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 IO address.
  91  *
  92  * WARNING: This will be called in real-mode on HV KVM and virtual
  93  *          mode on PR KVM
  94  */
  95 long kvmppc_ioba_validate(struct kvmppc_spapr_tce_table *stt,
  96                 unsigned long ioba, unsigned long npages)
  97 {
  98         unsigned long mask = (1ULL << stt->page_shift) - 1;
  99         unsigned long idx = ioba >> stt->page_shift;
 100
 101         if ((ioba & mask) || (idx < stt->offset) ||
 102                         (idx - stt->offset + npages > stt->size) ||
 103                         (idx + npages < idx))
 104                 return H_PARAMETER;
 105
 106         return H_SUCCESS;
 107 }
 108 EXPORT_SYMBOL_GPL(kvmppc_ioba_validate);
 109
 110 /*
 111  * Validates TCE address.
 112  * At the moment flags and page mask are validated.
 113  * As the host kernel does not access those addresses (just puts them
 114  * to the table and user space is supposed to process them), we can skip
 115  * checking other things (such as TCE is a guest RAM address or the page
 116  * was actually allocated).
 117  *
 118  * WARNING: This will be called in real-mode on HV KVM and virtual
 119  *          mode on PR KVM
 120  */
 121 long kvmppc_tce_validate(struct kvmppc_spapr_tce_table *stt, unsigned long tce)
 122 {
 123         unsigned long page_mask = ~((1ULL << stt->page_shift) - 1);
 124         unsigned long mask = ~(page_mask | TCE_PCI_WRITE | TCE_PCI_READ);
 125
 126         if (tce & mask)
 127                 return H_PARAMETER;
 128
 129         return H_SUCCESS;
 130 }
 131 EXPORT_SYMBOL_GPL(kvmppc_tce_validate);
 132
 133 /* Note on the use of page_address() in real mode,
 134  *
 135  * It is safe to use page_address() in real mode on ppc64 because
 136  * page_address() is always defined as lowmem_page_address()
 137  * which returns __va(PFN_PHYS(page_to_pfn(page))) which is arithmetic
 138  * operation and does not access page struct.
 139  *
 140  * Theoretically page_address() could be defined different
 141  * but either WANT_PAGE_VIRTUAL or HASHED_PAGE_VIRTUAL
 142  * would have to be enabled.
 143  * WANT_PAGE_VIRTUAL is never enabled on ppc32/ppc64,
 144  * HASHED_PAGE_VIRTUAL could be enabled for ppc32 only and only
 145  * if CONFIG_HIGHMEM is defined. As CONFIG_SPARSEMEM_VMEMMAP
 146  * is not expected to be enabled on ppc32, page_address()
 147  * is safe for ppc32 as well.
 148  *
 149  * WARNING: This will be called in real-mode on HV KVM and virtual
 150  *          mode on PR KVM
 151  */
 152 static u64 *kvmppc_page_address(struct page *page)
 153 {
 154 #if defined(HASHED_PAGE_VIRTUAL) || defined(WANT_PAGE_VIRTUAL)
 155 #error TODO: fix to avoid page_address() here
 156 #endif
 157         return (u64 *) page_address(page);
 158 }
 159
 160 /*
 161  * Handles TCE requests for emulated devices.
 162  * Puts guest TCE values to the table and expects user space to convert them.
 163  * Called in both real and virtual modes.
 164  * Cannot fail so kvmppc_tce_validate must be called before it.
