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/tlbflush.h>
  35 #include <asm/kvm_ppc.h>
  36 #include <asm/kvm_book3s.h>
  37 #include <asm/book3s/64/mmu-hash.h>
  38 #include <asm/hvcall.h>
  39 #include <asm/synch.h>
  40 #include <asm/ppc-opcode.h>
  41 #include <asm/kvm_host.h>
  42 #include <asm/udbg.h>
  43 #include <asm/iommu.h>
  44 #include <asm/tce.h>
  45 #include <asm/mmu_context.h>
  46
  47 static unsigned long kvmppc_tce_pages(unsigned long iommu_pages)
  48 {
  49         return ALIGN(iommu_pages * sizeof(u64), PAGE_SIZE) / PAGE_SIZE;
  50 }
  51
  52 static unsigned long kvmppc_stt_pages(unsigned long tce_pages)
  53 {
  54         unsigned long stt_bytes = sizeof(struct kvmppc_spapr_tce_table) +
  55                         (tce_pages * sizeof(struct page *));
  56
  57         return tce_pages + ALIGN(stt_bytes, PAGE_SIZE) / PAGE_SIZE;
  58 }
  59
  60 static long kvmppc_account_memlimit(unsigned long stt_pages, bool inc)
  61 {
  62         long ret = 0;
  63
  64         if (!current || !current->mm)
  65                 return ret; /* process exited */
  66
  67         down_write(&current->mm->mmap_sem);
  68
  69         if (inc) {
  70                 unsigned long locked, lock_limit;
  71
  72                 locked = current->mm->locked_vm + stt_pages;
  73                 lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
  74                 if (locked > lock_limit && !capable(CAP_IPC_LOCK))
  75                         ret = -ENOMEM;
  76                 else
  77                         current->mm->locked_vm += stt_pages;
  78         } else {
  79                 if (WARN_ON_ONCE(stt_pages > current->mm->locked_vm))
  80                         stt_pages = current->mm->locked_vm;
  81
  82                 current->mm->locked_vm -= stt_pages;
  83         }
  84
  85         pr_debug("[%d] RLIMIT_MEMLOCK KVM %c%ld %ld/%ld%s\n", current->pid,
  86                         inc ? '+' : '-',
  87                         stt_pages << PAGE_SHIFT,
  88                         current->mm->locked_vm << PAGE_SHIFT,
  89                         rlimit(RLIMIT_MEMLOCK),
  90                         ret ? " - exceeded" : "");
  91
  92         up_write(&current->mm->mmap_sem);
  93
  94         return ret;
  95 }
  96
  97 static void kvm_spapr_tce_iommu_table_free(struct rcu_head *head)
  98 {
  99         struct kvmppc_spapr_tce_iommu_table *stit = container_of(head,
 100                         struct kvmppc_spapr_tce_iommu_table, rcu);
 101
 102         iommu_tce_table_put(stit->tbl);
 103
 104         kfree(stit);
 105 }
 106
 107 static void kvm_spapr_tce_liobn_put(struct kref *kref)
 108 {
 109         struct kvmppc_spapr_tce_iommu_table *stit = container_of(kref,
 110                         struct kvmppc_spapr_tce_iommu_table, kref);
 111
 112         list_del_rcu(&stit->next);
 113
 114         call_rcu(&stit->rcu, kvm_spapr_tce_iommu_table_free);
 115 }
 116
 117 extern void kvm_spapr_tce_release_iommu_group(struct kvm *kvm,
 118                 struct iommu_group *grp)
 119 {
 120         int i;
 121         struct kvmppc_spapr_tce_table *stt;
 122         struct kvmppc_spapr_tce_iommu_table *stit, *tmp;
 123         struct iommu_table_group *table_group = NULL;
 124
 125         list_for_each_entry_rcu(stt, &kvm->arch.spapr_tce_tables, list) {
 126
