arch/powerpc/kvm/book3s_64_vio.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  *
   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>
   7  */
   8
   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>
  22 #include <linux/mm.h>
  23 #include <linux/rcupdate_wait.h>
  24
  25 #include <asm/kvm_ppc.h>
  26 #include <asm/kvm_book3s.h>
  27 #include <asm/book3s/64/mmu-hash.h>
  28 #include <asm/hvcall.h>
  29 #include <asm/synch.h>
  30 #include <asm/ppc-opcode.h>
  31 #include <asm/udbg.h>
  32 #include <asm/iommu.h>
  33 #include <asm/tce.h>
  34 #include <asm/mmu_context.h>
  35
  36 static struct kvmppc_spapr_tce_table *kvmppc_find_table(struct kvm *kvm,
  37         unsigned long liobn)
  38 {
  39         struct kvmppc_spapr_tce_table *stt;
  40
  41         list_for_each_entry_lockless(stt, &kvm->arch.spapr_tce_tables, list)
  42                 if (stt->liobn == liobn)
  43                         return stt;
  44
  45         return NULL;
  46 }
  47
  48 static unsigned long kvmppc_tce_pages(unsigned long iommu_pages)
  49 {
  50         return ALIGN(iommu_pages * sizeof(u64), PAGE_SIZE) / PAGE_SIZE;
  51 }
  52
  53 static unsigned long kvmppc_stt_pages(unsigned long tce_pages)
  54 {
  55         unsigned long stt_bytes = sizeof(struct kvmppc_spapr_tce_table) +
  56                         (tce_pages * sizeof(struct page *));
  57
  58         return tce_pages + ALIGN(stt_bytes, PAGE_SIZE) / PAGE_SIZE;
  59 }
  60
  61 static void kvm_spapr_tce_iommu_table_free(struct rcu_head *head)
  62 {
  63         struct kvmppc_spapr_tce_iommu_table *stit = container_of(head,
  64                         struct kvmppc_spapr_tce_iommu_table, rcu);
  65
  66         iommu_tce_table_put(stit->tbl);
  67
  68         kfree(stit);
  69 }
  70
  71 static void kvm_spapr_tce_liobn_put(struct kref *kref)
  72 {
  73         struct kvmppc_spapr_tce_iommu_table *stit = container_of(kref,
  74                         struct kvmppc_spapr_tce_iommu_table, kref);
  75
  76         list_del_rcu(&stit->next);
  77
  78         call_rcu(&stit->rcu, kvm_spapr_tce_iommu_table_free);
  79 }
  80
  81 void kvm_spapr_tce_release_iommu_group(struct kvm *kvm,
  82                                        struct iommu_group *grp)
  83 {
  84         int i;
  85         struct kvmppc_spapr_tce_table *stt;
  86         struct kvmppc_spapr_tce_iommu_table *stit, *tmp;
  87         struct iommu_table_group *table_group = NULL;
  88
  89         rcu_read_lock();
  90         list_for_each_entry_rcu(stt, &kvm->arch.spapr_tce_tables, list) {
  91
  92                 table_group = iommu_group_get_iommudata(grp);
  93                 if (WARN_ON(!table_group))
  94                         continue;
  95
  96                 list_for_each_entry_safe(stit, tmp, &stt->iommu_tables, next) {
  97                         for (i = 0; i < IOMMU_TABLE_GROUP_MAX_TABLES; ++i) {
  98                                 if (table_group->tables[i] != stit->tbl)
  99                                         continue;
 100
 101                                 kref_put(&stit->kref, kvm_spapr_tce_liobn_put);
 102                         }
 103                 }
 104                 cond_resched_rcu();
 105         }
 106         rcu_read_unlock();
 107 }
 108
 109 long kvm_spapr_tce_attach_iommu_group(struct kvm *kvm, int tablefd,
 110                                       struct iommu_group *grp)
 111 {
 112         struct kvmppc_spapr_tce_table *stt = NULL;
 113         bool found = false;