 165  *
 166  * WARNING: This will be called in real-mode on HV KVM and virtual
 167  *          mode on PR KVM
 168  */
 169 void kvmppc_tce_put(struct kvmppc_spapr_tce_table *stt,
 170                 unsigned long idx, unsigned long tce)
 171 {
 172         struct page *page;
 173         u64 *tbl;
 174
 175         idx -= stt->offset;
 176         page = stt->pages[idx / TCES_PER_PAGE];
 177         tbl = kvmppc_page_address(page);
 178
 179         tbl[idx % TCES_PER_PAGE] = tce;
 180 }
 181 EXPORT_SYMBOL_GPL(kvmppc_tce_put);
 182
 183 long kvmppc_gpa_to_ua(struct kvm *kvm, unsigned long gpa,
 184                 unsigned long *ua, unsigned long **prmap)
 185 {
 186         unsigned long gfn = gpa >> PAGE_SHIFT;
 187         struct kvm_memory_slot *memslot;
 188
 189         memslot = search_memslots(kvm_memslots(kvm), gfn);
 190         if (!memslot)
 191                 return -EINVAL;
 192
 193         *ua = __gfn_to_hva_memslot(memslot, gfn) |
 194                 (gpa & ~(PAGE_MASK | TCE_PCI_READ | TCE_PCI_WRITE));
 195
 196 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
 197         if (prmap)
 198                 *prmap = &memslot->arch.rmap[gfn - memslot->base_gfn];
 199 #endif
 200
 201         return 0;
 202 }
 203 EXPORT_SYMBOL_GPL(kvmppc_gpa_to_ua);
 204
 205 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
 206 static void kvmppc_rm_clear_tce(struct iommu_table *tbl, unsigned long entry)
 207 {
 208         unsigned long hpa = 0;
 209         enum dma_data_direction dir = DMA_NONE;
 210
 211         iommu_tce_xchg_rm(tbl, entry, &hpa, &dir);
 212 }
 213
 214 static long kvmppc_rm_tce_iommu_mapped_dec(struct kvm *kvm,
 215                 struct iommu_table *tbl, unsigned long entry)
 216 {
 217         struct mm_iommu_table_group_mem_t *mem = NULL;
 218         const unsigned long pgsize = 1ULL << tbl->it_page_shift;
 219         unsigned long *pua = IOMMU_TABLE_USERSPACE_ENTRY(tbl, entry);
 220
 221         if (!pua)
 222                 /* it_userspace allocation might be delayed */
 223                 return H_TOO_HARD;
 224
 225         pua = (void *) vmalloc_to_phys(pua);
 226         if (WARN_ON_ONCE_RM(!pua))
 227                 return H_HARDWARE;
 228
 229         mem = mm_iommu_lookup_rm(kvm->mm, *pua, pgsize);
 230         if (!mem)
 231                 return H_TOO_HARD;
 232
 233         mm_iommu_mapped_dec(mem);
 234
 235         *pua = 0;
 236
 237         return H_SUCCESS;
 238 }
 239
 240 static long kvmppc_rm_tce_iommu_unmap(struct kvm *kvm,
 241                 struct iommu_table *tbl, unsigned long entry)
 242 {
 243         enum dma_data_direction dir = DMA_NONE;
 244         unsigned long hpa = 0;
 245         long ret;
 246
 247         if (iommu_tce_xchg_rm(tbl, entry, &hpa, &dir))
 248                 /*
 249                  * real mode xchg can fail if struct page crosses
 250                  * a page boundary
 251                  */
 252                 return H_TOO_HARD;
 253
 254         if (dir == DMA_NONE)
 255                 return H_SUCCESS;
 256
 257         ret = kvmppc_rm_tce_iommu_mapped_dec(kvm, tbl, entry);
 258         if (ret)
 259                 iommu_tce_xchg_rm(tbl, entry, &hpa, &dir);
 260
 261         return ret;
 262 }
 263
 264 static long kvmppc_rm_tce_iommu_map(struct kvm *kvm, struct iommu_table *tbl,
 265                 unsigned long entry, unsigned long ua,
 266                 enum dma_data_direction dir)
 267 {
 268         long ret;
 269         unsigned long hpa = 0;
 270         unsigned long *pua = IOMMU_TABLE_USERSPACE_ENTRY(tbl, entry);
 271         struct mm_iommu_table_group_mem_t *mem;
