arch/powerpc/kvm/book3s_64_vio_hv.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/hugetlb.h>
  17 #include <linux/list.h>
  18 #include <linux/stringify.h>
  19
  20 #include <asm/kvm_ppc.h>
  21 #include <asm/kvm_book3s.h>
  22 #include <asm/book3s/64/mmu-hash.h>
  23 #include <asm/mmu_context.h>
  24 #include <asm/hvcall.h>
  25 #include <asm/synch.h>
  26 #include <asm/ppc-opcode.h>
  27 #include <asm/kvm_host.h>
  28 #include <asm/udbg.h>
  29 #include <asm/iommu.h>
  30 #include <asm/tce.h>
  31 #include <asm/pte-walk.h>
  32
  33 #ifdef CONFIG_BUG
  34
  35 #define WARN_ON_ONCE_RM(condition)      ({                      \
  36         static bool __section(.data.unlikely) __warned;         \
  37         int __ret_warn_once = !!(condition);                    \
  38                                                                 \
  39         if (unlikely(__ret_warn_once && !__warned)) {           \
  40                 __warned = true;                                \
  41                 pr_err("WARN_ON_ONCE_RM: (%s) at %s:%u\n",      \
  42                                 __stringify(condition),         \
  43                                 __func__, __LINE__);            \
  44                 dump_stack();                                   \
  45         }                                                       \
  46         unlikely(__ret_warn_once);                              \
  47 })
  48
  49 #else
  50
  51 #define WARN_ON_ONCE_RM(condition) ({                           \
  52         int __ret_warn_on = !!(condition);                      \
  53         unlikely(__ret_warn_on);                                \
  54 })
  55
  56 #endif
  57
  58 /*
  59  * Finds a TCE table descriptor by LIOBN.
  60  *
  61  * WARNING: This will be called in real or virtual mode on HV KVM and virtual
  62  *          mode on PR KVM
  63  */
  64 struct kvmppc_spapr_tce_table *kvmppc_find_table(struct kvm *kvm,
  65                 unsigned long liobn)
  66 {
  67         struct kvmppc_spapr_tce_table *stt;
  68
  69         list_for_each_entry_lockless(stt, &kvm->arch.spapr_tce_tables, list)
  70                 if (stt->liobn == liobn)
  71                         return stt;
  72
  73         return NULL;
  74 }
  75 EXPORT_SYMBOL_GPL(kvmppc_find_table);
  76
  77 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
  78 static long kvmppc_rm_tce_to_ua(struct kvm *kvm, unsigned long tce,
  79                 unsigned long *ua, unsigned long **prmap)
  80 {
  81         unsigned long gfn = tce >> PAGE_SHIFT;
  82         struct kvm_memory_slot *memslot;
  83
  84         memslot = search_memslots(kvm_memslots_raw(kvm), gfn);
  85         if (!memslot)
  86                 return -EINVAL;
  87
  88         *ua = __gfn_to_hva_memslot(memslot, gfn) |
  89                 (tce & ~(PAGE_MASK | TCE_PCI_READ | TCE_PCI_WRITE));
  90
  91         if (prmap)
  92                 *prmap = &memslot->arch.rmap[gfn - memslot->base_gfn];
  93
  94         return 0;
  95 }
  96
  97 /*
  98  * Validates TCE address.
  99  * At the moment flags and page mask are validated.
 100  * As the host kernel does not access those addresses (just puts them
 101  * to the table and user space is supposed to process them), we can skip
 102  * checking other things (such as TCE is a guest RAM address or the page
 103  * was actually allocated).