 114         struct iommu_table *tbl = NULL;
 115         struct iommu_table_group *table_group;
 116         long i;
 117         struct kvmppc_spapr_tce_iommu_table *stit;
 118         struct fd f;
 119
 120         f = fdget(tablefd);
 121         if (!f.file)
 122                 return -EBADF;
 123
 124         rcu_read_lock();
 125         list_for_each_entry_rcu(stt, &kvm->arch.spapr_tce_tables, list) {
 126                 if (stt == f.file->private_data) {
 127                         found = true;
 128                         break;
 129                 }
 130         }
 131         rcu_read_unlock();
 132
 133         fdput(f);
 134
 135         if (!found)
 136                 return -EINVAL;
 137
 138         table_group = iommu_group_get_iommudata(grp);
 139         if (WARN_ON(!table_group))
 140                 return -EFAULT;
 141
 142         for (i = 0; i < IOMMU_TABLE_GROUP_MAX_TABLES; ++i) {
 143                 struct iommu_table *tbltmp = table_group->tables[i];
 144
 145                 if (!tbltmp)
 146                         continue;
 147                 /* Make sure hardware table parameters are compatible */
 148                 if ((tbltmp->it_page_shift <= stt->page_shift) &&
 149                                 (tbltmp->it_offset << tbltmp->it_page_shift ==
 150                                  stt->offset << stt->page_shift) &&
 151                                 (tbltmp->it_size << tbltmp->it_page_shift >=
 152                                  stt->size << stt->page_shift)) {
 153                         /*
 154                          * Reference the table to avoid races with
 155                          * add/remove DMA windows.
 156                          */
 157                         tbl = iommu_tce_table_get(tbltmp);
 158                         break;
 159                 }
 160         }
 161         if (!tbl)
 162                 return -EINVAL;
 163
 164         rcu_read_lock();
 165         list_for_each_entry_rcu(stit, &stt->iommu_tables, next) {
 166                 if (tbl != stit->tbl)
 167                         continue;
 168
 169                 if (!kref_get_unless_zero(&stit->kref)) {
 170                         /* stit is being destroyed */
 171                         iommu_tce_table_put(tbl);
 172                         rcu_read_unlock();
 173                         return -ENOTTY;
 174                 }
 175                 /*
 176                  * The table is already known to this KVM, we just increased
 177                  * its KVM reference counter and can return.
 178                  */
 179                 rcu_read_unlock();
 180                 return 0;
 181         }
 182         rcu_read_unlock();
 183
 184         stit = kzalloc(sizeof(*stit), GFP_KERNEL);
 185         if (!stit) {
 186                 iommu_tce_table_put(tbl);
 187                 return -ENOMEM;
 188         }
 189
 190         stit->tbl = tbl;
 191         kref_init(&stit->kref);
 192
 193         list_add_rcu(&stit->next, &stt->iommu_tables);
 194
 195         return 0;
 196 }
 197
 198 static void release_spapr_tce_table(struct rcu_head *head)
 199 {
 200         struct kvmppc_spapr_tce_table *stt = container_of(head,
 201                         struct kvmppc_spapr_tce_table, rcu);
 202         unsigned long i, npages = kvmppc_tce_pages(stt->size);
 203
 204         for (i = 0; i < npages; i++)
 205                 if (stt->pages[i])
 206                         __free_page(stt->pages[i]);
 207