 272
 273         if (!pua)
 274                 /* it_userspace allocation might be delayed */
 275                 return H_TOO_HARD;
 276
 277         mem = mm_iommu_lookup_rm(kvm->mm, ua, 1ULL << tbl->it_page_shift);
 278         if (!mem)
 279                 return H_TOO_HARD;
 280
 281         if (WARN_ON_ONCE_RM(mm_iommu_ua_to_hpa_rm(mem, ua, &hpa)))
 282                 return H_HARDWARE;
 283
 284         pua = (void *) vmalloc_to_phys(pua);
 285         if (WARN_ON_ONCE_RM(!pua))
 286                 return H_HARDWARE;
 287
 288         if (WARN_ON_ONCE_RM(mm_iommu_mapped_inc(mem)))
 289                 return H_CLOSED;
 290
 291         ret = iommu_tce_xchg_rm(tbl, entry, &hpa, &dir);
 292         if (ret) {
 293                 mm_iommu_mapped_dec(mem);
 294                 /*
 295                  * real mode xchg can fail if struct page crosses
 296                  * a page boundary
 297                  */
 298                 return H_TOO_HARD;
 299         }
 300
 301         if (dir != DMA_NONE)
 302                 kvmppc_rm_tce_iommu_mapped_dec(kvm, tbl, entry);
 303
 304         *pua = ua;
 305
 306         return 0;
 307 }
 308
 309 long kvmppc_rm_h_put_tce(struct kvm_vcpu *vcpu, unsigned long liobn,
 310                 unsigned long ioba, unsigned long tce)
 311 {
 312         struct kvmppc_spapr_tce_table *stt;
 313         long ret;
 314         struct kvmppc_spapr_tce_iommu_table *stit;
 315         unsigned long entry, ua = 0;
 316         enum dma_data_direction dir;
 317
 318         /* udbg_printf("H_PUT_TCE(): liobn=0x%lx ioba=0x%lx, tce=0x%lx\n", */
 319         /*          liobn, ioba, tce); */
 320
 321         /* For radix, we might be in virtual mode, so punt */
 322         if (kvm_is_radix(vcpu->kvm))
 323                 return H_TOO_HARD;
 324
 325         stt = kvmppc_find_table(vcpu->kvm, liobn);
 326         if (!stt)
 327                 return H_TOO_HARD;
 328
 329         ret = kvmppc_ioba_validate(stt, ioba, 1);
 330         if (ret != H_SUCCESS)
 331                 return ret;
 332
 333         ret = kvmppc_tce_validate(stt, tce);
 334         if (ret != H_SUCCESS)
 335                 return ret;
 336
 337         dir = iommu_tce_direction(tce);
 338         if ((dir != DMA_NONE) && kvmppc_gpa_to_ua(vcpu->kvm,
 339                         tce & ~(TCE_PCI_READ | TCE_PCI_WRITE), &ua, NULL))
 340                 return H_PARAMETER;
 341
 342         entry = ioba >> stt->page_shift;
 343
 344         list_for_each_entry_lockless(stit, &stt->iommu_tables, next) {
 345                 if (dir == DMA_NONE)
 346                         ret = kvmppc_rm_tce_iommu_unmap(vcpu->kvm,
 347                                         stit->tbl, entry);
 348                 else
 349                         ret = kvmppc_rm_tce_iommu_map(vcpu->kvm,
 350                                         stit->tbl, entry, ua, dir);
 351
 352                 if (ret == H_SUCCESS)
 353                         continue;
 354
 355                 if (ret == H_TOO_HARD)
 356                         return ret;
 357
 358                 WARN_ON_ONCE_RM(1);
 359                 kvmppc_rm_clear_tce(stit->tbl, entry);
 360         }
 361
 362         kvmppc_tce_put(stt, entry, tce);
 363
 364         return H_SUCCESS;
 365 }
 366
 367 static long kvmppc_rm_ua_to_hpa(struct kvm_vcpu *vcpu,
 368                 unsigned long ua, unsigned long *phpa)
 369 {
 370         pte_t *ptep, pte;
 371         unsigned shift = 0;
 372
 373         ptep = __find_linux_pte_or_hugepte(vcpu->arch.pgdir, ua, NULL, &shift);
 374         if (!ptep || !pte_present(*ptep))