 127                 table_group = iommu_group_get_iommudata(grp);
 128                 if (WARN_ON(!table_group))
 129                         continue;
 130
 131                 list_for_each_entry_safe(stit, tmp, &stt->iommu_tables, next) {
 132                         for (i = 0; i < IOMMU_TABLE_GROUP_MAX_TABLES; ++i) {
 133                                 if (table_group->tables[i] != stit->tbl)
 134                                         continue;
 135
 136                                 kref_put(&stit->kref, kvm_spapr_tce_liobn_put);
 137                                 return;
 138                         }
 139                 }
 140         }
 141 }
 142
 143 extern long kvm_spapr_tce_attach_iommu_group(struct kvm *kvm, int tablefd,
 144                 struct iommu_group *grp)
 145 {
 146         struct kvmppc_spapr_tce_table *stt = NULL;
 147         bool found = false;
 148         struct iommu_table *tbl = NULL;
 149         struct iommu_table_group *table_group;
 150         long i;
 151         struct kvmppc_spapr_tce_iommu_table *stit;
 152         struct fd f;
 153
 154         f = fdget(tablefd);
 155         if (!f.file)
 156                 return -EBADF;
 157
 158         list_for_each_entry_rcu(stt, &kvm->arch.spapr_tce_tables, list) {
 159                 if (stt == f.file->private_data) {
 160                         found = true;
 161                         break;
 162                 }
 163         }
 164
 165         fdput(f);
 166
 167         if (!found)
 168                 return -EINVAL;
 169
 170         table_group = iommu_group_get_iommudata(grp);
 171         if (WARN_ON(!table_group))
 172                 return -EFAULT;
 173
 174         for (i = 0; i < IOMMU_TABLE_GROUP_MAX_TABLES; ++i) {
 175                 struct iommu_table *tbltmp = table_group->tables[i];
 176
 177                 if (!tbltmp)
 178                         continue;
 179                 /*
 180                  * Make sure hardware table parameters are exactly the same;
 181                  * this is used in the TCE handlers where boundary checks
 182                  * use only the first attached table.
 183                  */
 184                 if ((tbltmp->it_page_shift == stt->page_shift) &&
 185                                 (tbltmp->it_offset == stt->offset) &&
 186                                 (tbltmp->it_size == stt->size)) {
 187                         /*
 188                          * Reference the table to avoid races with
 189                          * add/remove DMA windows.
 190                          */
 191                         tbl = iommu_tce_table_get(tbltmp);
 192                         break;
 193                 }
 194         }
 195         if (!tbl)
 196                 return -EINVAL;
 197
 198         list_for_each_entry_rcu(stit, &stt->iommu_tables, next) {
 199                 if (tbl != stit->tbl)
 200                         continue;
 201
 202                 if (!kref_get_unless_zero(&stit->kref)) {
 203                         /* stit is being destroyed */
 204                         iommu_tce_table_put(tbl);
 205                         return -ENOTTY;
 206                 }
 207                 /*
 208                  * The table is already known to this KVM, we just increased
 209                  * its KVM reference counter and can return.
