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 */ 17 18 #include <linux/types.h> 19 #include <linux/string.h> 20 #include <linux/kvm.h> 21 #include <linux/kvm_host.h> 22 #include <linux/highmem.h> 23 #include <linux/gfp.h> 24 #include <linux/slab.h> 25 #include <linux/hugetlb.h> 26 #include <linux/vmalloc.h> 27 #include <linux/srcu.h> 28 #include <linux/anon_inodes.h> 29 #include <linux/file.h> 30 31 #include <asm/tlbflush.h> 32 #include <asm/kvm_ppc.h> 33 #include <asm/kvm_book3s.h> 34 #include <asm/mmu-hash64.h> 35 #include <asm/hvcall.h> 36 #include <asm/synch.h> 37 #include <asm/ppc-opcode.h> 38 #include <asm/cputable.h> 39 40 #include "book3s_hv_cma.h" 41 42 /* POWER7 has 10-bit LPIDs, PPC970 has 6-bit LPIDs */ 43 #define MAX_LPID_970 63 44 45 /* Power architecture requires HPT is at least 256kB */ 46 #define PPC_MIN_HPT_ORDER 18 47 48 static long kvmppc_virtmode_do_h_enter(struct kvm *kvm, unsigned long flags, 49 long pte_index, unsigned long pteh, 50 unsigned long ptel, unsigned long *pte_idx_ret); 51 static void kvmppc_rmap_reset(struct kvm *kvm); 52 53 long kvmppc_alloc_hpt(struct kvm *kvm, u32 *htab_orderp) 54 { 55 unsigned long hpt; 56 struct revmap_entry *rev; 57 struct page *page = NULL; 58 long order = KVM_DEFAULT_HPT_ORDER; 59 60 if (htab_orderp) { 61 order = *htab_orderp; 62 if (order < PPC_MIN_HPT_ORDER) 63 order = PPC_MIN_HPT_ORDER; 64 } 65 66 kvm->arch.hpt_cma_alloc = 0; 67 /* 68 * try first to allocate it from the kernel page allocator. 69 * We keep the CMA reserved for failed allocation. 70 */ 71 hpt = __get_free_pages(GFP_KERNEL | __GFP_ZERO | __GFP_REPEAT | 72 __GFP_NOWARN, order - PAGE_SHIFT); 73 74 /* Next try to allocate from the preallocated pool */ 75 if (!hpt) { 76 VM_BUG_ON(order < KVM_CMA_CHUNK_ORDER); 77 page = kvm_alloc_hpt(1 << (order - PAGE_SHIFT)); 78 if (page) { 79 hpt = (unsigned long)pfn_to_kaddr(page_to_pfn(page)); 80 kvm->arch.hpt_cma_alloc = 1; 81 } else 82 --order; 83 } 84 85 /* Lastly try successively smaller sizes from the page allocator */ 86 while (!hpt && order > PPC_MIN_HPT_ORDER) { 87 hpt = __get_free_pages(GFP_KERNEL|__GFP_ZERO|__GFP_REPEAT| 88 __GFP_NOWARN, order - PAGE_SHIFT); 89 if (!hpt) 90 --order; 91 } 92 93 if (!hpt) 94 return -ENOMEM; 95 96 kvm->arch.hpt_virt = hpt; 97 kvm->arch.hpt_order = order; 98 /* HPTEs are 2**4 bytes long */ 99 kvm->arch.hpt_npte = 1ul << (order - 4); 100 /* 128 (2**7) bytes in each HPTEG */ 101 kvm->arch.hpt_mask = (1ul << (order - 7)) - 1; 102 103 /* Allocate reverse map array */ 104 rev = vmalloc(sizeof(struct revmap_entry) * kvm->arch.hpt_npte); 105 if (!rev) { 106 pr_err("kvmppc_alloc_hpt: Couldn't alloc reverse map array\n"); 107 goto out_freehpt; 108 } 109 kvm->arch.revmap = rev; 110 kvm->arch.sdr1 = __pa(hpt) | (order - 18); 111 112 pr_info("KVM guest htab at %lx (order %ld), LPID %x\n", 113 hpt, order, kvm->arch.lpid); 114 115 if (htab_orderp) 116 *htab_orderp = order; 117 return 0; 118 119 out_freehpt: 120 if (kvm->arch.hpt_cma_alloc) 121 kvm_release_hpt(page, 1 << (order - PAGE_SHIFT)); 122 else 123 free_pages(hpt, order - PAGE_SHIFT); 124 return -ENOMEM; 125 } 126 127 long kvmppc_alloc_reset_hpt(struct kvm *kvm, u32 *htab_orderp) 128 { 129 long err = -EBUSY; 130 long order; 131 132 mutex_lock(&kvm->lock); 133 if (kvm->arch.rma_setup_done) { 134 kvm->arch.rma_setup_done = 0; 135 /* order rma_setup_done vs. vcpus_running */ 136 smp_mb(); 137 if (atomic_read(&kvm->arch.vcpus_running)) { 138 kvm->arch.rma_setup_done = 1; 139 goto out; 140 } 141 } 142 if (kvm->arch.hpt_virt) { 143 order = kvm->arch.hpt_order; 144 /* Set the entire HPT to 0, i.e. invalid HPTEs */ 145 memset((void *)kvm->arch.hpt_virt, 0, 1ul << order); 146 /* 147 * Reset all the reverse-mapping chains for all memslots 148 */ 149 kvmppc_rmap_reset(kvm); 150 /* Ensure that each vcpu will flush its TLB on next entry. */ 151 cpumask_setall(&kvm->arch.need_tlb_flush); 152 *htab_orderp = order; 153 err = 0; 154 } else { 155 err = kvmppc_alloc_hpt(kvm, htab_orderp); 156 order = *htab_orderp; 157 } 158 out: 159 mutex_unlock(&kvm->lock); 160 return err; 161 } 162 163 void kvmppc_free_hpt(struct kvm *kvm) 164 { 165 kvmppc_free_lpid(kvm->arch.lpid); 166 vfree(kvm->arch.revmap); 167 if (kvm->arch.hpt_cma_alloc) 168 kvm_release_hpt(virt_to_page(kvm->arch.hpt_virt), 169 1 << (kvm->arch.hpt_order - PAGE_SHIFT)); 170 else 171 free_pages(kvm->arch.hpt_virt, 172 kvm->arch.hpt_order - PAGE_SHIFT); 173 } 174 175 /* Bits in first HPTE dword for pagesize 4k, 64k or 16M */ 176 static inline unsigned long hpte0_pgsize_encoding(unsigned long pgsize) 177 { 178 return (pgsize > 0x1000) ? HPTE_V_LARGE : 0; 179 } 180 181 /* Bits in second HPTE dword for pagesize 4k, 64k or 16M */ 182 static inline unsigned long hpte1_pgsize_encoding(unsigned long pgsize) 183 { 184 return (pgsize == 0x10000) ? 