 375                 return -ENXIO;
 376         pte = *ptep;
 377
 378         if (!shift)
 379                 shift = PAGE_SHIFT;
 380
 381         /* Avoid handling anything potentially complicated in realmode */
 382         if (shift > PAGE_SHIFT)
 383                 return -EAGAIN;
 384
 385         if (!pte_young(pte))
 386                 return -EAGAIN;
 387
 388         *phpa = (pte_pfn(pte) << PAGE_SHIFT) | (ua & ((1ULL << shift) - 1)) |
 389                         (ua & ~PAGE_MASK);
 390
 391         return 0;
 392 }
 393
 394 long kvmppc_rm_h_put_tce_indirect(struct kvm_vcpu *vcpu,
 395                 unsigned long liobn, unsigned long ioba,
 396                 unsigned long tce_list, unsigned long npages)
 397 {
 398         struct kvmppc_spapr_tce_table *stt;
 399         long i, ret = H_SUCCESS;
 400         unsigned long tces, entry, ua = 0;
 401         unsigned long *rmap = NULL;
 402         bool prereg = false;
 403         struct kvmppc_spapr_tce_iommu_table *stit;
 404
 405         /* For radix, we might be in virtual mode, so punt */
 406         if (kvm_is_radix(vcpu->kvm))
 407                 return H_TOO_HARD;
 408
 409         stt = kvmppc_find_table(vcpu->kvm, liobn);
 410         if (!stt)
 411                 return H_TOO_HARD;
 412
 413         entry = ioba >> stt->page_shift;
 414         /*
 415          * The spec says that the maximum size of the list is 512 TCEs
 416          * so the whole table addressed resides in 4K page
 417          */
 418         if (npages > 512)
 419                 return H_PARAMETER;
 420
 421         if (tce_list & (SZ_4K - 1))
 422                 return H_PARAMETER;
 423
 424         ret = kvmppc_ioba_validate(stt, ioba, npages);
 425         if (ret != H_SUCCESS)
 426                 return ret;
 427
 428         if (mm_iommu_preregistered(vcpu->kvm->mm)) {
 429                 /*
 430                  * We get here if guest memory was pre-registered which
 431                  * is normally VFIO case and gpa->hpa translation does not
 432                  * depend on hpt.
 433                  */
 434                 struct mm_iommu_table_group_mem_t *mem;
 435
 436                 if (kvmppc_gpa_to_ua(vcpu->kvm, tce_list, &ua, NULL))
 437                         return H_TOO_HARD;
 438
 439                 mem = mm_iommu_lookup_rm(vcpu->kvm->mm, ua, IOMMU_PAGE_SIZE_4K);
 440                 if (mem)
 441                         prereg = mm_iommu_ua_to_hpa_rm(mem, ua, &tces) == 0;
 442         }
 443
 444         if (!prereg) {
 445                 /*
 446                  * This is usually a case of a guest with emulated devices only
 447                  * when TCE list is not in preregistered memory.
 448                  * We do not require memory to be preregistered in this case
 449                  * so lock rmap and do __find_linux_pte_or_hugepte().
 450                  */
 451                 if (kvmppc_gpa_to_ua(vcpu->kvm, tce_list, &ua, &rmap))
 452                         return H_TOO_HARD;
 453
 454                 rmap = (void *) vmalloc_to_phys(rmap);
 455                 if (WARN_ON_ONCE_RM(!rmap))
 456                         return H_HARDWARE;
 457
 458                 /*
 459                  * Synchronize with the MMU notifier callbacks in
 460                  * book3s_64_mmu_hv.c (kvm_unmap_hva_hv etc.).
 461                  * While we have the rmap lock, code running on other CPUs
 462                  * cannot finish unmapping the host real page that backs
 463                  * this guest real page, so we are OK to access the host
 464                  * real page.
