arch/powerpc/kvm/book3s_64_vio.c

   1 /*
   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.
   5  *
   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.
  10  *
  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.
  14  *
  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>
  18  */
  19
  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>
  33
  34 #include <asm/kvm_ppc.h>
  35 #include <asm/kvm_book3s.h>
  36 #include <asm/book3s/64/mmu-hash.h>
  37 #include <asm/hvcall.h>
  38 #include <asm/synch.h>
  39 #include <asm/ppc-opcode.h>
  40 #include <asm/kvm_host.h>
  41 #include <asm/udbg.h>
  42 #include <asm/iommu.h>
  43 #include <asm/tce.h>
  44 #include <asm/mmu_context.h>
  45
  46 static unsigned long kvmppc_tce_pages(unsigned long iommu_pages)
  47 {
  48         return ALIGN(iommu_pages * sizeof(u64), PAGE_SIZE) / PAGE_SIZE;
  49 }
  50
  51 static unsigned long kvmppc_stt_pages(unsigned long tce_pages)
  52 {
  53         unsigned long stt_bytes = sizeof(struct kvmppc_spapr_tce_table) +
  54                         (tce_pages * sizeof(struct page *));
  55
  56         return tce_pages + ALIGN(stt_bytes, PAGE_SIZE) / PAGE_SIZE;
  57 }
  58
  59 static long kvmppc_account_memlimit(unsigned long stt_pages, bool inc)
  60 {
  61         long ret = 0;
  62
  63         if (!current || !current->mm)
  64                 return ret; /* process exited */
  65
  66         down_write(&current->mm->mmap_sem);
  67
  68         if (inc) {
  69                 unsigned long locked, lock_limit;
  70
  71                 locked = current->mm->locked_vm + stt_pages;
  72                 lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
  73                 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
  74                         ret = -ENOMEM;
  75                 else
  76                         current->mm->locked_vm += stt_pages;
  77         } else {
  78                 if (WARN_ON_ONCE(stt_pages > current->mm->locked_vm))
  79                         stt_pages = current->mm->locked_vm;
  80
  81                 current->mm->locked_vm -= stt_pages;
  82         }
  83
  84         pr_debug("[%d] RLIMIT_MEMLOCK KVM %c%ld %ld/%ld%s\n", current->pid,
  85                         inc ? '+' : '-',
  86                         stt_pages << PAGE_SHIFT,
  87                         current->mm->locked_vm << PAGE_SHIFT,
  88                         rlimit(RLIMIT_MEMLOCK),
  89                         ret ? " - exceeded" : "");
  90
  91         up_write(&current->mm->mmap_sem);
  92
  93         return ret;
  94 }
  95
  96 static void kvm_spapr_tce_iommu_table_free(struct rcu_head *head)
  97 {
  98         struct kvmppc_spapr_tce_iommu_table *stit = container_of(head,
  99                         struct kvmppc_spapr_tce_iommu_table, rcu);
 100
 101         iommu_tce_table_put(stit->tbl);
 102
 103         kfree(stit);
 104 }
 105
 106 static void kvm_spapr_tce_liobn_put(struct kref *kref)
 107 {
 108         struct kvmppc_spapr_tce_iommu_table *stit = container_of(kref,
 109                         struct kvmppc_spapr_tce_iommu_table, kref);
 110
 111         list_del_rcu(&stit->next);
 112
 113         call_rcu(&stit->rcu, kvm_spapr_tce_iommu_table_free);
 114 }
 115
 116 extern void kvm_spapr_tce_release_iommu_group(struct kvm *kvm,
 117                 struct iommu_group *grp)
 118 {
 119         int i;
 120         struct kvmppc_spapr_tce_table *stt;
 121         struct kvmppc_spapr_tce_iommu_table *stit, *tmp;