0x1000 : 0; 185 } 186 187 void kvmppc_map_vrma(struct kvm_vcpu *vcpu, struct kvm_memory_slot *memslot, 188 unsigned long porder) 189 { 190 unsigned long i; 191 unsigned long npages; 192 unsigned long hp_v, hp_r; 193 unsigned long addr, hash; 194 unsigned long psize; 195 unsigned long hp0, hp1; 196 unsigned long idx_ret; 197 long ret; 198 struct kvm *kvm = vcpu->kvm; 199 200 psize = 1ul << porder; 201 npages = memslot->npages >> (porder - PAGE_SHIFT); 202 203 /* VRMA can't be > 1TB */ 204 if (npages > 1ul << (40 - porder)) 205 npages = 1ul << (40 - porder); 206 /* Can't use more than 1 HPTE per HPTEG */ 207 if (npages > kvm->arch.hpt_mask + 1) 208 npages = kvm->arch.hpt_mask + 1; 209 210 hp0 = HPTE_V_1TB_SEG | (VRMA_VSID << (40 - 16)) | 211 HPTE_V_BOLTED | hpte0_pgsize_encoding(psize); 212 hp1 = hpte1_pgsize_encoding(psize) | 213 HPTE_R_R | HPTE_R_C | HPTE_R_M | PP_RWXX; 214 215 for (i = 0; i < npages; ++i) { 216 addr = i << porder; 217 /* can't use hpt_hash since va > 64 bits */ 218 hash = (i ^ (VRMA_VSID ^ (VRMA_VSID << 25))) & kvm->arch.hpt_mask; 219 /* 220 * We assume that the hash table is empty and no 221 * vcpus are using it at this stage. Since we create 222 * at most one HPTE per HPTEG, we just assume entry 7 223 * is available and use it. 224 */ 225 hash = (hash << 3) + 7; 226 hp_v = hp0 | ((addr >> 16) & ~0x7fUL); 227 hp_r = hp1 | addr; 228 ret = kvmppc_virtmode_do_h_enter(kvm, H_EXACT, hash, hp_v, hp_r, 229 &idx_ret); 230 if (ret != H_SUCCESS) { 231 pr_err("KVM: map_vrma at %lx failed, ret=%ld\n", 232 addr, ret); 233 break; 234 } 235 } 236 } 237 238 int kvmppc_mmu_hv_init(void) 239 { 240 unsigned long host_lpid, rsvd_lpid; 241 242 if (!cpu_has_feature(CPU_FTR_HVMODE)) 243 return -EINVAL; 244 245 /* POWER7 has 10-bit LPIDs, PPC970 and e500mc have 6-bit LPIDs */ 246 if (cpu_has_feature(CPU_FTR_ARCH_206)) { 247 host_lpid = mfspr(SPRN_LPID); /* POWER7 */ 248 rsvd_lpid = LPID_RSVD; 249 } else { 250 host_lpid = 0; /* PPC970 */ 251 rsvd_lpid = MAX_LPID_970; 252 } 253 254 kvmppc_init_lpid(rsvd_lpid + 1); 255 256 kvmppc_claim_lpid(host_lpid); 257 /* rsvd_lpid is reserved for use in partition switching */ 258 kvmppc_claim_lpid(rsvd_lpid); 259 260 return 0; 261 } 262 263 static void kvmppc_mmu_book3s_64_hv_reset_msr(struct kvm_vcpu *vcpu) 264 { 265 unsigned long msr = vcpu->arch.intr_msr; 266 267 /* If transactional, change to suspend mode on IRQ delivery */ 268 if (MSR_TM_TRANSACTIONAL(vcpu->arch.shregs.msr)) 269 msr |= MSR_TS_S; 270 else 271 msr |= vcpu->arch.shregs.msr & MSR_TS_MASK; 272 kvmppc_set_msr(vcpu, msr); 273 } 274 275 /* 276 * This is called to get a reference to a guest page if there isn't 277 * one already in the memslot->arch.slot_phys[] array. 278 */ 279 static long kvmppc_get_guest_page(struct kvm *kvm, unsigned long gfn, 280 struct kvm_memory_slot *memslot, 281 unsigned long psize) 282 { 283 unsigned long start; 284 long np, err; 285 struct page *page, *hpage, *pages[1]; 286 unsigned long s, pgsize; 287 unsigned long *physp; 288 unsigned int is_io, got, pgorder; 289 struct vm_area_struct *vma; 290 unsigned long pfn, i, npages; 291 292 physp = memslot->arch.slot_phys; 293 if (!physp) 294 return -EINVAL; 295 if (physp[gfn - memslot->base_gfn]) 296 return 0; 297 298 is_io = 0; 299 got = 0; 300 page = NULL; 301 pgsize = psize; 302 err = -EINVAL; 303 start = gfn_to_hva_memslot(memslot, gfn); 304 305 /* Instantiate and get the page we want access to */ 306 np = get_user_pages_fast(start, 1, 1, pages); 307 if (np != 1) { 308 /* Look up the vma for the page */ 309 down_read(¤t->mm->mmap_sem); 310 vma = find_vma(current->mm, start); 311 if (!vma || vma->vm_start > start || 312 start + psize > vma->vm_end || 313 !(vma->vm_flags & VM_PFNMAP)) 314 goto up_err; 315 is_io = hpte_cache_bits(pgprot_val(vma->vm_page_prot)); 316 pfn = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT); 317 /* check alignment of pfn vs. requested page size */ 318 if (psize > PAGE_SIZE && (pfn & ((psize >> PAGE_SHIFT) - 1))) 319 goto up_err; 320 up_read(¤t->mm->mmap_sem); 321 322 } else { 323 page = pages[0]; 324 got = KVMPPC_GOT_PAGE; 325 326 /* See if this is a large page */ 327 s = PAGE_SIZE; 328 if (PageHuge(page)) { 329 hpage = compound_head(page); 330 s <<= compound_order(hpage); 331 /* Get the whole large page if slot alignment is ok */ 332 if (s > psize && slot_is_aligned(memslot, s) && 333 !