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