 465                  */
 466                 lock_rmap(rmap);
 467                 if (kvmppc_rm_ua_to_hpa(vcpu, ua, &tces)) {
 468                         ret = H_TOO_HARD;
 469                         goto unlock_exit;
 470                 }
 471         }
 472
 473         for (i = 0; i < npages; ++i) {
 474                 unsigned long tce = be64_to_cpu(((u64 *)tces)[i]);
 475
 476                 ret = kvmppc_tce_validate(stt, tce);
 477                 if (ret != H_SUCCESS)
 478                         goto unlock_exit;
 479
 480                 ua = 0;
 481                 if (kvmppc_gpa_to_ua(vcpu->kvm,
 482                                 tce & ~(TCE_PCI_READ | TCE_PCI_WRITE),
 483                                 &ua, NULL))
 484                         return H_PARAMETER;
 485
 486                 list_for_each_entry_lockless(stit, &stt->iommu_tables, next) {
 487                         ret = kvmppc_rm_tce_iommu_map(vcpu->kvm,
 488                                         stit->tbl, entry + i, ua,
 489                                         iommu_tce_direction(tce));
 490
 491                         if (ret == H_SUCCESS)
 492                                 continue;
 493
 494                         if (ret == H_TOO_HARD)
 495                                 goto unlock_exit;
 496
 497                         WARN_ON_ONCE_RM(1);
 498                         kvmppc_rm_clear_tce(stit->tbl, entry);
 499                 }
 500
 501                 kvmppc_tce_put(stt, entry + i, tce);
 502         }
 503
 504 unlock_exit:
 505         if (rmap)
 506                 unlock_rmap(rmap);
 507
 508         return ret;
 509 }
 510
 511 long kvmppc_rm_h_stuff_tce(struct kvm_vcpu *vcpu,
 512                 unsigned long liobn, unsigned long ioba,
 513                 unsigned long tce_value, unsigned long npages)
 514 {
 515         struct kvmppc_spapr_tce_table *stt;
 516         long i, ret;
 517         struct kvmppc_spapr_tce_iommu_table *stit;
 518
 519         /* For radix, we might be in virtual mode, so punt */
 520         if (kvm_is_radix(vcpu->kvm))
 521                 return H_TOO_HARD;
 522
 523         stt = kvmppc_find_table(vcpu->kvm, liobn);
 524         if (!stt)
 525                 return H_TOO_HARD;
 526
 527         ret = kvmppc_ioba_validate(stt, ioba, npages);
 528         if (ret != H_SUCCESS)
 529                 return ret;
 530
 531         /* Check permission bits only to allow userspace poison TCE for debug */
 532         if (tce_value & (TCE_PCI_WRITE | TCE_PCI_READ))
 533                 return H_PARAMETER;
 534
 535         list_for_each_entry_lockless(stit, &stt->iommu_tables, next) {
 536                 unsigned long entry = ioba >> stit->tbl->it_page_shift;
 537
 538                 for (i = 0; i < npages; ++i) {
 539                         ret = kvmppc_rm_tce_iommu_unmap(vcpu->kvm,
 540                                         stit->tbl, entry + i);
 541
 542                         if (ret == H_SUCCESS)
 543                                 continue;
 544
 545                         if (ret == H_TOO_HARD)
 546                                 return ret;
 547
 548                         WARN_ON_ONCE_RM(1);
 549                         kvmppc_rm_clear_tce(stit->tbl, entry);
 550                 }
 551         }
 552
 553         for (i = 0; i < npages; ++i, ioba += (1ULL << stt->page_shift))
 554                 kvmppc_tce_put(stt, ioba >> stt->page_shift, tce_value);
 555
 556         return H_SUCCESS;
 557 }
 558
 559 /* This can be called in either virtual mode or real mode */
 560 long kvmppc_h_get_tce(struct kvm_vcpu *vcpu, unsigned long liobn,
 561                       unsigned long ioba)
 562 {
 563         struct kvmppc_spapr_tce_table *stt;
 564         long ret;
 565         unsigned long idx;
 566         struct page *page;
 567         u64 *tbl;
 568
 569         stt = kvmppc_find_table(vcpu->kvm, liobn);
 570         if (!stt)
 571                 return H_TOO_HARD;
 572
 573         ret = kvmppc_ioba_validate(stt, ioba, 1);
 574         if (ret != H_SUCCESS)
 575                 return ret;
 576
 577         idx = (ioba >> stt->page_shift) - stt->offset;
 578         page = stt->pages[idx / TCES_PER_PAGE];
 579         tbl = (u64 *)page_address(page);
 580
 581         vcpu->arch.gpr[4] = tbl[idx % TCES_PER_PAGE];
 582
 583         return H_SUCCESS;
 584 }
 585 EXPORT_SYMBOL_GPL(kvmppc_h_get_tce);
 586
 587 #endif /* KVM_BOOK3S_HV_POSSIBLE */