(memslot->userspace_addr & (s - 1))) { 334 start &= ~(s - 1); 335 pgsize = s; 336 get_page(hpage); 337 put_page(page); 338 page = hpage; 339 } 340 } 341 if (s < psize) 342 goto out; 343 pfn = page_to_pfn(page); 344 } 345 346 npages = pgsize >> PAGE_SHIFT; 347 pgorder = __ilog2(npages); 348 physp += (gfn - memslot->base_gfn) & ~(npages - 1); 349 spin_lock(&kvm->arch.slot_phys_lock); 350 for (i = 0; i < npages; ++i) { 351 if (!physp[i]) { 352 physp[i] = ((pfn + i) << PAGE_SHIFT) + 353 got + is_io + pgorder; 354 got = 0; 355 } 356 } 357 spin_unlock(&kvm->arch.slot_phys_lock); 358 err = 0; 359 360 out: 361 if (got) 362 put_page(page); 363 return err; 364 365 up_err: 366 up_read(¤t->mm->mmap_sem); 367 return err; 368 } 369 370 long kvmppc_virtmode_do_h_enter(struct kvm *kvm, unsigned long flags, 371 long pte_index, unsigned long pteh, 372 unsigned long ptel, unsigned long *pte_idx_ret) 373 { 374 unsigned long psize, gpa, gfn; 375 struct kvm_memory_slot *memslot; 376 long ret; 377 378 if (kvm->arch.using_mmu_notifiers) 379 goto do_insert; 380 381 psize = hpte_page_size(pteh, ptel); 382 if (!psize) 383 return H_PARAMETER; 384 385 pteh &= ~(HPTE_V_HVLOCK | HPTE_V_ABSENT | HPTE_V_VALID); 386 387 /* Find the memslot (if any) for this address */ 388 gpa = (ptel & HPTE_R_RPN) & ~(psize - 1); 389 gfn = gpa >> PAGE_SHIFT; 390 memslot = gfn_to_memslot(kvm, gfn); 391 if (memslot && !(memslot->flags & KVM_MEMSLOT_INVALID)) { 392 if (!slot_is_aligned(memslot, psize)) 393 return H_PARAMETER; 394 if (kvmppc_get_guest_page(kvm, gfn, memslot, psize) < 0) 395 return H_PARAMETER; 396 } 397 398 do_insert: 399 /* Protect linux PTE lookup from page table destruction */ 400 rcu_read_lock_sched(); /* this disables preemption too */ 401 ret = kvmppc_do_h_enter(kvm, flags, pte_index, pteh, ptel, 402 current->mm->pgd, false, pte_idx_ret); 403 rcu_read_unlock_sched(); 404 if (ret == H_TOO_HARD) { 405 /* this can't happen */ 406 pr_err("KVM: Oops, kvmppc_h_enter returned too hard!\n"); 407 ret = H_RESOURCE; /* or something */ 408 } 409 return ret; 410 411 } 412 413 /* 414 * We come here on a H_ENTER call from the guest when we are not 415 * using mmu notifiers and we don't have the requested page pinned 416 * already. 417 */ 418 long kvmppc_virtmode_h_enter(struct kvm_vcpu *vcpu, unsigned long flags, 419 long pte_index, unsigned long pteh, 420 unsigned long ptel) 421 { 422 return kvmppc_virtmode_do_h_enter(vcpu->kvm, flags, pte_index, 423 pteh, ptel, &vcpu->arch.gpr[4]); 424 } 425 426 static struct kvmppc_slb *kvmppc_mmu_book3s_hv_find_slbe(struct kvm_vcpu *vcpu, 427 gva_t eaddr) 428 { 429 u64 mask; 430 int i; 431 432 for (i = 0; i < vcpu->arch.slb_nr; i++) { 433 if (!(vcpu->arch.slb[i].orige & SLB_ESID_V)) 434 continue; 435 436 if (vcpu->arch.slb[i].origv & SLB_VSID_B_1T) 437 mask = ESID_MASK_1T; 438 else 439 mask = ESID_MASK; 440 441 if (((vcpu->arch.slb[i].orige ^ eaddr) & mask) == 0) 442 return &vcpu->arch.slb[i]; 443 } 444 return NULL; 445 } 446 447 static unsigned long kvmppc_mmu_get_real_addr(unsigned long v, unsigned long r, 448 unsigned long ea) 449 { 450 unsigned long ra_mask; 451 452 ra_mask = hpte_page_size(v, r) - 1; 453 return (r & HPTE_R_RPN & ~ra_mask) | (ea & ra_mask); 454 } 455 456 static int kvmppc_mmu_book3s_64_hv_xlate(struct kvm_vcpu *vcpu, gva_t eaddr, 457 struct kvmppc_pte *gpte, bool data, bool iswrite) 458 { 459 struct kvm *kvm = vcpu->kvm; 460 struct kvmppc_slb *slbe; 461 unsigned long slb_v; 462 unsigned long pp, key; 463 unsigned long v, gr; 464 unsigned long *hptep; 465 int index; 466 int virtmode = vcpu->arch.shregs.msr & (data ? MSR_DR : MSR_IR); 467 468 /* Get SLB entry */ 469 if (virtmode) { 470 slbe = kvmppc_mmu_book3s_hv_find_slbe(vcpu, eaddr); 471 if (!slbe) 472 return -EINVAL; 473 slb_v = slbe->origv; 474 } else { 475 /* real mode access */ 476 slb_v = vcpu->kvm->arch.vrma_slb_v; 477 } 478 479 preempt_disable(); 480 /* Find the HPTE in the hash table */ 481 index = kvmppc_hv_find_lock_hpte(kvm, eaddr, slb_v, 482 HPTE_V_VALID | HPTE_V_ABSENT); 483 if (index < 0) { 484 preempt_enable(); 485 return -ENOENT; 486 } 487 hptep = (unsigned long *)(kvm->arch.hpt_virt + (index << 4)); 488 v = hptep[0] & ~HPTE_V_HVLOCK; 489 gr = kvm->arch.revmap[index].guest_rpte; 490 491 /* Unlock the HPTE */ 492 asm volatile("lwsync" : : : "memory"); 493 hptep[0] = v; 494 preempt_enable(); 495 496 gpte->eaddr = eaddr; 497 gpte->vpage = ((v & HPTE_V_AVPN) << 4) | ((eaddr >> 12) & 0xfff); 498 499 /* Get PP bits and key for permission check */ 500 pp = gr & (HPTE_R_PP0 | HPTE_R_PP); 501 key = (vcpu->arch.shregs.msr & MSR_PR) ? SLB_VSID_KP : SLB_VSID_KS; 502 key &= slb_v; 503 504 /* Calculate permissions */ 505 gpte->may_read = hpte_read_permission(pp, key); 506 gpte->may_write = hpte_write_permission(pp, key); 507 gpte->may_execute = gpte->may_read && !