 122         struct iommu_table_group *table_group = NULL;
 123
 124         list_for_each_entry_rcu(stt, &kvm->arch.spapr_tce_tables, list) {
 125
 126                 table_group = iommu_group_get_iommudata(grp);
 127                 if (WARN_ON(!table_group))
 128                         continue;
 129
 130                 list_for_each_entry_safe(stit, tmp, &stt->iommu_tables, next) {
 131                         for (i = 0; i < IOMMU_TABLE_GROUP_MAX_TABLES; ++i) {
 132                                 if (table_group->tables[i] != stit->tbl)
 133                                         continue;
 134
 135                                 kref_put(&stit->kref, kvm_spapr_tce_liobn_put);
 136                                 return;
 137                         }
 138                 }
 139         }
 140 }
 141
 142 extern long kvm_spapr_tce_attach_iommu_group(struct kvm *kvm, int tablefd,
 143                 struct iommu_group *grp)
 144 {
 145         struct kvmppc_spapr_tce_table *stt = NULL;
 146         bool found = false;
 147         struct iommu_table *tbl = NULL;
 148         struct iommu_table_group *table_group;
 149         long i;
 150         struct kvmppc_spapr_tce_iommu_table *stit;
 151         struct fd f;
 152
 153         f = fdget(tablefd);
 154         if (!f.file)
 155                 return -EBADF;
 156
 157         list_for_each_entry_rcu(stt, &kvm->arch.spapr_tce_tables, list) {
 158                 if (stt == f.file->private_data) {
 159                         found = true;
 160                         break;
 161                 }
 162         }
 163
 164         fdput(f);
 165
 166         if (!found)
 167                 return -EINVAL;
 168
 169         table_group = iommu_group_get_iommudata(grp);
 170         if (WARN_ON(!table_group))
 171                 return -EFAULT;
 172
 173         for (i = 0; i < IOMMU_TABLE_GROUP_MAX_TABLES; ++i) {
 174                 struct iommu_table *tbltmp = table_group->tables[i];
 175
 176                 if (!tbltmp)
 177                         continue;
 178                 /* Make sure hardware table parameters are compatible */
 179                 if ((tbltmp->it_page_shift <= stt->page_shift) &&
 180                                 (tbltmp->it_offset << tbltmp->it_page_shift ==
 181                                  stt->offset << stt->page_shift) &&
 182                                 (tbltmp->it_size << tbltmp->it_page_shift >=
 183                                  stt->size << stt->page_shift)) {
 184                         /*
 185                          * Reference the table to avoid races with
 186                          * add/remove DMA windows.
 187                          */
 188                         tbl = iommu_tce_table_get(tbltmp);
 189                         break;
 190                 }
 191         }
 192         if (!tbl)
 193                 return -EINVAL;
 194
 195         list_for_each_entry_rcu(stit, &stt->iommu_tables, next) {
 196                 if (tbl != stit->tbl)
 197                         continue;
 198
 199                 if (!kref_get_unless_zero(&stit->kref)) {
 200                         /* stit is being destroyed */
 201                         iommu_tce_table_put(tbl);
 202                         return -ENOTTY;
 203                 }
 204                 /*
 205                  * The table is already known to this KVM, we just increased
 206                  * its KVM reference counter and can return.
