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