(gr & (HPTE_R_N | HPTE_R_G)); 508 509 /* Storage key permission check for POWER7 */ 510 if (data && virtmode && cpu_has_feature(CPU_FTR_ARCH_206)) { 511 int amrfield = hpte_get_skey_perm(gr, vcpu->arch.amr); 512 if (amrfield & 1) 513 gpte->may_read = 0; 514 if (amrfield & 2) 515 gpte->may_write = 0; 516 } 517 518 /* Get the guest physical address */ 519 gpte->raddr = kvmppc_mmu_get_real_addr(v, gr, eaddr); 520 return 0; 521 } 522 523 /* 524 * Quick test for whether an instruction is a load or a store. 525 * If the instruction is a load or a store, then this will indicate 526 * which it is, at least on server processors. (Embedded processors 527 * have some external PID instructions that don't follow the rule 528 * embodied here.) If the instruction isn't a load or store, then 529 * this doesn't return anything useful. 530 */ 531 static int instruction_is_store(unsigned int instr) 532 { 533 unsigned int mask; 534 535 mask = 0x10000000; 536 if ((instr & 0xfc000000) == 0x7c000000) 537 mask = 0x100; /* major opcode 31 */ 538 return (instr & mask) != 0; 539 } 540 541 static int kvmppc_hv_emulate_mmio(struct kvm_run *run, struct kvm_vcpu *vcpu, 542 unsigned long gpa, gva_t ea, int is_store) 543 { 544 int ret; 545 u32 last_inst; 546 unsigned long srr0 = kvmppc_get_pc(vcpu); 547 548 /* We try to load the last instruction. We don't let 549 * emulate_instruction do it as it doesn't check what 550 * kvmppc_ld returns. 551 * If we fail, we just return to the guest and try executing it again. 552 */ 553 if (vcpu->arch.last_inst == KVM_INST_FETCH_FAILED) { 554 ret = kvmppc_ld(vcpu, &srr0, sizeof(u32), &last_inst, false); 555 if (ret != EMULATE_DONE || last_inst == KVM_INST_FETCH_FAILED) 556 return RESUME_GUEST; 557 vcpu->arch.last_inst = last_inst; 558 } 559 560 /* 561 * WARNING: We do not know for sure whether the instruction we just 562 * read from memory is the same that caused the fault in the first 563 * place. If the instruction we read is neither an load or a store, 564 * then it can't access memory, so we don't need to worry about 565 * enforcing access permissions. So, assuming it is a load or 566 * store, we just check that its direction (load or store) is 567 * consistent with the original fault, since that's what we 568 * checked the access permissions against. If there is a mismatch 569 * we just return and retry the instruction. 570 */ 571 572 if (instruction_is_store(kvmppc_get_last_inst(vcpu)) != !!is_store) 573 return RESUME_GUEST; 574 575 /* 576 * Emulated accesses are emulated by looking at the hash for 577 * translation once, then performing the access later. The 578 * translation could be invalidated in the meantime in which 579 * point performing the subsequent memory access on the old 580 * physical address could possibly be a security hole for the 581 * guest (but not the host). 582 * 583 * This is less of an issue for MMIO stores since they aren't 584 * globally visible. It could be an issue for MMIO loads to 585 * a certain extent but we'll ignore it for now. 586 */ 587 588 vcpu->arch.paddr_accessed = gpa; 589 vcpu->arch.vaddr_accessed = ea; 590 return kvmppc_emulate_mmio(run, vcpu); 591 } 592 593 int kvmppc_book3s_hv_page_fault(struct kvm_run *run, struct kvm_vcpu *vcpu, 594 unsigned long ea, unsigned long dsisr) 595 { 596 struct kvm *kvm = vcpu->kvm; 597 unsigned long *hptep, hpte[3], r; 598 unsigned long mmu_seq, psize, pte_size; 599 unsigned long gpa, gfn, hva, pfn; 600 struct kvm_memory_slot *memslot; 601 unsigned long *rmap; 602 struct revmap_entry *rev; 603 struct page *page, *pages[1]; 604 long index, ret, npages; 605 unsigned long is_io; 606 unsigned int writing, write_ok; 607 struct vm_area_struct *vma; 608 unsigned long rcbits; 609 610 /* 611 * Real-mode code has already searched the HPT and found the 612 * entry we're interested in. Lock the entry and check that 613 * it hasn't changed. If it has, just return and re-execute the 614 * instruction. 615 */ 616 if (ea != vcpu->arch.pgfault_addr) 617 return RESUME_GUEST; 618 index = vcpu->arch.pgfault_index; 619 hptep = (unsigned long *)(kvm->arch.hpt_virt + (index << 4)); 620 rev = &kvm->arch.revmap[index]; 621 preempt_disable(); 622 while (!try_lock_hpte(hptep, HPTE_V_HVLOCK)) 623 cpu_relax(); 624 hpte[0] = hptep[0] & ~HPTE_V_HVLOCK; 625 hpte[1] = hptep[1]; 626 hpte[2] = r = rev->guest_rpte; 627 asm volatile("lwsync" : : : "memory"); 628 hptep[0] = hpte[0]; 629 preempt_enable(); 630 631 if (hpte[0] != vcpu->arch.pgfault_hpte[0] || 632 hpte[1] != vcpu->arch.pgfault_hpte[1]) 633 return RESUME_GUEST; 634 635 /* Translate the logical address and get the page */ 636 psize = hpte_page_size(hpte[0], r); 637 gpa = (r & HPTE_R_RPN & ~(psize - 1)) | (ea & (psize - 1)); 638 gfn = gpa >> PAGE_SHIFT; 639 memslot = gfn_to_memslot(kvm, gfn); 640 641 /* No memslot means it's an emulated MMIO region */ 642 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID)) 643 return