 210                  */
 211                 return 0;
 212         }
 213
 214         stit = kzalloc(sizeof(*stit), GFP_KERNEL);
 215         if (!stit) {
 216                 iommu_tce_table_put(tbl);
 217                 return -ENOMEM;
 218         }
 219
 220         stit->tbl = tbl;
 221         kref_init(&stit->kref);
 222
 223         list_add_rcu(&stit->next, &stt->iommu_tables);
 224
 225         return 0;
 226 }
 227
 228 static void release_spapr_tce_table(struct rcu_head *head)
 229 {
 230         struct kvmppc_spapr_tce_table *stt = container_of(head,
 231                         struct kvmppc_spapr_tce_table, rcu);
 232         unsigned long i, npages = kvmppc_tce_pages(stt->size);
 233
 234         for (i = 0; i < npages; i++)
 235                 __free_page(stt->pages[i]);
 236
 237         kfree(stt);
 238 }
 239
 240 static int kvm_spapr_tce_fault(struct vm_fault *vmf)
 241 {
 242         struct kvmppc_spapr_tce_table *stt = vmf->vma->vm_file->private_data;
 243         struct page *page;
 244
 245         if (vmf->pgoff >= kvmppc_tce_pages(stt->size))
 246                 return VM_FAULT_SIGBUS;
 247
 248         page = stt->pages[vmf->pgoff];
 249         get_page(page);
 250         vmf->page = page;
 251         return 0;
 252 }
 253
 254 static const struct vm_operations_struct kvm_spapr_tce_vm_ops = {
 255         .fault = kvm_spapr_tce_fault,
 256 };
 257
 258 static int kvm_spapr_tce_mmap(struct file *file, struct vm_area_struct *vma)
 259 {
 260         vma->vm_ops = &kvm_spapr_tce_vm_ops;
 261         return 0;
 262 }
 263
 264 static int kvm_spapr_tce_release(struct inode *inode, struct file *filp)
 265 {
 266         struct kvmppc_spapr_tce_table *stt = filp->private_data;
 267         struct kvmppc_spapr_tce_iommu_table *stit, *tmp;
 268         struct kvm *kvm = stt->kvm;
 269
 270         mutex_lock(&kvm->lock);
 271         list_del_rcu(&stt->list);
 272         mutex_unlock(&kvm->lock);
 273
 274         list_for_each_entry_safe(stit, tmp, &stt->iommu_tables, next) {
 275                 WARN_ON(!kref_read(&stit->kref));
 276                 while (1) {
 277                         if (kref_put(&stit->kref, kvm_spapr_tce_liobn_put))
 278                                 break;
 279                 }
 280         }
 281
 282         kvm_put_kvm(stt->kvm);
 283
 284         kvmppc_account_memlimit(
 285                 kvmppc_stt_pages(kvmppc_tce_pages(stt->size)), false);
 286         call_rcu(&stt->rcu, release_spapr_tce_table);
 287
 288         return 0;
 289 }
 290
 291 static const struct file_operations kvm_spapr_tce_fops = {
 292         .mmap           = kvm_spapr_tce_mmap,
 293         .release        = kvm_spapr_tce_release,
 294 };
 295
 296 long kvm_vm_ioctl_create_spapr_tce(struct kvm *kvm,
 297                                    struct kvm_create_spapr_tce_64 *args)
 298 {
 299         struct kvmppc_spapr_tce_table *stt = NULL;
 300         struct kvmppc_spapr_tce_table *siter;
 301         unsigned long npages, size;
 302         int ret = -ENOMEM;
 303         int i;
 304
 305         if (!args->size)
 306                 return -EINVAL;
 307
 308         size = _ALIGN_UP(args->size, PAGE_SIZE >> 3);
 309         npages = kvmppc_tce_pages(size);
 310         ret = kvmppc_account_memlimit(kvmppc_stt_pages(npages), true);
 311         if (ret)
 312                 return ret;
 313
 314         ret = -ENOMEM;
 315         stt = kzalloc(sizeof(*stt) + npages * sizeof(struct page *),
 316                       GFP_KERNEL);
 317         if (!stt)
 318                 goto fail_acct;
 319
 320         stt->liobn = args->liobn;
 321         stt->page_shift = args->page_shift;
 322         stt->offset = args->offset;