 208         kfree(stt);
 209 }
 210
 211 static struct page *kvm_spapr_get_tce_page(struct kvmppc_spapr_tce_table *stt,
 212                 unsigned long sttpage)
 213 {
 214         struct page *page = stt->pages[sttpage];
 215
 216         if (page)
 217                 return page;
 218
 219         mutex_lock(&stt->alloc_lock);
 220         page = stt->pages[sttpage];
 221         if (!page) {
 222                 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
 223                 WARN_ON_ONCE(!page);
 224                 if (page)
 225                         stt->pages[sttpage] = page;
 226         }
 227         mutex_unlock(&stt->alloc_lock);
 228
 229         return page;
 230 }
 231
 232 static vm_fault_t kvm_spapr_tce_fault(struct vm_fault *vmf)
 233 {
 234         struct kvmppc_spapr_tce_table *stt = vmf->vma->vm_file->private_data;
 235         struct page *page;
 236
 237         if (vmf->pgoff >= kvmppc_tce_pages(stt->size))
 238                 return VM_FAULT_SIGBUS;
 239
 240         page = kvm_spapr_get_tce_page(stt, vmf->pgoff);
 241         if (!page)
 242                 return VM_FAULT_OOM;
 243
 244         get_page(page);
 245         vmf->page = page;
 246         return 0;
 247 }
 248
 249 static const struct vm_operations_struct kvm_spapr_tce_vm_ops = {
 250         .fault = kvm_spapr_tce_fault,
 251 };
 252
 253 static int kvm_spapr_tce_mmap(struct file *file, struct vm_area_struct *vma)
 254 {
 255         vma->vm_ops = &kvm_spapr_tce_vm_ops;
 256         return 0;
 257 }
 258
 259 static int kvm_spapr_tce_release(struct inode *inode, struct file *filp)
 260 {
 261         struct kvmppc_spapr_tce_table *stt = filp->private_data;
 262         struct kvmppc_spapr_tce_iommu_table *stit, *tmp;
 263         struct kvm *kvm = stt->kvm;
 264
 265         mutex_lock(&kvm->lock);
 266         list_del_rcu(&stt->list);
 267         mutex_unlock(&kvm->lock);
 268
 269         list_for_each_entry_safe(stit, tmp, &stt->iommu_tables, next) {
 270                 WARN_ON(!kref_read(&stit->kref));
 271                 while (1) {
 272                         if (kref_put(&stit->kref, kvm_spapr_tce_liobn_put))
 273                                 break;
 274                 }
 275         }
 276
 277         account_locked_vm(kvm->mm,
 278                 kvmppc_stt_pages(kvmppc_tce_pages(stt->size)), false);
 279
 280         kvm_put_kvm(stt->kvm);
 281
 282         call_rcu(&stt->rcu, release_spapr_tce_table);
 283
 284         return 0;
 285 }
 286
 287 static const struct file_operations kvm_spapr_tce_fops = {
 288         .mmap           = kvm_spapr_tce_mmap,
 289         .release        = kvm_spapr_tce_release,
 290 };
 291
 292 int kvm_vm_ioctl_create_spapr_tce(struct kvm *kvm,
 293                                   struct kvm_create_spapr_tce_64 *args)
 294 {
 295         struct kvmppc_spapr_tce_table *stt = NULL;
 296         struct kvmppc_spapr_tce_table *siter;
 297         struct mm_struct *mm = kvm->mm;
 298         unsigned long npages;
 299         int ret;
 300
 301         if (!args->size || args->page_shift < 12 || args->page_shift > 34 ||
 302                 (args->offset + args->size > (ULLONG_MAX >> args->page_shift)))
 303                 return -EINVAL;
 304
 305         npages = kvmppc_tce_pages(args->size);
 306         ret = account_locked_vm(mm, kvmppc_stt_pages(npages), true);
 307         if (ret)
 308                 return ret;
 309
 310         ret = -ENOMEM;