kvmppc_hv_emulate_mmio(run, vcpu, gpa, ea, 644 dsisr & DSISR_ISSTORE); 645 646 if (!kvm->arch.using_mmu_notifiers) 647 return -EFAULT; /* should never get here */ 648 649 /* used to check for invalidations in progress */ 650 mmu_seq = kvm->mmu_notifier_seq; 651 smp_rmb(); 652 653 is_io = 0; 654 pfn = 0; 655 page = NULL; 656 pte_size = PAGE_SIZE; 657 writing = (dsisr & DSISR_ISSTORE) != 0; 658 /* If writing != 0, then the HPTE must allow writing, if we get here */ 659 write_ok = writing; 660 hva = gfn_to_hva_memslot(memslot, gfn); 661 npages = get_user_pages_fast(hva, 1, writing, pages); 662 if (npages < 1) { 663 /* Check if it's an I/O mapping */ 664 down_read(¤t->mm->mmap_sem); 665 vma = find_vma(current->mm, hva); 666 if (vma && vma->vm_start <= hva && hva + psize <= vma->vm_end && 667 (vma->vm_flags & VM_PFNMAP)) { 668 pfn = vma->vm_pgoff + 669 ((hva - vma->vm_start) >> PAGE_SHIFT); 670 pte_size = psize; 671 is_io = hpte_cache_bits(pgprot_val(vma->vm_page_prot)); 672 write_ok = vma->vm_flags & VM_WRITE; 673 } 674 up_read(¤t->mm->mmap_sem); 675 if (!pfn) 676 return -EFAULT; 677 } else { 678 page = pages[0]; 679 pfn = page_to_pfn(page); 680 if (PageHuge(page)) { 681 page = compound_head(page); 682 pte_size <<= compound_order(page); 683 } 684 /* if the guest wants write access, see if that is OK */ 685 if (!writing && hpte_is_writable(r)) { 686 unsigned int hugepage_shift; 687 pte_t *ptep, pte; 688 689 /* 690 * We need to protect against page table destruction 691 * while looking up and updating the pte. 692 */ 693 rcu_read_lock_sched(); 694 ptep = find_linux_pte_or_hugepte(current->mm->pgd, 695 hva, &hugepage_shift); 696 if (ptep) { 697 pte = kvmppc_read_update_linux_pte(ptep, 1, 698 hugepage_shift); 699 if (pte_write(pte)) 700 write_ok = 1; 701 } 702 rcu_read_unlock_sched(); 703 } 704 } 705 706 ret = -EFAULT; 707 if (psize > pte_size) 708 goto out_put; 709 710 /* Check WIMG vs. the actual page we're accessing */ 711 if (!hpte_cache_flags_ok(r, is_io)) { 712 if (is_io) 713 return -EFAULT; 714 /* 715 * Allow guest to map emulated device memory as 716 * uncacheable, but actually make it cacheable. 717 */ 718 r = (r & ~(HPTE_R_W|HPTE_R_I|HPTE_R_G)) | HPTE_R_M; 719 } 720 721 /* 722 * Set the HPTE to point to pfn. 723 * Since the pfn is at PAGE_SIZE granularity, make sure we 724 * don't mask out lower-order bits if psize < PAGE_SIZE. 725 */ 726 if (psize < PAGE_SIZE) 727 psize = PAGE_SIZE; 728 r = (r & ~(HPTE_R_PP0 - psize)) | ((pfn << PAGE_SHIFT) & ~(psize - 1)); 729 if (hpte_is_writable(r) && !write_ok) 730 r = hpte_make_readonly(r); 731 ret = RESUME_GUEST; 732 preempt_disable(); 733 while (!try_lock_hpte(hptep, HPTE_V_HVLOCK)) 734 cpu_relax(); 735 if ((hptep[0] & ~HPTE_V_HVLOCK) != hpte[0] || hptep[1] != hpte[1] || 736 rev->guest_rpte != hpte[2]) 737 /* HPTE has been changed under us; let the guest retry */ 738 goto out_unlock; 739 hpte[0] = (hpte[0] & ~HPTE_V_ABSENT) | HPTE_V_VALID; 740 741 rmap = &memslot->arch.rmap[gfn - memslot->base_gfn]; 742 lock_rmap(rmap); 743 744 /* Check if we might have been invalidated; let the guest retry if so */ 745 ret = RESUME_GUEST; 746 if (mmu_notifier_retry(vcpu->kvm, mmu_seq)) { 747 unlock_rmap(rmap); 748 goto out_unlock; 749 } 750 751 /* Only set R/C in real HPTE if set in both *rmap and guest_rpte */ 752 rcbits = *rmap >> KVMPPC_RMAP_RC_SHIFT; 753 r &= rcbits | ~(HPTE_R_R | HPTE_R_C); 754 755 if (hptep[0] & HPTE_V_VALID) { 756 /* HPTE was previously valid, so we need to invalidate it */ 757 unlock_rmap(rmap); 758 hptep[0] |= HPTE_V_ABSENT; 759 kvmppc_invalidate_hpte(kvm, hptep, index); 760 /* don't lose previous R and C bits */ 761 r |= hptep[1] & (HPTE_R_R | HPTE_R_C); 762 } else { 763 kvmppc_add_revmap_chain(kvm, rev, rmap, index, 0); 764 } 765 766 hptep[1] = r; 767 eieio(); 768 hptep[0] = hpte[0]; 769 asm volatile("ptesync" : : : "memory"); 770 preempt_enable(); 771 if (page && hpte_is_writable(r)) 772 SetPageDirty(page); 773 774 out_put: 775 if (page) { 776 /* 777 * We drop pages[0] here, not page because page might 778 * have been set to the head page of a compound, but 779 * we have to drop the reference on the correct tail 780 * page to match the get inside gup() 781 */ 782 put_page(pages[0]); 783 } 784 return ret; 785 786 out_unlock: 787 hptep[0] &= ~HPTE_V_HVLOCK; 788 preempt_enable(); 789 goto out_put; 790 } 791 792 static void kvmppc_rmap_reset(struct kvm *kvm) 793 { 794 struct kvm_memslots *slots; 795 struct kvm_memory_slot *memslot; 796 int srcu_idx; 797 798 srcu_idx = srcu_read_lock(&kvm->srcu); 799 slots = kvm->memslots; 800 kvm_for_each_memslot(memslot, slots) { 801 /* 802 * This assumes it is acceptable to lose reference and 803 * change bits across a reset. 