 207                  */
 208                 return 0;
 209         }
 210
 211         stit = kzalloc(sizeof(*stit), GFP_KERNEL);
 212         if (!stit) {
 213                 iommu_tce_table_put(tbl);
 214                 return -ENOMEM;
 215         }
 216
 217         stit->tbl = tbl;
 218         kref_init(&stit->kref);
 219
 220         list_add_rcu(&stit->next, &stt->iommu_tables);
 221
 222         return 0;
 223 }
 224
 225 static void release_spapr_tce_table(struct rcu_head *head)
 226 {
 227         struct kvmppc_spapr_tce_table *stt = container_of(head,
 228                         struct kvmppc_spapr_tce_table, rcu);
 229         unsigned long i, npages = kvmppc_tce_pages(stt->size);
 230
 231         for (i = 0; i < npages; i++)
 232                 __free_page(stt->pages[i]);
 233
 234         kfree(stt);
 235 }
 236
 237 static vm_fault_t kvm_spapr_tce_fault(struct vm_fault *vmf)
 238 {
 239         struct kvmppc_spapr_tce_table *stt = vmf->vma->vm_file->private_data;
 240         struct page *page;
 241
 242         if (vmf->pgoff >= kvmppc_tce_pages(stt->size))
 243                 return VM_FAULT_SIGBUS;
 244
 245         page = stt->pages[vmf->pgoff];
 246         get_page(page);
 247         vmf->page = page;
 248         return 0;
 249 }
 250
 251 static const struct vm_operations_struct kvm_spapr_tce_vm_ops = {
 252         .fault = kvm_spapr_tce_fault,
 253 };
 254
 255 static int kvm_spapr_tce_mmap(struct file *file, struct vm_area_struct *vma)
 256 {
 257         vma->vm_ops = &kvm_spapr_tce_vm_ops;
 258         return 0;
 259 }
 260
 261 static int kvm_spapr_tce_release(struct inode *inode, struct file *filp)
 262 {
 263         struct kvmppc_spapr_tce_table *stt = filp->private_data;
 264         struct kvmppc_spapr_tce_iommu_table *stit, *tmp;
 265         struct kvm *kvm = stt->kvm;
 266
 267         mutex_lock(&kvm->lock);
 268         list_del_rcu(&stt->list);
 269         mutex_unlock(&kvm->lock);
 270
 271         list_for_each_entry_safe(stit, tmp, &stt->iommu_tables, next) {
 272                 WARN_ON(!kref_read(&stit->kref));
 273                 while (1) {
 274                         if (kref_put(&stit->kref, kvm_spapr_tce_liobn_put))
 275                                 break;
 276                 }
 277         }
 278
 279         kvm_put_kvm(stt->kvm);
 280
 281         kvmppc_account_memlimit(
 282                 kvmppc_stt_pages(kvmppc_tce_pages(stt->size)), false);
 283         call_rcu(&stt->rcu, release_spapr_tce_table);
 284
 285         return 0;
 286 }
 287
 288 static const struct file_operations kvm_spapr_tce_fops = {
 289         .mmap           = kvm_spapr_tce_mmap,
 290         .release        = kvm_spapr_tce_release,
 291 };
 292
 293 long kvm_vm_ioctl_create_spapr_tce(struct kvm *kvm,
 294                                    struct kvm_create_spapr_tce_64 *args)
 295 {
 296         struct kvmppc_spapr_tce_table *stt = NULL;
 297         struct kvmppc_spapr_tce_table *siter;
 298         unsigned long npages, size = args->size;
 299         int ret = -ENOMEM;
 300         int i;
 301
 302         if (!args->size || args->page_shift < 12 || args->page_shift > 34 ||
 303                 (args->offset + args->size > (ULLONG_MAX >> args->page_shift)))
 304                 return -EINVAL;
 305
 306         npages = kvmppc_tce_pages(size);
 307         ret = kvmppc_account_memlimit(kvmppc_stt_pages(npages), true);
 308         if (ret)
 309                 return ret;
 310
 311         ret = -ENOMEM;
 312         stt = kzalloc(sizeof(*stt) + npages * sizeof(struct page *),
 313                       GFP_KERNEL);
 314         if (!stt)
 315                 goto fail_acct;
 316
 317         stt->liobn = args->liobn;
 318         stt->page_shift = args->page_shift;
 319         stt->offset = args->offset;
 320         stt->size = size;