 323         stt->size = size;
 324         stt->kvm = kvm;
 325         INIT_LIST_HEAD_RCU(&stt->iommu_tables);
 326
 327         for (i = 0; i < npages; i++) {
 328                 stt->pages[i] = alloc_page(GFP_KERNEL | __GFP_ZERO);
 329                 if (!stt->pages[i])
 330                         goto fail;
 331         }
 332
 333         mutex_lock(&kvm->lock);
 334
 335         /* Check this LIOBN hasn't been previously allocated */
 336         ret = 0;
 337         list_for_each_entry(siter, &kvm->arch.spapr_tce_tables, list) {
 338                 if (siter->liobn == args->liobn) {
 339                         ret = -EBUSY;
 340                         break;
 341                 }
 342         }
 343
 344         if (!ret)
 345                 ret = anon_inode_getfd("kvm-spapr-tce", &kvm_spapr_tce_fops,
 346                                        stt, O_RDWR | O_CLOEXEC);
 347
 348         if (ret >= 0) {
 349                 list_add_rcu(&stt->list, &kvm->arch.spapr_tce_tables);
 350                 kvm_get_kvm(kvm);
 351         }
 352
 353         mutex_unlock(&kvm->lock);
 354
 355         if (ret >= 0)
 356                 return ret;
 357
 358  fail:
 359         for (i = 0; i < npages; i++)
 360                 if (stt->pages[i])
 361                         __free_page(stt->pages[i]);
 362
 363         kfree(stt);
 364  fail_acct:
 365         kvmppc_account_memlimit(kvmppc_stt_pages(npages), false);
 366         return ret;
 367 }
 368
 369 static void kvmppc_clear_tce(struct iommu_table *tbl, unsigned long entry)
 370 {
 371         unsigned long hpa = 0;
 372         enum dma_data_direction dir = DMA_NONE;
 373
 374         iommu_tce_xchg(tbl, entry, &hpa, &dir);
 375 }
 376
 377 static long kvmppc_tce_iommu_mapped_dec(struct kvm *kvm,
 378                 struct iommu_table *tbl, unsigned long entry)
 379 {
 380         struct mm_iommu_table_group_mem_t *mem = NULL;
 381         const unsigned long pgsize = 1ULL << tbl->it_page_shift;
 382         unsigned long *pua = IOMMU_TABLE_USERSPACE_ENTRY(tbl, entry);
 383
 384         if (!pua)
 385                 /* it_userspace allocation might be delayed */
 386                 return H_TOO_HARD;
 387
 388         mem = mm_iommu_lookup(kvm->mm, *pua, pgsize);
 389         if (!mem)
 390                 return H_TOO_HARD;
 391
 392         mm_iommu_mapped_dec(mem);
 393
 394         *pua = 0;
 395
 396         return H_SUCCESS;
 397 }
 398
 399 static long kvmppc_tce_iommu_unmap(struct kvm *kvm,
 400                 struct iommu_table *tbl, unsigned long entry)
 401 {
 402         enum dma_data_direction dir = DMA_NONE;
 403         unsigned long hpa = 0;
 404         long ret;
 405
 406         if (WARN_ON_ONCE(iommu_tce_xchg(tbl, entry, &hpa, &dir)))
 407                 return H_HARDWARE;
 408
 409         if (dir == DMA_NONE)
 410                 return H_SUCCESS;
 411
 412         ret = kvmppc_tce_iommu_mapped_dec(kvm, tbl, entry);
 413         if (ret != H_SUCCESS)
 414                 iommu_tce_xchg(tbl, entry, &hpa, &dir);
 415
 416         return ret;
 417 }
 418
 419 long kvmppc_tce_iommu_map(struct kvm *kvm, struct iommu_table *tbl,
 420                 unsigned long entry, unsigned long ua,
 421                 enum dma_data_direction dir)
 422 {
 423         long ret;
 424         unsigned long hpa, *pua = IOMMU_TABLE_USERSPACE_ENTRY(tbl, entry);
 425         struct mm_iommu_table_group_mem_t *mem;
 426
 427         if (!pua)
 428                 /* it_userspace allocation might be delayed */
 429                 return H_TOO_HARD;
 430
 431         mem = mm_iommu_lookup(kvm->mm, ua, 1ULL << tbl->it_page_shift);
 432         if (!mem)