 104  */
 105 static long kvmppc_rm_tce_validate(struct kvmppc_spapr_tce_table *stt,
 106                 unsigned long tce)
 107 {
 108         unsigned long gpa = tce & ~(TCE_PCI_READ | TCE_PCI_WRITE);
 109         enum dma_data_direction dir = iommu_tce_direction(tce);
 110         struct kvmppc_spapr_tce_iommu_table *stit;
 111         unsigned long ua = 0;
 112
 113         /* Allow userspace to poison TCE table */
 114         if (dir == DMA_NONE)
 115                 return H_SUCCESS;
 116
 117         if (iommu_tce_check_gpa(stt->page_shift, gpa))
 118                 return H_PARAMETER;
 119
 120         if (kvmppc_rm_tce_to_ua(stt->kvm, tce, &ua, NULL))
 121                 return H_TOO_HARD;
 122
 123         list_for_each_entry_lockless(stit, &stt->iommu_tables, next) {
 124                 unsigned long hpa = 0;
 125                 struct mm_iommu_table_group_mem_t *mem;
 126                 long shift = stit->tbl->it_page_shift;
 127
 128                 mem = mm_iommu_lookup_rm(stt->kvm->mm, ua, 1ULL << shift);
 129                 if (!mem)
 130                         return H_TOO_HARD;
 131
 132                 if (mm_iommu_ua_to_hpa_rm(mem, ua, shift, &hpa))
 133                         return H_TOO_HARD;
 134         }
 135
 136         return H_SUCCESS;
 137 }
 138
 139 /* Note on the use of page_address() in real mode,
 140  *
 141  * It is safe to use page_address() in real mode on ppc64 because
 142  * page_address() is always defined as lowmem_page_address()
 143  * which returns __va(PFN_PHYS(page_to_pfn(page))) which is arithmetic
 144  * operation and does not access page struct.
 145  *
 146  * Theoretically page_address() could be defined different
 147  * but either WANT_PAGE_VIRTUAL or HASHED_PAGE_VIRTUAL
 148  * would have to be enabled.
 149  * WANT_PAGE_VIRTUAL is never enabled on ppc32/ppc64,
 150  * HASHED_PAGE_VIRTUAL could be enabled for ppc32 only and only
 151  * if CONFIG_HIGHMEM is defined. As CONFIG_SPARSEMEM_VMEMMAP
 152  * is not expected to be enabled on ppc32, page_address()
 153  * is safe for ppc32 as well.
 154  *
 155  * WARNING: This will be called in real-mode on HV KVM and virtual
 156  *          mode on PR KVM
 157  */
 158 static u64 *kvmppc_page_address(struct page *page)
 159 {
 160 #if defined(HASHED_PAGE_VIRTUAL) || defined(WANT_PAGE_VIRTUAL)
 161 #error TODO: fix to avoid page_address() here
 162 #endif
 163         return (u64 *) page_address(page);
 164 }
 165
 166 /*
 167  * Handles TCE requests for emulated devices.
 168  * Puts guest TCE values to the table and expects user space to convert them.
 169  * Cannot fail so kvmppc_rm_tce_validate must be called before it.
 170  */
 171 static void kvmppc_rm_tce_put(struct kvmppc_spapr_tce_table *stt,
 172                 unsigned long idx, unsigned long tce)
 173 {
 174         struct page *page;
 175         u64 *tbl;
 176
 177         idx -= stt->offset;
 178         page = stt->pages[idx / TCES_PER_PAGE];
 179         /*
 180          * page must not be NULL in real mode,
 181          * kvmppc_rm_ioba_validate() must have taken care of this.
 182          */
 183         WARN_ON_ONCE_RM(!page);
 184         tbl = kvmppc_page_address(page);
 185
 186         tbl[idx % TCES_PER_PAGE] = tce;
 187 }
 188
 189 /*
 190  * TCEs pages are allocated in kvmppc_rm_tce_put() which won't be able to do so
 191  * in real mode.
 192  * Check if kvmppc_rm_tce_put() can succeed in real mode, i.e. a TCEs page is
 193  * allocated or not required (when clearing a tce entry).
 194  */
 195 static long kvmppc_rm_ioba_validate(struct kvmppc_spapr_tce_table *stt,
 196                 unsigned long ioba, unsigned long npages, bool clearing)
 197 {
 198         unsigned long i, idx, sttpage, sttpages;
 199         unsigned long ret = kvmppc_ioba_validate(stt, ioba, npages);
 200
 201         if (ret)
 202                 return ret;
 203         /*
 204          * clearing==true says kvmppc_rm_tce_put won't be allocating pages
 205          * for empty tces.