 321         stt->kvm = kvm;
 322         INIT_LIST_HEAD_RCU(&stt->iommu_tables);
 323
 324         for (i = 0; i < npages; i++) {
 325                 stt->pages[i] = alloc_page(GFP_KERNEL | __GFP_ZERO);
 326                 if (!stt->pages[i])
 327                         goto fail;
 328         }
 329
 330         mutex_lock(&kvm->lock);
 331
 332         /* Check this LIOBN hasn't been previously allocated */
 333         ret = 0;
 334         list_for_each_entry(siter, &kvm->arch.spapr_tce_tables, list) {
 335                 if (siter->liobn == args->liobn) {
 336                         ret = -EBUSY;
 337                         break;
 338                 }
 339         }
 340
 341         if (!ret)
 342                 ret = anon_inode_getfd("kvm-spapr-tce", &kvm_spapr_tce_fops,
 343                                        stt, O_RDWR | O_CLOEXEC);
 344
 345         if (ret >= 0) {
 346                 list_add_rcu(&stt->list, &kvm->arch.spapr_tce_tables);
 347                 kvm_get_kvm(kvm);
 348         }
 349
 350         mutex_unlock(&kvm->lock);
 351
 352         if (ret >= 0)
 353                 return ret;
 354
 355  fail:
 356         for (i = 0; i < npages; i++)
 357                 if (stt->pages[i])
 358                         __free_page(stt->pages[i]);
 359
 360         kfree(stt);
 361  fail_acct:
 362         kvmppc_account_memlimit(kvmppc_stt_pages(npages), false);
 363         return ret;
 364 }
 365
 366 static long kvmppc_tce_validate(struct kvmppc_spapr_tce_table *stt,
 367                 unsigned long tce)
 368 {
 369         unsigned long gpa = tce & ~(TCE_PCI_READ | TCE_PCI_WRITE);
 370         enum dma_data_direction dir = iommu_tce_direction(tce);
 371         struct kvmppc_spapr_tce_iommu_table *stit;
 372         unsigned long ua = 0;
 373
 374         /* Allow userspace to poison TCE table */
 375         if (dir == DMA_NONE)
 376                 return H_SUCCESS;
 377
 378         if (iommu_tce_check_gpa(stt->page_shift, gpa))
 379                 return H_TOO_HARD;
 380
 381         if (kvmppc_gpa_to_ua(stt->kvm, tce & ~(TCE_PCI_READ | TCE_PCI_WRITE),
 382                                 &ua, NULL))
 383                 return H_TOO_HARD;
 384
 385         list_for_each_entry_rcu(stit, &stt->iommu_tables, next) {
 386                 unsigned long hpa = 0;
 387                 struct mm_iommu_table_group_mem_t *mem;
 388                 long shift = stit->tbl->it_page_shift;
 389
 390                 mem = mm_iommu_lookup(stt->kvm->mm, ua, 1ULL << shift);
 391                 if (!mem)
 392                         return H_TOO_HARD;
 393
 394                 if (mm_iommu_ua_to_hpa(mem, ua, shift, &hpa))
 395                         return H_TOO_HARD;
 396         }
 397
 398         return H_SUCCESS;
 399 }
 400
 401 static void kvmppc_clear_tce(struct iommu_table *tbl, unsigned long entry)
 402 {
 403         unsigned long hpa = 0;
 404         enum dma_data_direction dir = DMA_NONE;
 405
 406         iommu_tce_xchg(tbl, entry, &hpa, &dir);
 407 }
 408
 409 static long kvmppc_tce_iommu_mapped_dec(struct kvm *kvm,
 410                 struct iommu_table *tbl, unsigned long entry)
 411 {
 412         struct mm_iommu_table_group_mem_t *mem = NULL;
 413         const unsigned long pgsize = 1ULL << tbl->it_page_shift;
 414         __be64 *pua = IOMMU_TABLE_USERSPACE_ENTRY(tbl, entry);
 415
 416         if (!pua)
 417                 /* it_userspace allocation might be delayed */
 418                 return H_TOO_HARD;
 419
 420         mem = mm_iommu_lookup(kvm->mm, be64_to_cpu(*pua), pgsize);
 421         if (!mem)
 422                 return H_TOO_HARD;
 423
 424         mm_iommu_mapped_dec(mem);
 425
 426         *pua = cpu_to_be64(0);
 427
 428         return H_SUCCESS;
 429 }
 430
 431 static long kvmppc_tce_iommu_do_unmap(struct kvm *kvm,