 311         stt = kzalloc(struct_size(stt, pages, npages), GFP_KERNEL | __GFP_NOWARN);
 312         if (!stt)
 313                 goto fail_acct;
 314
 315         stt->liobn = args->liobn;
 316         stt->page_shift = args->page_shift;
 317         stt->offset = args->offset;
 318         stt->size = args->size;
 319         stt->kvm = kvm;
 320         mutex_init(&stt->alloc_lock);
 321         INIT_LIST_HEAD_RCU(&stt->iommu_tables);
 322
 323         mutex_lock(&kvm->lock);
 324
 325         /* Check this LIOBN hasn't been previously allocated */
 326         ret = 0;
 327         list_for_each_entry(siter, &kvm->arch.spapr_tce_tables, list) {
 328                 if (siter->liobn == args->liobn) {
 329                         ret = -EBUSY;
 330                         break;
 331                 }
 332         }
 333
 334         kvm_get_kvm(kvm);
 335         if (!ret)
 336                 ret = anon_inode_getfd("kvm-spapr-tce", &kvm_spapr_tce_fops,
 337                                        stt, O_RDWR | O_CLOEXEC);
 338
 339         if (ret >= 0)
 340                 list_add_rcu(&stt->list, &kvm->arch.spapr_tce_tables);
 341         else
 342                 kvm_put_kvm_no_destroy(kvm);
 343
 344         mutex_unlock(&kvm->lock);
 345
 346         if (ret >= 0)
 347                 return ret;
 348
 349         kfree(stt);
 350  fail_acct:
 351         account_locked_vm(mm, kvmppc_stt_pages(npages), false);
 352         return ret;
 353 }
 354
 355 static long kvmppc_tce_to_ua(struct kvm *kvm, unsigned long tce,
 356                 unsigned long *ua)
 357 {
 358         unsigned long gfn = tce >> PAGE_SHIFT;
 359         struct kvm_memory_slot *memslot;
 360
 361         memslot = __gfn_to_memslot(kvm_memslots(kvm), gfn);
 362         if (!memslot)
 363                 return -EINVAL;
 364
 365         *ua = __gfn_to_hva_memslot(memslot, gfn) |
 366                 (tce & ~(PAGE_MASK | TCE_PCI_READ | TCE_PCI_WRITE));
 367
 368         return 0;
 369 }
 370
 371 static long kvmppc_tce_validate(struct kvmppc_spapr_tce_table *stt,
 372                 unsigned long tce)
 373 {
 374         unsigned long gpa = tce & ~(TCE_PCI_READ | TCE_PCI_WRITE);
 375         enum dma_data_direction dir = iommu_tce_direction(tce);
 376         struct kvmppc_spapr_tce_iommu_table *stit;
 377         unsigned long ua = 0;
 378
 379         /* Allow userspace to poison TCE table */
 380         if (dir == DMA_NONE)
 381                 return H_SUCCESS;
 382
 383         if (iommu_tce_check_gpa(stt->page_shift, gpa))
 384                 return H_TOO_HARD;
 385
 386         if (kvmppc_tce_to_ua(stt->kvm, tce, &ua))
 387                 return H_TOO_HARD;
 388
 389         rcu_read_lock();
 390         list_for_each_entry_rcu(stit, &stt->iommu_tables, next) {
 391                 unsigned long hpa = 0;
 392                 struct mm_iommu_table_group_mem_t *mem;
 393                 long shift = stit->tbl->it_page_shift;
 394
 395                 mem = mm_iommu_lookup(stt->kvm->mm, ua, 1ULL << shift);
 396                 if (!mem || mm_iommu_ua_to_hpa(mem, ua, shift, &hpa)) {
 397                         rcu_read_unlock();
 398                         return H_TOO_HARD;
 399                 }
 400         }
 401         rcu_read_unlock();
 402
 403         return H_SUCCESS;
 404 }
 405
 406 /*
 407  * Handles TCE requests for emulated devices.
 408  * Puts guest TCE values to the table and expects user space to convert them.
 409  * Cannot fail so kvmppc_tce_validate must be called before it.