 206          */
 207         if (clearing)
 208                 return H_SUCCESS;
 209
 210         idx = (ioba >> stt->page_shift) - stt->offset;
 211         sttpage = idx / TCES_PER_PAGE;
 212         sttpages = _ALIGN_UP(idx % TCES_PER_PAGE + npages, TCES_PER_PAGE) /
 213                         TCES_PER_PAGE;
 214         for (i = sttpage; i < sttpage + sttpages; ++i)
 215                 if (!stt->pages[i])
 216                         return H_TOO_HARD;
 217
 218         return H_SUCCESS;
 219 }
 220
 221 static long iommu_tce_xchg_no_kill_rm(struct mm_struct *mm,
 222                 struct iommu_table *tbl,
 223                 unsigned long entry, unsigned long *hpa,
 224                 enum dma_data_direction *direction)
 225 {
 226         long ret;
 227
 228         ret = tbl->it_ops->xchg_no_kill(tbl, entry, hpa, direction, true);
 229
 230         if (!ret && ((*direction == DMA_FROM_DEVICE) ||
 231                                 (*direction == DMA_BIDIRECTIONAL))) {
 232                 __be64 *pua = IOMMU_TABLE_USERSPACE_ENTRY_RO(tbl, entry);
 233                 /*
 234                  * kvmppc_rm_tce_iommu_do_map() updates the UA cache after
 235                  * calling this so we still get here a valid UA.
 236                  */
 237                 if (pua && *pua)
 238                         mm_iommu_ua_mark_dirty_rm(mm, be64_to_cpu(*pua));
 239         }
 240
 241         return ret;
 242 }
 243
 244 extern void iommu_tce_kill_rm(struct iommu_table *tbl,
 245                 unsigned long entry, unsigned long pages)
 246 {
 247         if (tbl->it_ops->tce_kill)
 248                 tbl->it_ops->tce_kill(tbl, entry, pages, true);
 249 }
 250
 251 static void kvmppc_rm_clear_tce(struct kvm *kvm, struct iommu_table *tbl,
 252                 unsigned long entry)
 253 {
 254         unsigned long hpa = 0;
 255         enum dma_data_direction dir = DMA_NONE;
 256
 257         iommu_tce_xchg_no_kill_rm(kvm->mm, tbl, entry, &hpa, &dir);
 258 }
 259
 260 static long kvmppc_rm_tce_iommu_mapped_dec(struct kvm *kvm,
 261                 struct iommu_table *tbl, unsigned long entry)
 262 {
 263         struct mm_iommu_table_group_mem_t *mem = NULL;
 264         const unsigned long pgsize = 1ULL << tbl->it_page_shift;
 265         __be64 *pua = IOMMU_TABLE_USERSPACE_ENTRY_RO(tbl, entry);
 266
 267         if (!pua)
 268                 /* it_userspace allocation might be delayed */
 269                 return H_TOO_HARD;
 270
 271         mem = mm_iommu_lookup_rm(kvm->mm, be64_to_cpu(*pua), pgsize);
 272         if (!mem)
 273                 return H_TOO_HARD;
 274
 275         mm_iommu_mapped_dec(mem);
 276
 277         *pua = cpu_to_be64(0);
 278
 279         return H_SUCCESS;
 280 }
 281
 282 static long kvmppc_rm_tce_iommu_do_unmap(struct kvm *kvm,
 283                 struct iommu_table *tbl, unsigned long entry)
 284 {
 285         enum dma_data_direction dir = DMA_NONE;
 286         unsigned long hpa = 0;
 287         long ret;
 288
 289         if (iommu_tce_xchg_no_kill_rm(kvm->mm, tbl, entry, &hpa, &dir))
 290                 /*
 291                  * real mode xchg can fail if struct page crosses
 292                  * a page boundary
 293                  */
 294                 return H_TOO_HARD;
 295
 296         if (dir == DMA_NONE)
 297                 return H_SUCCESS;
 298
 299         ret = kvmppc_rm_tce_iommu_mapped_dec(kvm, tbl, entry);
 300         if (ret)
 301                 iommu_tce_xchg_no_kill_rm(kvm->mm, tbl, entry, &hpa, &dir);
 302
 303         return ret;
 304 }
 305
 306 static long kvmppc_rm_tce_iommu_unmap(struct kvm *kvm,
 307                 struct kvmppc_spapr_tce_table *stt, struct iommu_table *tbl,
 308                 unsigned long entry)
 309 {
 310         unsigned long i, ret = H_SUCCESS;