 432                 struct iommu_table *tbl, unsigned long entry)
 433 {
 434         enum dma_data_direction dir = DMA_NONE;
 435         unsigned long hpa = 0;
 436         long ret;
 437
 438         if (WARN_ON_ONCE(iommu_tce_xchg(tbl, entry, &hpa, &dir)))
 439                 return H_TOO_HARD;
 440
 441         if (dir == DMA_NONE)
 442                 return H_SUCCESS;
 443
 444         ret = kvmppc_tce_iommu_mapped_dec(kvm, tbl, entry);
 445         if (ret != H_SUCCESS)
 446                 iommu_tce_xchg(tbl, entry, &hpa, &dir);
 447
 448         return ret;
 449 }
 450
 451 static long kvmppc_tce_iommu_unmap(struct kvm *kvm,
 452                 struct kvmppc_spapr_tce_table *stt, struct iommu_table *tbl,
 453                 unsigned long entry)
 454 {
 455         unsigned long i, ret = H_SUCCESS;
 456         unsigned long subpages = 1ULL << (stt->page_shift - tbl->it_page_shift);
 457         unsigned long io_entry = entry * subpages;
 458
 459         for (i = 0; i < subpages; ++i) {
 460                 ret = kvmppc_tce_iommu_do_unmap(kvm, tbl, io_entry + i);
 461                 if (ret != H_SUCCESS)
 462                         break;
 463         }
 464
 465         return ret;
 466 }
 467
 468 long kvmppc_tce_iommu_do_map(struct kvm *kvm, struct iommu_table *tbl,
 469                 unsigned long entry, unsigned long ua,
 470                 enum dma_data_direction dir)
 471 {
 472         long ret;
 473         unsigned long hpa;
 474         __be64 *pua = IOMMU_TABLE_USERSPACE_ENTRY(tbl, entry);
 475         struct mm_iommu_table_group_mem_t *mem;
 476
 477         if (!pua)
 478                 /* it_userspace allocation might be delayed */
 479                 return H_TOO_HARD;
 480
 481         mem = mm_iommu_lookup(kvm->mm, ua, 1ULL << tbl->it_page_shift);
 482         if (!mem)
 483                 /* This only handles v2 IOMMU type, v1 is handled via ioctl() */
 484                 return H_TOO_HARD;
 485
 486         if (WARN_ON_ONCE(mm_iommu_ua_to_hpa(mem, ua, tbl->it_page_shift, &hpa)))
 487                 return H_TOO_HARD;
 488
 489         if (mm_iommu_mapped_inc(mem))
 490                 return H_TOO_HARD;
 491
 492         ret = iommu_tce_xchg(tbl, entry, &hpa, &dir);
 493         if (WARN_ON_ONCE(ret)) {
 494                 mm_iommu_mapped_dec(mem);
 495                 return H_TOO_HARD;
 496         }
 497
 498         if (dir != DMA_NONE)
 499                 kvmppc_tce_iommu_mapped_dec(kvm, tbl, entry);
 500
 501         *pua = cpu_to_be64(ua);
 502
 503         return 0;
 504 }
 505
 506 static long kvmppc_tce_iommu_map(struct kvm *kvm,
 507                 struct kvmppc_spapr_tce_table *stt, struct iommu_table *tbl,
 508                 unsigned long entry, unsigned long ua,
 509                 enum dma_data_direction dir)
 510 {
 511         unsigned long i, pgoff, ret = H_SUCCESS;
 512         unsigned long subpages = 1ULL << (stt->page_shift - tbl->it_page_shift);
 513         unsigned long io_entry = entry * subpages;
 514
 515         for (i = 0, pgoff = 0; i < subpages;
 516                         ++i, pgoff += IOMMU_PAGE_SIZE(tbl)) {
 517
 518                 ret = kvmppc_tce_iommu_do_map(kvm, tbl,
 519                                 io_entry + i, ua + pgoff, dir);
 520                 if (ret != H_SUCCESS)
 521                         break;
 522         }
 523
 524         return ret;
 525 }
 526
 527 long kvmppc_h_put_tce(struct kvm_vcpu *vcpu, unsigned long liobn,
 528                       unsigned long ioba, unsigned long tce)
 529 {
 530         struct kvmppc_spapr_tce_table *stt;
 531         long ret, idx;
 532         struct kvmppc_spapr_tce_iommu_table *stit;
 533         unsigned long entry, ua = 0;
 534         enum dma_data_direction dir;
 535
