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