 311         unsigned long subpages = 1ULL << (stt->page_shift - tbl->it_page_shift);
 312         unsigned long io_entry = entry * subpages;
 313
 314         for (i = 0; i < subpages; ++i) {
 315                 ret = kvmppc_rm_tce_iommu_do_unmap(kvm, tbl, io_entry + i);
 316                 if (ret != H_SUCCESS)
 317                         break;
 318         }
 319
 320         return ret;
 321 }
 322
 323 static long kvmppc_rm_tce_iommu_do_map(struct kvm *kvm, struct iommu_table *tbl,
 324                 unsigned long entry, unsigned long ua,
 325                 enum dma_data_direction dir)
 326 {
 327         long ret;
 328         unsigned long hpa = 0;
 329         __be64 *pua = IOMMU_TABLE_USERSPACE_ENTRY_RO(tbl, entry);
 330         struct mm_iommu_table_group_mem_t *mem;
 331
 332         if (!pua)
 333                 /* it_userspace allocation might be delayed */
 334                 return H_TOO_HARD;
 335
 336         mem = mm_iommu_lookup_rm(kvm->mm, ua, 1ULL << tbl->it_page_shift);
 337         if (!mem)
 338                 return H_TOO_HARD;
 339
 340         if (WARN_ON_ONCE_RM(mm_iommu_ua_to_hpa_rm(mem, ua, tbl->it_page_shift,
 341                         &hpa)))
 342                 return H_TOO_HARD;
 343
 344         if (WARN_ON_ONCE_RM(mm_iommu_mapped_inc(mem)))
 345                 return H_TOO_HARD;
 346
 347         ret = iommu_tce_xchg_no_kill_rm(kvm->mm, tbl, entry, &hpa, &dir);
 348         if (ret) {
 349                 mm_iommu_mapped_dec(mem);
 350                 /*
 351                  * real mode xchg can fail if struct page crosses
 352                  * a page boundary
 353                  */
 354                 return H_TOO_HARD;
 355         }
 356
 357         if (dir != DMA_NONE)
 358                 kvmppc_rm_tce_iommu_mapped_dec(kvm, tbl, entry);
 359
 360         *pua = cpu_to_be64(ua);
 361
 362         return 0;
 363 }
 364
 365 static long kvmppc_rm_tce_iommu_map(struct kvm *kvm,
 366                 struct kvmppc_spapr_tce_table *stt, struct iommu_table *tbl,
 367                 unsigned long entry, unsigned long ua,
 368                 enum dma_data_direction dir)
 369 {
 370         unsigned long i, pgoff, ret = H_SUCCESS;
 371         unsigned long subpages = 1ULL << (stt->page_shift - tbl->it_page_shift);
 372         unsigned long io_entry = entry * subpages;
 373
 374         for (i = 0, pgoff = 0; i < subpages;
 375                         ++i, pgoff += IOMMU_PAGE_SIZE(tbl)) {
 376
 377                 ret = kvmppc_rm_tce_iommu_do_map(kvm, tbl,
 378                                 io_entry + i, ua + pgoff, dir);
 379                 if (ret != H_SUCCESS)
 380                         break;
 381         }
 382
 383         return ret;
 384 }
 385
 386 long kvmppc_rm_h_put_tce(struct kvm_vcpu *vcpu, unsigned long liobn,
 387                 unsigned long ioba, unsigned long tce)
 388 {
 389         struct kvmppc_spapr_tce_table *stt;
 390         long ret;
 391         struct kvmppc_spapr_tce_iommu_table *stit;
 392         unsigned long entry, ua = 0;
 393         enum dma_data_direction dir;
 394
 395         /* udbg_printf("H_PUT_TCE(): liobn=0x%lx ioba=0x%lx, tce=0x%lx\n", */
 396         /*          liobn, ioba, tce); */
 397
 398         /* For radix, we might be in virtual mode, so punt */
 399         if (kvm_is_radix(vcpu->kvm))
 400                 return H_TOO_HARD;
 401
 402         stt = kvmppc_find_table(vcpu->kvm, liobn);
 403         if (!stt)
 404                 return H_TOO_HARD;
 405
 406         ret = kvmppc_rm_ioba_validate(stt, ioba, 1, tce == 0);
 407         if (ret != H_SUCCESS)
 408                 return ret;
 409
 410         ret = kvmppc_rm_tce_validate(stt, tce);
 411         if (ret != H_SUCCESS)
 412                 return ret;
 413
 414         dir = iommu_tce_direction(tce);
 415         if ((dir != DMA_NONE) && kvmppc_rm_tce_to_ua(vcpu->kvm, tce, &ua, NULL))