 536         /* udbg_printf("H_PUT_TCE(): liobn=0x%lx ioba=0x%lx, tce=0x%lx\n", */
 537         /*          liobn, ioba, tce); */
 538
 539         stt = kvmppc_find_table(vcpu->kvm, liobn);
 540         if (!stt)
 541                 return H_TOO_HARD;
 542
 543         ret = kvmppc_ioba_validate(stt, ioba, 1);
 544         if (ret != H_SUCCESS)
 545                 return ret;
 546
 547         ret = kvmppc_tce_validate(stt, tce);
 548         if (ret != H_SUCCESS)
 549                 return ret;
 550
 551         dir = iommu_tce_direction(tce);
 552
 553         idx = srcu_read_lock(&vcpu->kvm->srcu);
 554
 555         if ((dir != DMA_NONE) && kvmppc_gpa_to_ua(vcpu->kvm,
 556                         tce & ~(TCE_PCI_READ | TCE_PCI_WRITE), &ua, NULL)) {
 557                 ret = H_PARAMETER;
 558                 goto unlock_exit;
 559         }
 560
 561         entry = ioba >> stt->page_shift;
 562
 563         list_for_each_entry_lockless(stit, &stt->iommu_tables, next) {
 564                 if (dir == DMA_NONE)
 565                         ret = kvmppc_tce_iommu_unmap(vcpu->kvm, stt,
 566                                         stit->tbl, entry);
 567                 else
 568                         ret = kvmppc_tce_iommu_map(vcpu->kvm, stt, stit->tbl,
 569                                         entry, ua, dir);
 570
 571                 if (ret == H_SUCCESS)
 572                         continue;
 573
 574                 if (ret == H_TOO_HARD)
 575                         goto unlock_exit;
 576
 577                 WARN_ON_ONCE(1);
 578                 kvmppc_clear_tce(stit->tbl, entry);
 579         }
 580
 581         kvmppc_tce_put(stt, entry, tce);
 582
 583 unlock_exit:
 584         srcu_read_unlock(&vcpu->kvm->srcu, idx);
 585
 586         return ret;
 587 }
 588 EXPORT_SYMBOL_GPL(kvmppc_h_put_tce);
 589
 590 long kvmppc_h_put_tce_indirect(struct kvm_vcpu *vcpu,
 591                 unsigned long liobn, unsigned long ioba,
 592                 unsigned long tce_list, unsigned long npages)
 593 {
 594         struct kvmppc_spapr_tce_table *stt;
 595         long i, ret = H_SUCCESS, idx;
 596         unsigned long entry, ua = 0;
 597         u64 __user *tces;
 598         u64 tce;
 599         struct kvmppc_spapr_tce_iommu_table *stit;
 600
 601         stt = kvmppc_find_table(vcpu->kvm, liobn);
 602         if (!stt)
 603                 return H_TOO_HARD;
 604
 605         entry = ioba >> stt->page_shift;
 606         /*
 607          * SPAPR spec says that the maximum size of the list is 512 TCEs
 608          * so the whole table fits in 4K page
 609          */
 610         if (npages > 512)
 611                 return H_PARAMETER;
 612
 613         if (tce_list & (SZ_4K - 1))
 614                 return H_PARAMETER;
 615
 616         ret = kvmppc_ioba_validate(stt, ioba, npages);
 617         if (ret != H_SUCCESS)
 618                 return ret;
 619
 620         idx = srcu_read_lock(&vcpu->kvm->srcu);
 621         if (kvmppc_gpa_to_ua(vcpu->kvm, tce_list, &ua, NULL)) {
 622                 ret = H_TOO_HARD;
 623                 goto unlock_exit;
 624         }
 625         tces = (u64 __user *) ua;
 626
 627         for (i = 0; i < npages; ++i) {
 628                 if (get_user(tce, tces + i)) {
 629                         ret = H_TOO_HARD;
 630                         goto unlock_exit;
 631                 }
 632                 tce = be64_to_cpu(tce);
 633
 634                 ret = kvmppc_tce_validate(stt, tce);
 635                 if (ret != H_SUCCESS)
 636                         goto unlock_exit;
 637         }
 638
 639         for (i = 0; i < npages; ++i) {
 640                 /*
 641                  * This looks unsafe, because we validate, then regrab
 642                  * the TCE from userspace which could have been changed by
 643                  * another thread.
