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