 416                 return H_PARAMETER;
 417
 418         entry = ioba >> stt->page_shift;
 419
 420         list_for_each_entry_lockless(stit, &stt->iommu_tables, next) {
 421                 if (dir == DMA_NONE)
 422                         ret = kvmppc_rm_tce_iommu_unmap(vcpu->kvm, stt,
 423                                         stit->tbl, entry);
 424                 else
 425                         ret = kvmppc_rm_tce_iommu_map(vcpu->kvm, stt,
 426                                         stit->tbl, entry, ua, dir);
 427
 428                 iommu_tce_kill_rm(stit->tbl, entry, 1);
 429
 430                 if (ret != H_SUCCESS) {
 431                         kvmppc_rm_clear_tce(vcpu->kvm, stit->tbl, entry);
 432                         return ret;
 433                 }
 434         }
 435
 436         kvmppc_rm_tce_put(stt, entry, tce);
 437
 438         return H_SUCCESS;
 439 }
 440
 441 static long kvmppc_rm_ua_to_hpa(struct kvm_vcpu *vcpu,
 442                 unsigned long ua, unsigned long *phpa)
 443 {
 444         pte_t *ptep, pte;
 445         unsigned shift = 0;
 446
 447         /*
 448          * Called in real mode with MSR_EE = 0. We are safe here.
 449          * It is ok to do the lookup with arch.pgdir here, because
 450          * we are doing this on secondary cpus and current task there
 451          * is not the hypervisor. Also this is safe against THP in the
 452          * host, because an IPI to primary thread will wait for the secondary
 453          * to exit which will agains result in the below page table walk
 454          * to finish.
 455          */
 456         ptep = __find_linux_pte(vcpu->arch.pgdir, ua, NULL, &shift);
 457         if (!ptep || !pte_present(*ptep))
 458                 return -ENXIO;
 459         pte = *ptep;
 460
 461         if (!shift)
 462                 shift = PAGE_SHIFT;
 463
 464         /* Avoid handling anything potentially complicated in realmode */
 465         if (shift > PAGE_SHIFT)
 466                 return -EAGAIN;
 467
 468         if (!pte_young(pte))
 469                 return -EAGAIN;
 470
 471         *phpa = (pte_pfn(pte) << PAGE_SHIFT) | (ua & ((1ULL << shift) - 1)) |
 472                         (ua & ~PAGE_MASK);
 473
 474         return 0;
 475 }
 476
 477 long kvmppc_rm_h_put_tce_indirect(struct kvm_vcpu *vcpu,
 478                 unsigned long liobn, unsigned long ioba,
 479                 unsigned long tce_list, unsigned long npages)
 480 {
 481         struct kvmppc_spapr_tce_table *stt;
 482         long i, ret = H_SUCCESS;
 483         unsigned long tces, entry, ua = 0;
 484         unsigned long *rmap = NULL;
 485         bool prereg = false;
 486         struct kvmppc_spapr_tce_iommu_table *stit;
 487
 488         /* For radix, we might be in virtual mode, so punt */
 489         if (kvm_is_radix(vcpu->kvm))
 490                 return H_TOO_HARD;
 491
 492         stt = kvmppc_find_table(vcpu->kvm, liobn);
 493         if (!stt)
 494                 return H_TOO_HARD;
 495
 496         entry = ioba >> stt->page_shift;
 497         /*
 498          * The spec says that the maximum size of the list is 512 TCEs
 499          * so the whole table addressed resides in 4K page
 500          */
 501         if (npages > 512)
 502                 return H_PARAMETER;
 503
 504         if (tce_list & (SZ_4K - 1))
 505                 return H_PARAMETER;
 506
 507         ret = kvmppc_rm_ioba_validate(stt, ioba, npages, false);
 508         if (ret != H_SUCCESS)
 509                 return ret;
 510
 511         if (mm_iommu_preregistered(vcpu->kvm->mm)) {
 512                 /*
 513                  * We get here if guest memory was pre-registered which
 514                  * is normally VFIO case and gpa->hpa translation does not
 515                  * depend on hpt.