 644                  *
 645                  * But it actually is safe, because the relevant checks will be
 646                  * re-executed in the following code.  If userspace tries to
 647                  * change this dodgily it will result in a messier failure mode
 648                  * but won't threaten the host.
 649                  */
 650                 if (get_user(tce, tces + i)) {
 651                         ret = H_TOO_HARD;
 652                         goto unlock_exit;
 653                 }
 654                 tce = be64_to_cpu(tce);
 655
 656                 if (kvmppc_gpa_to_ua(vcpu->kvm,
 657                                 tce & ~(TCE_PCI_READ | TCE_PCI_WRITE),
 658                                 &ua, NULL))
 659                         return H_PARAMETER;
 660
 661                 list_for_each_entry_lockless(stit, &stt->iommu_tables, next) {
 662                         ret = kvmppc_tce_iommu_map(vcpu->kvm, stt,
 663                                         stit->tbl, entry + i, ua,
 664                                         iommu_tce_direction(tce));
 665
 666                         if (ret == H_SUCCESS)
 667                                 continue;
 668
 669                         if (ret == H_TOO_HARD)
 670                                 goto unlock_exit;
 671
 672                         WARN_ON_ONCE(1);
 673                         kvmppc_clear_tce(stit->tbl, entry);
 674                 }
 675
 676                 kvmppc_tce_put(stt, entry + i, tce);
 677         }
 678
 679 unlock_exit:
 680         srcu_read_unlock(&vcpu->kvm->srcu, idx);
 681
 682         return ret;
 683 }
 684 EXPORT_SYMBOL_GPL(kvmppc_h_put_tce_indirect);
 685
 686 long kvmppc_h_stuff_tce(struct kvm_vcpu *vcpu,
 687                 unsigned long liobn, unsigned long ioba,
 688                 unsigned long tce_value, unsigned long npages)
 689 {
 690         struct kvmppc_spapr_tce_table *stt;
 691         long i, ret;
 692         struct kvmppc_spapr_tce_iommu_table *stit;
 693
 694         stt = kvmppc_find_table(vcpu->kvm, liobn);
 695         if (!stt)
 696                 return H_TOO_HARD;
 697
 698         ret = kvmppc_ioba_validate(stt, ioba, npages);
 699         if (ret != H_SUCCESS)
 700                 return ret;
 701
 702         /* Check permission bits only to allow userspace poison TCE for debug */
 703         if (tce_value & (TCE_PCI_WRITE | TCE_PCI_READ))
 704                 return H_PARAMETER;
 705
 706         list_for_each_entry_lockless(stit, &stt->iommu_tables, next) {
 707                 unsigned long entry = ioba >> stt->page_shift;
 708
 709                 for (i = 0; i < npages; ++i) {
 710                         ret = kvmppc_tce_iommu_unmap(vcpu->kvm, stt,
 711                                         stit->tbl, entry + i);
 712
 713                         if (ret == H_SUCCESS)
 714                                 continue;
 715
 716                         if (ret == H_TOO_HARD)
 717                                 return ret;
 718
 719                         WARN_ON_ONCE(1);
 720                         kvmppc_clear_tce(stit->tbl, entry);
 721                 }
 722         }
 723
 724         for (i = 0; i < npages; ++i, ioba += (1ULL << stt->page_shift))
 725                 kvmppc_tce_put(stt, ioba >> stt->page_shift, tce_value);
 726
 727         return H_SUCCESS;
 728 }
 729 EXPORT_SYMBOL_GPL(kvmppc_h_stuff_tce);