 410  */
 411 static void kvmppc_tce_put(struct kvmppc_spapr_tce_table *stt,
 412                 unsigned long idx, unsigned long tce)
 413 {
 414         struct page *page;
 415         u64 *tbl;
 416         unsigned long sttpage;
 417
 418         idx -= stt->offset;
 419         sttpage = idx / TCES_PER_PAGE;
 420         page = stt->pages[sttpage];
 421
 422         if (!page) {
 423                 /* We allow any TCE, not just with read|write permissions */
 424                 if (!tce)
 425                         return;
 426
 427                 page = kvm_spapr_get_tce_page(stt, sttpage);
 428                 if (!page)
 429                         return;
 430         }
 431         tbl = page_to_virt(page);
 432
 433         tbl[idx % TCES_PER_PAGE] = tce;
 434 }
 435
 436 static void kvmppc_clear_tce(struct mm_struct *mm, struct kvmppc_spapr_tce_table *stt,
 437                 struct iommu_table *tbl, unsigned long entry)
 438 {
 439         unsigned long i;
 440         unsigned long subpages = 1ULL << (stt->page_shift - tbl->it_page_shift);
 441         unsigned long io_entry = entry << (stt->page_shift - tbl->it_page_shift);
 442
 443         for (i = 0; i < subpages; ++i) {
 444                 unsigned long hpa = 0;
 445                 enum dma_data_direction dir = DMA_NONE;
 446
 447                 iommu_tce_xchg_no_kill(mm, tbl, io_entry + i, &hpa, &dir);
 448         }
 449 }
 450
 451 static long kvmppc_tce_iommu_mapped_dec(struct kvm *kvm,
 452                 struct iommu_table *tbl, unsigned long entry)
 453 {
 454         struct mm_iommu_table_group_mem_t *mem = NULL;
 455         const unsigned long pgsize = 1ULL << tbl->it_page_shift;
 456         __be64 *pua = IOMMU_TABLE_USERSPACE_ENTRY_RO(tbl, entry);
 457
 458         if (!pua)
 459                 return H_SUCCESS;
 460
 461         mem = mm_iommu_lookup(kvm->mm, be64_to_cpu(*pua), pgsize);
 462         if (!mem)
 463                 return H_TOO_HARD;
 464
 465         mm_iommu_mapped_dec(mem);
 466
 467         *pua = cpu_to_be64(0);
 468
 469         return H_SUCCESS;
 470 }
 471
 472 static long kvmppc_tce_iommu_do_unmap(struct kvm *kvm,
 473                 struct iommu_table *tbl, unsigned long entry)
 474 {
 475         enum dma_data_direction dir = DMA_NONE;
 476         unsigned long hpa = 0;
 477         long ret;
 478
 479         if (WARN_ON_ONCE(iommu_tce_xchg_no_kill(kvm->mm, tbl, entry, &hpa,
 480                                         &dir)))
 481                 return H_TOO_HARD;
 482
 483         if (dir == DMA_NONE)
 484                 return H_SUCCESS;
 485
 486         ret = kvmppc_tce_iommu_mapped_dec(kvm, tbl, entry);
 487         if (ret != H_SUCCESS)
 488                 iommu_tce_xchg_no_kill(kvm->mm, tbl, entry, &hpa, &dir);
 489
 490         return ret;
 491 }
 492
 493 static long kvmppc_tce_iommu_unmap(struct kvm *kvm,
 494                 struct kvmppc_spapr_tce_table *stt, struct iommu_table *tbl,
 495                 unsigned long entry)
 496 {
 497         unsigned long i, ret = H_SUCCESS;
 498         unsigned long subpages = 1ULL << (stt->page_shift - tbl->it_page_shift);
 499         unsigned long io_entry = entry * subpages;
 500
 501         for (i = 0; i < subpages; ++i) {
 502                 ret = kvmppc_tce_iommu_do_unmap(kvm, tbl, io_entry + i);
 503                 if (ret != H_SUCCESS)
 504                         break;