804 */ 805 memset(memslot->arch.rmap, 0, 806 memslot->npages * sizeof(*memslot->arch.rmap)); 807 } 808 srcu_read_unlock(&kvm->srcu, srcu_idx); 809 } 810 811 static int kvm_handle_hva_range(struct kvm *kvm, 812 unsigned long start, 813 unsigned long end, 814 int (*handler)(struct kvm *kvm, 815 unsigned long *rmapp, 816 unsigned long gfn)) 817 { 818 int ret; 819 int retval = 0; 820 struct kvm_memslots *slots; 821 struct kvm_memory_slot *memslot; 822 823 slots = kvm_memslots(kvm); 824 kvm_for_each_memslot(memslot, slots) { 825 unsigned long hva_start, hva_end; 826 gfn_t gfn, gfn_end; 827 828 hva_start = max(start, memslot->userspace_addr); 829 hva_end = min(end, memslot->userspace_addr + 830 (memslot->npages << PAGE_SHIFT)); 831 if (hva_start >= hva_end) 832 continue; 833 /* 834 * {gfn(page) | page intersects with [hva_start, hva_end)} = 835 * {gfn, gfn+1, ..., gfn_end-1}. 836 */ 837 gfn = hva_to_gfn_memslot(hva_start, memslot); 838 gfn_end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, memslot); 839 840 for (; gfn < gfn_end; ++gfn) { 841 gfn_t gfn_offset = gfn - memslot->base_gfn; 842 843 ret = handler(kvm, &memslot->arch.rmap[gfn_offset], gfn); 844 retval |= ret; 845 } 846 } 847 848 return retval; 849 } 850 851 static int kvm_handle_hva(struct kvm *kvm, unsigned long hva, 852 int (*handler)(struct kvm *kvm, unsigned long *rmapp, 853 unsigned long gfn)) 854 { 855 return kvm_handle_hva_range(kvm, hva, hva + 1, handler); 856 } 857 858 static int kvm_unmap_rmapp(struct kvm *kvm, unsigned long *rmapp, 859 unsigned long gfn) 860 { 861 struct revmap_entry *rev = kvm->arch.revmap; 862 unsigned long h, i, j; 863 unsigned long *hptep; 864 unsigned long ptel, psize, rcbits; 865 866 for (;;) { 867 lock_rmap(rmapp); 868 if (!(*rmapp & KVMPPC_RMAP_PRESENT)) { 869 unlock_rmap(rmapp); 870 break; 871 } 872 873 /* 874 * To avoid an ABBA deadlock with the HPTE lock bit, 875 * we can't spin on the HPTE lock while holding the 876 * rmap chain lock. 877 */ 878 i = *rmapp & KVMPPC_RMAP_INDEX; 879 hptep = (unsigned long *) (kvm->arch.hpt_virt + (i << 4)); 880 if (!try_lock_hpte(hptep, HPTE_V_HVLOCK)) { 881 /* unlock rmap before spinning on the HPTE lock */ 882 unlock_rmap(rmapp); 883 while (hptep[0] & HPTE_V_HVLOCK) 884 cpu_relax(); 885 continue; 886 } 887 j = rev[i].forw; 888 if (j == i) { 889 /* chain is now empty */ 890 *rmapp &= ~(KVMPPC_RMAP_PRESENT | KVMPPC_RMAP_INDEX); 891 } else { 892 /* remove i from chain */ 893 h = rev[i].back; 894 rev[h].forw = j; 895 rev[j].back = h; 896 rev[i].forw = rev[i].back = i; 897 *rmapp = (*rmapp & ~KVMPPC_RMAP_INDEX) | j; 898 } 899 900 /* Now check and modify the HPTE */ 901 ptel = rev[i].guest_rpte; 902 psize = hpte_page_size(hptep[0], ptel); 903 if ((hptep[0] & HPTE_V_VALID) && 904 hpte_rpn(ptel, psize) == gfn) { 905 if (kvm->arch.using_mmu_notifiers) 906 hptep[0] |= HPTE_V_ABSENT; 907 kvmppc_invalidate_hpte(kvm, hptep, i); 908 /* Harvest R and C */ 909 rcbits = hptep[1] & (HPTE_R_R | HPTE_R_C); 910 *rmapp |= rcbits << KVMPPC_RMAP_RC_SHIFT; 911 if (rcbits & ~rev[i].guest_rpte) { 912 rev[i].guest_rpte = ptel | rcbits; 913 note_hpte_modification(kvm, &rev[i]); 914 } 915 } 916 unlock_rmap(rmapp); 917 hptep[0] &= ~HPTE_V_HVLOCK; 918 } 919 return 0; 920 } 921 922 int kvm_unmap_hva_hv(struct kvm *kvm, unsigned long hva) 923 { 924 if (kvm->arch.using_mmu_notifiers) 925 kvm_handle_hva(kvm, hva, kvm_unmap_rmapp); 926 return 0; 927 } 928 929 int kvm_unmap_hva_range_hv(struct kvm *kvm, unsigned long start, unsigned long end) 930 { 931 if (kvm->arch.using_mmu_notifiers) 932 kvm_handle_hva_range(kvm, start, end, kvm_unmap_rmapp); 933 return 0; 934 } 935 936 void kvmppc_core_flush_memslot_hv(struct kvm *kvm, 937 struct kvm_memory_slot *memslot) 938 { 939 unsigned long *rmapp; 940 unsigned long gfn; 941 unsigned long n; 942 943 rmapp = memslot->arch.rmap; 944 gfn = memslot->base_gfn; 945 for (n = memslot->npages; n; --n) { 946 /* 947 * Testing the present bit without locking is OK because 948 * the memslot has been marked invalid already, and hence 949 * no new HPTEs referencing this page can be created, 950 * thus the present bit can't go from 0 to 1. 951 */ 952 if (*rmapp & KVMPPC_RMAP_PRESENT) 953 kvm_unmap_rmapp(kvm, rmapp, gfn); 954 ++rmapp; 955 ++gfn; 956 } 957 } 958 959 static int kvm_age_rmapp(struct kvm *kvm, unsigned long *rmapp, 960 unsigned long gfn) 961 { 962 struct revmap_entry *rev = kvm->arch.revmap; 963 unsigned long head, i, j; 964 unsigned long *hptep; 965 int ret = 0; 966 967 retry: 968 lock_rmap(rmapp); 969 if (*rmapp & KVMPPC_RMAP_REFERENCED) { 970 *rmapp &= ~KVMPPC_RMAP_REFERENCED; 971 ret = 1; 972 } 973 if (!