 433                 /* This only handles v2 IOMMU type, v1 is handled via ioctl() */
 434                 return H_TOO_HARD;
 435
 436         if (WARN_ON_ONCE(mm_iommu_ua_to_hpa(mem, ua, &hpa)))
 437                 return H_HARDWARE;
 438
 439         if (mm_iommu_mapped_inc(mem))
 440                 return H_CLOSED;
 441
 442         ret = iommu_tce_xchg(tbl, entry, &hpa, &dir);
 443         if (WARN_ON_ONCE(ret)) {
 444                 mm_iommu_mapped_dec(mem);
 445                 return H_HARDWARE;
 446         }
 447
 448         if (dir != DMA_NONE)
 449                 kvmppc_tce_iommu_mapped_dec(kvm, tbl, entry);
 450
 451         *pua = ua;
 452
 453         return 0;
 454 }
 455
 456 long kvmppc_h_put_tce(struct kvm_vcpu *vcpu, unsigned long liobn,
 457                       unsigned long ioba, unsigned long tce)
 458 {
 459         struct kvmppc_spapr_tce_table *stt;
 460         long ret, idx;
 461         struct kvmppc_spapr_tce_iommu_table *stit;
 462         unsigned long entry, ua = 0;
 463         enum dma_data_direction dir;
 464
 465         /* udbg_printf("H_PUT_TCE(): liobn=0x%lx ioba=0x%lx, tce=0x%lx\n", */
 466         /*          liobn, ioba, tce); */
 467
 468         stt = kvmppc_find_table(vcpu->kvm, liobn);
 469         if (!stt)
 470                 return H_TOO_HARD;
 471
 472         ret = kvmppc_ioba_validate(stt, ioba, 1);
 473         if (ret != H_SUCCESS)
 474                 return ret;
 475
 476         ret = kvmppc_tce_validate(stt, tce);
 477         if (ret != H_SUCCESS)
 478                 return ret;
 479
 480         dir = iommu_tce_direction(tce);
 481         if ((dir != DMA_NONE) && kvmppc_gpa_to_ua(vcpu->kvm,
 482                         tce & ~(TCE_PCI_READ | TCE_PCI_WRITE), &ua, NULL))
 483                 return H_PARAMETER;
 484
 485         entry = ioba >> stt->page_shift;
 486
 487         list_for_each_entry_lockless(stit, &stt->iommu_tables, next) {
 488                 if (dir == DMA_NONE) {
 489                         ret = kvmppc_tce_iommu_unmap(vcpu->kvm,
 490                                         stit->tbl, entry);
 491                 } else {
 492                         idx = srcu_read_lock(&vcpu->kvm->srcu);
 493                         ret = kvmppc_tce_iommu_map(vcpu->kvm, stit->tbl,
 494                                         entry, ua, dir);
 495                         srcu_read_unlock(&vcpu->kvm->srcu, idx);
 496                 }
 497
 498                 if (ret == H_SUCCESS)
 499                         continue;
 500
 501                 if (ret == H_TOO_HARD)
 502                         return ret;
 503
 504                 WARN_ON_ONCE(1);
 505                 kvmppc_clear_tce(stit->tbl, entry);
 506         }
 507
 508         kvmppc_tce_put(stt, entry, tce);
 509
 510         return H_SUCCESS;
 511 }
 512 EXPORT_SYMBOL_GPL(kvmppc_h_put_tce);
 513
 514 long kvmppc_h_put_tce_indirect(struct kvm_vcpu *vcpu,
 515                 unsigned long liobn, unsigned long ioba,
 516                 unsigned long tce_list, unsigned long npages)
 517 {
 518         struct kvmppc_spapr_tce_table *stt;
 519         long i, ret = H_SUCCESS, idx;
 520         unsigned long entry, ua = 0;
 521         u64 __user *tces;
 522         u64 tce;
 523         struct kvmppc_spapr_tce_iommu_table *stit;
 524
 525         stt = kvmppc_find_table(vcpu->kvm, liobn);
 526         if (!stt)
 527                 return H_TOO_HARD;
 528
 529         entry = ioba >> stt->page_shift;
 530         /*
 531          * SPAPR spec says that the maximum size of the list is 512 TCEs
 532          * so the whole table fits in 4K page
 533          */
 534         if (npages > 512)
 535                 return H_PARAMETER;
 536
 537         if (tce_list & (SZ_4K - 1))