 505         }
 506
 507         iommu_tce_kill(tbl, io_entry, subpages);
 508
 509         return ret;
 510 }
 511
 512 static long kvmppc_tce_iommu_do_map(struct kvm *kvm, struct iommu_table *tbl,
 513                 unsigned long entry, unsigned long ua,
 514                 enum dma_data_direction dir)
 515 {
 516         long ret;
 517         unsigned long hpa;
 518         __be64 *pua = IOMMU_TABLE_USERSPACE_ENTRY(tbl, entry);
 519         struct mm_iommu_table_group_mem_t *mem;
 520
 521         if (!pua)
 522                 /* it_userspace allocation might be delayed */
 523                 return H_TOO_HARD;
 524
 525         mem = mm_iommu_lookup(kvm->mm, ua, 1ULL << tbl->it_page_shift);
 526         if (!mem)
 527                 /* This only handles v2 IOMMU type, v1 is handled via ioctl() */
 528                 return H_TOO_HARD;
 529
 530         if (WARN_ON_ONCE(mm_iommu_ua_to_hpa(mem, ua, tbl->it_page_shift, &hpa)))
 531                 return H_TOO_HARD;
 532
 533         if (mm_iommu_mapped_inc(mem))
 534                 return H_TOO_HARD;
 535
 536         ret = iommu_tce_xchg_no_kill(kvm->mm, tbl, entry, &hpa, &dir);
 537         if (WARN_ON_ONCE(ret)) {
 538                 mm_iommu_mapped_dec(mem);
 539                 return H_TOO_HARD;
 540         }
 541
 542         if (dir != DMA_NONE)
 543                 kvmppc_tce_iommu_mapped_dec(kvm, tbl, entry);
 544
 545         *pua = cpu_to_be64(ua);
 546
 547         return 0;
 548 }
 549
 550 static long kvmppc_tce_iommu_map(struct kvm *kvm,
 551                 struct kvmppc_spapr_tce_table *stt, struct iommu_table *tbl,
 552                 unsigned long entry, unsigned long ua,
 553                 enum dma_data_direction dir)
 554 {
 555         unsigned long i, pgoff, ret = H_SUCCESS;
 556         unsigned long subpages = 1ULL << (stt->page_shift - tbl->it_page_shift);
 557         unsigned long io_entry = entry * subpages;
 558
 559         for (i = 0, pgoff = 0; i < subpages;
 560                         ++i, pgoff += IOMMU_PAGE_SIZE(tbl)) {
 561
 562                 ret = kvmppc_tce_iommu_do_map(kvm, tbl,
 563                                 io_entry + i, ua + pgoff, dir);
 564                 if (ret != H_SUCCESS)
 565                         break;
 566         }
 567
 568         iommu_tce_kill(tbl, io_entry, subpages);
 569
 570         return ret;
 571 }
 572
 573 long kvmppc_h_put_tce(struct kvm_vcpu *vcpu, unsigned long liobn,
 574                       unsigned long ioba, unsigned long tce)
 575 {
 576         struct kvmppc_spapr_tce_table *stt;
 577         long ret, idx;
 578         struct kvmppc_spapr_tce_iommu_table *stit;
 579         unsigned long entry, ua = 0;
 580         enum dma_data_direction dir;
 581
 582         /* udbg_printf("H_PUT_TCE(): liobn=0x%lx ioba=0x%lx, tce=0x%lx\n", */
 583         /*          liobn, ioba, tce); */
 584
 585         stt = kvmppc_find_table(vcpu->kvm, liobn);
 586         if (!stt)
 587                 return H_TOO_HARD;
 588
 589         ret = kvmppc_ioba_validate(stt, ioba, 1);
 590         if (ret != H_SUCCESS)
 591                 return ret;
 592
 593         idx = srcu_read_lock(&vcpu->kvm->srcu);
 594
 595         ret = kvmppc_tce_validate(stt, tce);
 596         if (ret != H_SUCCESS)
 597                 goto unlock_exit;
 598
 599         dir = iommu_tce_direction(tce);
 600