 516                  */
 517                 struct mm_iommu_table_group_mem_t *mem;
 518
 519                 if (kvmppc_rm_tce_to_ua(vcpu->kvm, tce_list, &ua, NULL))
 520                         return H_TOO_HARD;
 521
 522                 mem = mm_iommu_lookup_rm(vcpu->kvm->mm, ua, IOMMU_PAGE_SIZE_4K);
 523                 if (mem)
 524                         prereg = mm_iommu_ua_to_hpa_rm(mem, ua,
 525                                         IOMMU_PAGE_SHIFT_4K, &tces) == 0;
 526         }
 527
 528         if (!prereg) {
 529                 /*
 530                  * This is usually a case of a guest with emulated devices only
 531                  * when TCE list is not in preregistered memory.
 532                  * We do not require memory to be preregistered in this case
 533                  * so lock rmap and do __find_linux_pte_or_hugepte().
 534                  */
 535                 if (kvmppc_rm_tce_to_ua(vcpu->kvm, tce_list, &ua, &rmap))
 536                         return H_TOO_HARD;
 537
 538                 rmap = (void *) vmalloc_to_phys(rmap);
 539                 if (WARN_ON_ONCE_RM(!rmap))
 540                         return H_TOO_HARD;
 541
 542                 /*
 543                  * Synchronize with the MMU notifier callbacks in
 544                  * book3s_64_mmu_hv.c (kvm_unmap_hva_range_hv etc.).
 545                  * While we have the rmap lock, code running on other CPUs
 546                  * cannot finish unmapping the host real page that backs
 547                  * this guest real page, so we are OK to access the host
 548                  * real page.
 549                  */
 550                 lock_rmap(rmap);
 551                 if (kvmppc_rm_ua_to_hpa(vcpu, ua, &tces)) {
 552                         ret = H_TOO_HARD;
 553                         goto unlock_exit;
 554                 }
 555         }
 556
 557         for (i = 0; i < npages; ++i) {
 558                 unsigned long tce = be64_to_cpu(((u64 *)tces)[i]);
 559
 560                 ret = kvmppc_rm_tce_validate(stt, tce);
 561                 if (ret != H_SUCCESS)
 562                         goto unlock_exit;
 563         }
 564
 565         for (i = 0; i < npages; ++i) {
 566                 unsigned long tce = be64_to_cpu(((u64 *)tces)[i]);
 567
 568                 ua = 0;
 569                 if (kvmppc_rm_tce_to_ua(vcpu->kvm, tce, &ua, NULL)) {
 570                         ret = H_PARAMETER;
 571                         goto invalidate_exit;
 572                 }
 573
 574                 list_for_each_entry_lockless(stit, &stt->iommu_tables, next) {
 575                         ret = kvmppc_rm_tce_iommu_map(vcpu->kvm, stt,
 576                                         stit->tbl, entry + i, ua,
 577                                         iommu_tce_direction(tce));
 578
 579                         if (ret != H_SUCCESS) {
 580                                 kvmppc_rm_clear_tce(vcpu->kvm, stit->tbl,
 581                                                 entry);
 582                                 goto invalidate_exit;
 583                         }
 584                 }
 585
 586                 kvmppc_rm_tce_put(stt, entry + i, tce);
 587         }
 588
 589 invalidate_exit:
 590         list_for_each_entry_lockless(stit, &stt->iommu_tables, next)