 538                 return H_PARAMETER;
 539
 540         ret = kvmppc_ioba_validate(stt, ioba, npages);
 541         if (ret != H_SUCCESS)
 542                 return ret;
 543
 544         idx = srcu_read_lock(&vcpu->kvm->srcu);
 545         if (kvmppc_gpa_to_ua(vcpu->kvm, tce_list, &ua, NULL)) {
 546                 ret = H_TOO_HARD;
 547                 goto unlock_exit;
 548         }
 549         tces = (u64 __user *) ua;
 550
 551         for (i = 0; i < npages; ++i) {
 552                 if (get_user(tce, tces + i)) {
 553                         ret = H_TOO_HARD;
 554                         goto unlock_exit;
 555                 }
 556                 tce = be64_to_cpu(tce);
 557
 558                 ret = kvmppc_tce_validate(stt, tce);
 559                 if (ret != H_SUCCESS)
 560                         goto unlock_exit;
 561
 562                 if (kvmppc_gpa_to_ua(vcpu->kvm,
 563                                 tce & ~(TCE_PCI_READ | TCE_PCI_WRITE),
 564                                 &ua, NULL))
 565                         return H_PARAMETER;
 566
 567                 list_for_each_entry_lockless(stit, &stt->iommu_tables, next) {
 568                         ret = kvmppc_tce_iommu_map(vcpu->kvm,
 569                                         stit->tbl, entry + i, ua,
 570                                         iommu_tce_direction(tce));
 571
 572                         if (ret == H_SUCCESS)
 573                                 continue;
 574
 575                         if (ret == H_TOO_HARD)
 576                                 goto unlock_exit;
 577
 578                         WARN_ON_ONCE(1);
 579                         kvmppc_clear_tce(stit->tbl, entry);
 580                 }
 581
 582                 kvmppc_tce_put(stt, entry + i, tce);
 583         }
 584
 585 unlock_exit:
 586         srcu_read_unlock(&vcpu->kvm->srcu, idx);
 587
 588         return ret;
 589 }
 590 EXPORT_SYMBOL_GPL(kvmppc_h_put_tce_indirect);
 591
 592 long kvmppc_h_stuff_tce(struct kvm_vcpu *vcpu,
 593                 unsigned long liobn, unsigned long ioba,
 594                 unsigned long tce_value, unsigned long npages)
 595 {
 596         struct kvmppc_spapr_tce_table *stt;
 597         long i, ret;
 598         struct kvmppc_spapr_tce_iommu_table *stit;
 599
 600         stt = kvmppc_find_table(vcpu->kvm, liobn);
 601         if (!stt)
 602                 return H_TOO_HARD;
 603
 604         ret = kvmppc_ioba_validate(stt, ioba, npages);
 605         if (ret != H_SUCCESS)
 606                 return ret;
 607
 608         /* Check permission bits only to allow userspace poison TCE for debug */
 609         if (tce_value & (TCE_PCI_WRITE | TCE_PCI_READ))
 610                 return H_PARAMETER;
 611
 612         list_for_each_entry_lockless(stit, &stt->iommu_tables, next) {
 613                 unsigned long entry = ioba >> stit->tbl->it_page_shift;
 614
 615                 for (i = 0; i < npages; ++i) {
 616                         ret = kvmppc_tce_iommu_unmap(vcpu->kvm,
 617                                         stit->tbl, entry + i);
 618
 619                         if (ret == H_SUCCESS)
 620                                 continue;
 621
 622                         if (ret == H_TOO_HARD)
 623                                 return ret;
 624
 625                         WARN_ON_ONCE(1);
 626                         kvmppc_clear_tce(stit->tbl, entry);
 627                 }
 628         }
 629
 630         for (i = 0; i < npages; ++i, ioba += (1ULL << stt->page_shift))
 631                 kvmppc_tce_put(stt, ioba >> stt->page_shift, tce_value);
 632
 633         return H_SUCCESS;
 634 }
 635 EXPORT_SYMBOL_GPL(kvmppc_h_stuff_tce);