 601         if ((dir != DMA_NONE) && kvmppc_tce_to_ua(vcpu->kvm, tce, &ua)) {
 602                 ret = H_PARAMETER;
 603                 goto unlock_exit;
 604         }
 605
 606         entry = ioba >> stt->page_shift;
 607
 608         list_for_each_entry_lockless(stit, &stt->iommu_tables, next) {
 609                 if (dir == DMA_NONE)
 610                         ret = kvmppc_tce_iommu_unmap(vcpu->kvm, stt,
 611                                         stit->tbl, entry);
 612                 else
 613                         ret = kvmppc_tce_iommu_map(vcpu->kvm, stt, stit->tbl,
 614                                         entry, ua, dir);
 615
 616
 617                 if (ret != H_SUCCESS) {
 618                         kvmppc_clear_tce(vcpu->kvm->mm, stt, stit->tbl, entry);
 619                         goto unlock_exit;
 620                 }
 621         }
 622
 623         kvmppc_tce_put(stt, entry, tce);
 624
 625 unlock_exit:
 626         srcu_read_unlock(&vcpu->kvm->srcu, idx);
 627
 628         return ret;
 629 }
 630 EXPORT_SYMBOL_GPL(kvmppc_h_put_tce);
 631
 632 long kvmppc_h_put_tce_indirect(struct kvm_vcpu *vcpu,
 633                 unsigned long liobn, unsigned long ioba,
 634                 unsigned long tce_list, unsigned long npages)
 635 {
 636         struct kvmppc_spapr_tce_table *stt;
 637         long i, ret = H_SUCCESS, idx;
 638         unsigned long entry, ua = 0;
 639         u64 __user *tces;
 640         u64 tce;
 641         struct kvmppc_spapr_tce_iommu_table *stit;
 642
 643         stt = kvmppc_find_table(vcpu->kvm, liobn);
 644         if (!stt)
 645                 return H_TOO_HARD;
 646
 647         entry = ioba >> stt->page_shift;
 648         /*
 649          * SPAPR spec says that the maximum size of the list is 512 TCEs
 650          * so the whole table fits in 4K page
 651          */
 652         if (npages > 512)
 653                 return H_PARAMETER;
 654
 655         if (tce_list & (SZ_4K - 1))
 656                 return H_PARAMETER;
 657
 658         ret = kvmppc_ioba_validate(stt, ioba, npages);
 659         if (ret != H_SUCCESS)
 660                 return ret;
 661
 662         idx = srcu_read_lock(&vcpu->kvm->srcu);
 663         if (kvmppc_tce_to_ua(vcpu->kvm, tce_list, &ua)) {
 664                 ret = H_TOO_HARD;
 665                 goto unlock_exit;
 666         }
 667         tces = (u64 __user *) ua;
 668
 669         for (i = 0; i < npages; ++i) {
 670                 if (get_user(tce, tces + i)) {
 671                         ret = H_TOO_HARD;
 672                         goto unlock_exit;
 673                 }
 674                 tce = be64_to_cpu(tce);
 675
 676                 ret = kvmppc_tce_validate(stt, tce);
 677                 if (ret != H_SUCCESS)
 678                         goto unlock_exit;
 679         }
 680
 681         for (i = 0; i < npages; ++i) {
 682                 /*
 683                  * This looks unsafe, because we validate, then regrab
 684                  * the TCE from userspace which could have been changed by
 685                  * another thread.
 686                  *
 687                  * But it actually is safe, because the relevant checks will be
 688                  * re-executed in the following code.  If userspace tries to
 689                  * change this dodgily it will result in a messier failure mode
 690                  * but won't threaten the host.