(*rmapp & KVMPPC_RMAP_PRESENT)) { 974 unlock_rmap(rmapp); 975 return ret; 976 } 977 978 i = head = *rmapp & KVMPPC_RMAP_INDEX; 979 do { 980 hptep = (unsigned long *) (kvm->arch.hpt_virt + (i << 4)); 981 j = rev[i].forw; 982 983 /* If this HPTE isn't referenced, ignore it */ 984 if (!(hptep[1] & HPTE_R_R)) 985 continue; 986 987 if (!try_lock_hpte(hptep, HPTE_V_HVLOCK)) { 988 /* unlock rmap before spinning on the HPTE lock */ 989 unlock_rmap(rmapp); 990 while (hptep[0] & HPTE_V_HVLOCK) 991 cpu_relax(); 992 goto retry; 993 } 994 995 /* Now check and modify the HPTE */ 996 if ((hptep[0] & HPTE_V_VALID) && (hptep[1] & HPTE_R_R)) { 997 kvmppc_clear_ref_hpte(kvm, hptep, i); 998 if (!(rev[i].guest_rpte & HPTE_R_R)) { 999 rev[i].guest_rpte |= HPTE_R_R; 1000 note_hpte_modification(kvm, &rev[i]); 1001 } 1002 ret = 1; 1003 } 1004 hptep[0] &= ~HPTE_V_HVLOCK; 1005 } while ((i = j) != head); 1006 1007 unlock_rmap(rmapp); 1008 return ret; 1009 } 1010 1011 int kvm_age_hva_hv(struct kvm *kvm, unsigned long hva) 1012 { 1013 if (!kvm->arch.using_mmu_notifiers) 1014 return 0; 1015 return kvm_handle_hva(kvm, hva, kvm_age_rmapp); 1016 } 1017 1018 static int kvm_test_age_rmapp(struct kvm *kvm, unsigned long *rmapp, 1019 unsigned long gfn) 1020 { 1021 struct revmap_entry *rev = kvm->arch.revmap; 1022 unsigned long head, i, j; 1023 unsigned long *hp; 1024 int ret = 1; 1025 1026 if (*rmapp & KVMPPC_RMAP_REFERENCED) 1027 return 1; 1028 1029 lock_rmap(rmapp); 1030 if (*rmapp & KVMPPC_RMAP_REFERENCED) 1031 goto out; 1032 1033 if (*rmapp & KVMPPC_RMAP_PRESENT) { 1034 i = head = *rmapp & KVMPPC_RMAP_INDEX; 1035 do { 1036 hp = (unsigned long *)(kvm->arch.hpt_virt + (i << 4)); 1037 j = rev[i].forw; 1038 if (hp[1] & HPTE_R_R) 1039 goto out; 1040 } while ((i = j) != head); 1041 } 1042 ret = 0; 1043 1044 out: 1045 unlock_rmap(rmapp); 1046 return ret; 1047 } 1048 1049 int kvm_test_age_hva_hv(struct kvm *kvm, unsigned long hva) 1050 { 1051 if (!kvm->arch.using_mmu_notifiers) 1052 return 0; 1053 return kvm_handle_hva(kvm, hva, kvm_test_age_rmapp); 1054 } 1055 1056 void kvm_set_spte_hva_hv(struct kvm *kvm, unsigned long hva, pte_t pte) 1057 { 1058 if (!kvm->arch.using_mmu_notifiers) 1059 return; 1060 kvm_handle_hva(kvm, hva, kvm_unmap_rmapp); 1061 } 1062 1063 static int kvm_test_clear_dirty(struct kvm *kvm, unsigned long *rmapp) 1064 { 1065 struct revmap_entry *rev = kvm->arch.revmap; 1066 unsigned long head, i, j; 1067 unsigned long *hptep; 1068 int ret = 0; 1069 1070 retry: 1071 lock_rmap(rmapp); 1072 if (*rmapp & KVMPPC_RMAP_CHANGED) { 1073 *rmapp &= ~KVMPPC_RMAP_CHANGED; 1074 ret = 1; 1075 } 1076 if (!(*rmapp & KVMPPC_RMAP_PRESENT)) { 1077 unlock_rmap(rmapp); 1078 return ret; 1079 } 1080 1081 i = head = *rmapp & KVMPPC_RMAP_INDEX; 1082 do { 1083 hptep = (unsigned long *) (kvm->arch.hpt_virt + (i << 4)); 1084 j = rev[i].forw; 1085 1086 if (!(hptep[1] & HPTE_R_C)) 1087 continue; 1088 1089 if (!try_lock_hpte(hptep, HPTE_V_HVLOCK)) { 1090 /* unlock rmap before spinning on the HPTE lock */ 1091 unlock_rmap(rmapp); 1092 while (hptep[0] & HPTE_V_HVLOCK) 1093 cpu_relax(); 1094 goto retry; 1095 } 1096 1097 /* Now check and modify the HPTE */ 1098 if ((hptep[0] & HPTE_V_VALID) && (hptep[1] & HPTE_R_C)) { 1099 /* need to make it temporarily absent to clear C */ 1100 hptep[0] |= HPTE_V_ABSENT; 1101 kvmppc_invalidate_hpte(kvm, hptep, i); 1102 hptep[1] &= ~HPTE_R_C; 1103 eieio(); 1104 hptep[0] = (hptep[0] & ~HPTE_V_ABSENT) | HPTE_V_VALID; 1105 if (!(rev[i].guest_rpte & HPTE_R_C)) { 1106 rev[i].guest_rpte |= HPTE_R_C; 1107 note_hpte_modification(kvm, &rev[i]); 1108 } 1109 ret = 1; 1110 } 1111 hptep[0] &= ~HPTE_V_HVLOCK; 1112 } while ((i = j) != head); 1113 1114 unlock_rmap(rmapp); 1115 return ret; 1116 } 1117 1118 static void harvest_vpa_dirty(struct kvmppc_vpa *vpa, 1119 struct kvm_memory_slot *memslot, 1120 unsigned long *map) 1121 { 1122 unsigned long gfn; 1123 1124 if (!vpa->dirty || !vpa->pinned_addr) 1125 return; 1126 gfn = vpa->gpa >> PAGE_SHIFT; 1127 if (gfn < memslot->base_gfn || 1128 gfn >= memslot->base_gfn + memslot->npages) 1129 return; 1130 1131 vpa->dirty = false; 1132 if (map) 1133 __set_bit_le(gfn - memslot->base_gfn, map); 1134 } 1135 1136 long kvmppc_hv_get_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot, 1137 unsigned long *map) 1138 { 1139 unsigned long i; 1140 unsigned long *rmapp; 1141 struct kvm_vcpu *vcpu; 1142 1143 preempt_disable(); 1144 rmapp = memslot->arch.rmap; 1145 for (i = 0; i < memslot->npages; ++i) { 1146 if (kvm_test_clear_dirty(kvm, rmapp) && map) 1147 __set_bit_le(i, map); 1148 ++rmapp; 1149 } 1150 1151 /* Harvest dirty bits from VPA and DTL updates */ 1152 /* Note: we never modify the SLB shadow buffer areas */ 1153 kvm_for_each_vcpu(i, vcpu, kvm) { 1154 spin_lock(&vcpu->arch.vpa_update_lock); 1155 harvest_vpa_dirty(&vcpu->arch.vpa, memslot, map); 1156 