 591                 iommu_tce_kill_rm(stit->tbl, entry, npages);
 592
 593 unlock_exit:
 594         if (rmap)
 595                 unlock_rmap(rmap);
 596
 597         return ret;
 598 }
 599
 600 long kvmppc_rm_h_stuff_tce(struct kvm_vcpu *vcpu,
 601                 unsigned long liobn, unsigned long ioba,
 602                 unsigned long tce_value, unsigned long npages)
 603 {
 604         struct kvmppc_spapr_tce_table *stt;
 605         long i, ret;
 606         struct kvmppc_spapr_tce_iommu_table *stit;
 607
 608         /* For radix, we might be in virtual mode, so punt */
 609         if (kvm_is_radix(vcpu->kvm))
 610                 return H_TOO_HARD;
 611
 612         stt = kvmppc_find_table(vcpu->kvm, liobn);
 613         if (!stt)
 614                 return H_TOO_HARD;
 615
 616         ret = kvmppc_rm_ioba_validate(stt, ioba, npages, tce_value == 0);
 617         if (ret != H_SUCCESS)
 618                 return ret;
 619
 620         /* Check permission bits only to allow userspace poison TCE for debug */
 621         if (tce_value & (TCE_PCI_WRITE | TCE_PCI_READ))
 622                 return H_PARAMETER;
 623
 624         list_for_each_entry_lockless(stit, &stt->iommu_tables, next) {
 625                 unsigned long entry = ioba >> stt->page_shift;
 626
 627                 for (i = 0; i < npages; ++i) {
 628                         ret = kvmppc_rm_tce_iommu_unmap(vcpu->kvm, stt,
 629                                         stit->tbl, entry + i);
 630
 631                         if (ret == H_SUCCESS)
 632                                 continue;
 633
 634                         if (ret == H_TOO_HARD)
 635                                 goto invalidate_exit;
 636
 637                         WARN_ON_ONCE_RM(1);
 638                         kvmppc_rm_clear_tce(vcpu->kvm, stit->tbl, entry);
 639                 }
 640         }
 641
 642         for (i = 0; i < npages; ++i, ioba += (1ULL << stt->page_shift))
 643                 kvmppc_rm_tce_put(stt, ioba >> stt->page_shift, tce_value);
 644
 645 invalidate_exit:
 646         list_for_each_entry_lockless(stit, &stt->iommu_tables, next)
 647                 iommu_tce_kill_rm(stit->tbl, ioba >> stt->page_shift, npages);
 648
 649         return ret;
 650 }
 651
 652 /* This can be called in either virtual mode or real mode */
 653 long kvmppc_h_get_tce(struct kvm_vcpu *vcpu, unsigned long liobn,
 654                       unsigned long ioba)
 655 {
 656         struct kvmppc_spapr_tce_table *stt;
 657         long ret;
 658         unsigned long idx;
 659         struct page *page;
 660         u64 *tbl;
 661
 662         stt = kvmppc_find_table(vcpu->kvm, liobn);
 663         if (!stt)
 664                 return H_TOO_HARD;
 665
 666         ret = kvmppc_ioba_validate(stt, ioba, 1);
 667         if (ret != H_SUCCESS)
 668                 return ret;
 669
 670         idx = (ioba >> stt->page_shift) - stt->offset;
 671         page = stt->pages[idx / TCES_PER_PAGE];
 672         if (!page) {
 673                 vcpu->arch.regs.gpr[4] = 0;
 674                 return H_SUCCESS;
 675         }
 676         tbl = (u64 *)page_address(page);
 677
 678         vcpu->arch.regs.gpr[4] = tbl[idx % TCES_PER_PAGE];
 679
 680         return H_SUCCESS;
 681 }
 682 EXPORT_SYMBOL_GPL(kvmppc_h_get_tce);
 683
 684 #endif /* KVM_BOOK3S_HV_POSSIBLE */