 691                  */
 692                 if (get_user(tce, tces + i)) {
 693                         ret = H_TOO_HARD;
 694                         goto unlock_exit;
 695                 }
 696                 tce = be64_to_cpu(tce);
 697
 698                 if (kvmppc_tce_to_ua(vcpu->kvm, tce, &ua)) {
 699                         ret = H_PARAMETER;
 700                         goto unlock_exit;
 701                 }
 702
 703                 list_for_each_entry_lockless(stit, &stt->iommu_tables, next) {
 704                         ret = kvmppc_tce_iommu_map(vcpu->kvm, stt,
 705                                         stit->tbl, entry + i, ua,
 706                                         iommu_tce_direction(tce));
 707
 708                         if (ret != H_SUCCESS) {
 709                                 kvmppc_clear_tce(vcpu->kvm->mm, stt, stit->tbl,
 710                                                  entry + i);
 711                                 goto unlock_exit;
 712                         }
 713                 }
 714
 715                 kvmppc_tce_put(stt, entry + i, tce);
 716         }
 717
 718 unlock_exit:
 719         srcu_read_unlock(&vcpu->kvm->srcu, idx);
 720
 721         return ret;
 722 }
 723 EXPORT_SYMBOL_GPL(kvmppc_h_put_tce_indirect);
 724
 725 long kvmppc_h_stuff_tce(struct kvm_vcpu *vcpu,
 726                 unsigned long liobn, unsigned long ioba,
 727                 unsigned long tce_value, unsigned long npages)
 728 {
 729         struct kvmppc_spapr_tce_table *stt;
 730         long i, ret;
 731         struct kvmppc_spapr_tce_iommu_table *stit;
 732
 733         stt = kvmppc_find_table(vcpu->kvm, liobn);
 734         if (!stt)
 735                 return H_TOO_HARD;
 736
 737         ret = kvmppc_ioba_validate(stt, ioba, npages);
 738         if (ret != H_SUCCESS)
 739                 return ret;
 740
 741         /* Check permission bits only to allow userspace poison TCE for debug */
 742         if (tce_value & (TCE_PCI_WRITE | TCE_PCI_READ))
 743                 return H_PARAMETER;
 744
 745         list_for_each_entry_lockless(stit, &stt->iommu_tables, next) {
 746                 unsigned long entry = ioba >> stt->page_shift;
 747
 748                 for (i = 0; i < npages; ++i) {
 749                         ret = kvmppc_tce_iommu_unmap(vcpu->kvm, stt,
 750                                         stit->tbl, entry + i);
 751
 752                         if (ret == H_SUCCESS)
 753                                 continue;
 754
 755                         if (ret == H_TOO_HARD)
 756                                 return ret;
 757
 758                         WARN_ON_ONCE(1);
 759                         kvmppc_clear_tce(vcpu->kvm->mm, stt, stit->tbl, entry + i);
 760                 }
 761         }
 762
 763         for (i = 0; i < npages; ++i, ioba += (1ULL << stt->page_shift))
 764                 kvmppc_tce_put(stt, ioba >> stt->page_shift, tce_value);
 765
 766         return ret;
 767 }
 768 EXPORT_SYMBOL_GPL(kvmppc_h_stuff_tce);
 769
 770 long kvmppc_h_get_tce(struct kvm_vcpu *vcpu, unsigned long liobn,
 771                       unsigned long ioba)
 772 {
 773         struct kvmppc_spapr_tce_table *stt;
 774         long ret;
 775         unsigned long idx;
 776         struct page *page;
 777         u64 *tbl;
 778
 779         stt = kvmppc_find_table(vcpu->kvm, liobn);
 780         if (!stt)
 781                 return H_TOO_HARD;
 782
 783         ret = kvmppc_ioba_validate(stt, ioba, 1);
 784         if (ret != H_SUCCESS)
 785                 return ret;
 786
 787         idx = (ioba >> stt->page_shift) - stt->offset;
 788         page = stt->pages[idx / TCES_PER_PAGE];
 789         if (!page) {
 790                 kvmppc_set_gpr(vcpu, 4, 0);
 791                 return H_SUCCESS;
 792         }
 793         tbl = (u64 *)page_address(page);
 794
 795         kvmppc_set_gpr(vcpu, 4, tbl[idx % TCES_PER_PAGE]);
 796
 797         return H_SUCCESS;
 798 }
 799 EXPORT_SYMBOL_GPL(kvmppc_h_get_tce);