long entry, ua = 0;
527 struct kvmppc_spapr_tce_iommu_table *stit;
529 stt = kvmppc_find_table(vcpu->kvm, liobn);
533 entry = ioba >> stt->page_shift;
535 * SPAPR spec says that the maximum size of the list is 512 TCEs
536 * so the whole table fits in 4K page
541 if (tce_list & (SZ_4K - 1))
544 ret = kvmppc_ioba_validate(stt, ioba, npages);
545 if (ret != H_SUCCESS)
548 idx = srcu_read_lock(&vcpu->kvm->srcu);
549 if (kvmppc_gpa_to_ua(vcpu->kvm, tce_list, &ua, NULL)) {
553 tces = (u64 __user *) ua;
555 for (i = 0; i < npages; ++i) {
556 if (get_user(tce, tces + i)) {
560 tce = be64_to_cpu(tce);
562 ret = kvmppc_tce_validate(stt, tce);
563 if (ret != H_SUCCESS)
566 if (kvmppc_gpa_to_ua(vcpu->kvm,
567 tce & ~(TCE_PCI_READ | TCE_PCI_WRITE),
573 list_for_each_entry_lockless(stit, &stt->iommu_tables, next) {
574 ret = kvmppc_tce_iommu_map(vcpu->kvm,
575 stit->tbl, entry + i, ua,
576 iommu_tce_direction(tce));
578 if (ret == H_SUCCESS)
581 if (ret == H_TOO_HARD)
585 kvmppc_clear_tce(stit->tbl, entry);
588 kvmppc_tce_put(stt, entry + i, tce);
592 srcu_read_unlock(&vcpu->kvm->srcu, idx);
596 EXPORT_SYMBOL_GPL(kvmppc_h_put_tce_indirect);
598 long kvmppc_h_stuff_tce(struct kvm_vcpu *vcpu,
599 unsigned long liobn, unsigned long ioba,
600 unsigned long tce_value, unsigned long npages)
602 struct kvmppc_spapr_tce_table *stt;
604 struct kvmppc_spapr_tce_iommu_table *stit;
606 stt = kvmppc_find_table(vcpu->kvm, liobn);
610 ret = kvmppc_ioba_validate(stt, ioba, npages);
611 if (ret != H_SUCCESS)
614 /* Check permission bits only to allow userspace poison TCE for debug */
615 if (tce_value & (TCE_PCI_WRITE | TCE_PCI_READ))
618 list_for_each_entry_lockless(stit, &stt->iommu_tables, next) {
619 unsigned long entry = ioba >> stit->tbl->it_page_shift;
621 for (i = 0; i < npages; ++i) {
622 ret = kvmppc_tce_iommu_unmap(vcpu->kvm,
623 stit->tbl, entry + i);
625 if (ret == H_SUCCESS)
628 if (ret == H_TOO_HARD)
632 kvmppc_clear_tce(stit->tbl, entry);
636 for (i = 0; i < npages; ++i, ioba += (1ULL << stt->page_shift))
637 kvmppc_tce_put(stt, ioba >> stt->page_shift, tce_value);
641 EXPORT_SYMBOL_GPL(kvmppc_h_stuff_tce);