harvest_vpa_dirty(&vcpu->arch.dtl, memslot, map); 1157 spin_unlock(&vcpu->arch.vpa_update_lock); 1158 } 1159 preempt_enable(); 1160 return 0; 1161 } 1162 1163 void *kvmppc_pin_guest_page(struct kvm *kvm, unsigned long gpa, 1164 unsigned long *nb_ret) 1165 { 1166 struct kvm_memory_slot *memslot; 1167 unsigned long gfn = gpa >> PAGE_SHIFT; 1168 struct page *page, *pages[1]; 1169 int npages; 1170 unsigned long hva, offset; 1171 unsigned long pa; 1172 unsigned long *physp; 1173 int srcu_idx; 1174 1175 srcu_idx = srcu_read_lock(&kvm->srcu); 1176 memslot = gfn_to_memslot(kvm, gfn); 1177 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID)) 1178 goto err; 1179 if (!kvm->arch.using_mmu_notifiers) { 1180 physp = memslot->arch.slot_phys; 1181 if (!physp) 1182 goto err; 1183 physp += gfn - memslot->base_gfn; 1184 pa = *physp; 1185 if (!pa) { 1186 if (kvmppc_get_guest_page(kvm, gfn, memslot, 1187 PAGE_SIZE) < 0) 1188 goto err; 1189 pa = *physp; 1190 } 1191 page = pfn_to_page(pa >> PAGE_SHIFT); 1192 get_page(page); 1193 } else { 1194 hva = gfn_to_hva_memslot(memslot, gfn); 1195 npages = get_user_pages_fast(hva, 1, 1, pages); 1196 if (npages < 1) 1197 goto err; 1198 page = pages[0]; 1199 } 1200 srcu_read_unlock(&kvm->srcu, srcu_idx); 1201 1202 offset = gpa & (PAGE_SIZE - 1); 1203 if (nb_ret) 1204 *nb_ret = PAGE_SIZE - offset; 1205 return page_address(page) + offset; 1206 1207 err: 1208 srcu_read_unlock(&kvm->srcu, srcu_idx); 1209 return NULL; 1210 } 1211 1212 void kvmppc_unpin_guest_page(struct kvm *kvm, void *va, unsigned long gpa, 1213 bool dirty) 1214 { 1215 struct page *page = virt_to_page(va); 1216 struct kvm_memory_slot *memslot; 1217 unsigned long gfn; 1218 unsigned long *rmap; 1219 int srcu_idx; 1220 1221 put_page(page); 1222 1223 if (!dirty || !kvm->arch.using_mmu_notifiers) 1224 return; 1225 1226 /* We need to mark this page dirty in the rmap chain */ 1227 gfn = gpa >> PAGE_SHIFT; 1228 srcu_idx = srcu_read_lock(&kvm->srcu); 1229 memslot = gfn_to_memslot(kvm, gfn); 1230 if (memslot) { 1231 rmap = &memslot->arch.rmap[gfn - memslot->base_gfn]; 1232 lock_rmap(rmap); 1233 *rmap |= KVMPPC_RMAP_CHANGED; 1234 unlock_rmap(rmap); 1235 } 1236 srcu_read_unlock(&kvm->srcu, srcu_idx); 1237 } 1238 1239 /* 1240 * Functions for reading and writing the hash table via reads and 1241 * writes on a file descriptor. 1242 * 1243 * Reads return the guest view of the hash table, which has to be 1244 * pieced together from the real hash table and the guest_rpte 1245 * values in the revmap array. 1246 * 1247 * On writes, each HPTE written is considered in turn, and if it 1248 * is valid, it is written to the HPT as if an H_ENTER with the 1249 * exact flag set was done. When the invalid count is non-zero 1250 * in the header written to the stream, the kernel will make 1251 * sure that that many HPTEs are invalid, and invalidate them 1252 * if not. 1253 */ 1254 1255 struct kvm_htab_ctx { 1256 unsigned long index; 1257 unsigned long flags; 1258 struct kvm *kvm; 1259 int first_pass; 1260 }; 1261 1262 #define HPTE_SIZE (2 * sizeof(unsigned long)) 1263 1264 /* 1265 * Returns 1 if this HPT entry has been modified or has pending 1266 * R/C bit changes. 1267 */ 1268 static int hpte_dirty(struct revmap_entry *revp, unsigned long *hptp) 1269 { 1270 unsigned long rcbits_unset; 1271 1272 if (revp->guest_rpte & HPTE_GR_MODIFIED) 1273 return 1; 1274 1275 /* Also need to consider changes in reference and changed bits */ 1276 rcbits_unset = ~revp->guest_rpte & (HPTE_R_R | HPTE_R_C); 1277 if ((hptp[0] & HPTE_V_VALID) && (hptp[1] & rcbits_unset)) 1278 return 1; 1279 1280 return 0; 1281 } 1282 1283 static long record_hpte(unsigned long flags, unsigned long *hptp, 1284 unsigned long *hpte, struct revmap_entry *revp, 1285 int want_valid, int first_pass) 1286 { 1287 unsigned long v, r; 1288 unsigned long rcbits_unset; 1289 int ok = 1; 1290 int valid, dirty; 1291 1292 /* Unmodified entries are uninteresting except on the first pass */ 1293 dirty = hpte_dirty(revp, hptp); 1294 if (!first_pass && !dirty) 1295 return 0; 1296 1297 valid = 0; 1298 if (hptp[0] & (HPTE_V_VALID | HPTE_V_ABSENT)) { 1299 valid = 1; 1300 if ((flags & KVM_GET_HTAB_BOLTED_ONLY) && 1301 !(hptp[0] & HPTE_V_BOLTED)) 1302 valid = 0; 1303 } 1304 if (valid != want_valid) 1305 return 0; 1306 1307 v = r = 0; 1308 if (valid || dirty) { 1309 /* lock the HPTE so it's stable and read it */ 1310 preempt_disable(); 1311 while (!try_lock_hpte(hptp, HPTE_V_HVLOCK)) 1312 cpu_relax(); 1313 v = hptp[0]; 1314 1315 /* re-evaluate valid and dirty from synchronized HPTE value */ 1316 valid = !!(v & HPTE_V_VALID); 1317 dirty = !!(revp->guest_rpte & HPTE_GR_MODIFIED); 1318 1319 /* Harvest R and C into guest view if necessary */ 1320 rcbits_unset = ~revp->guest_rpte & (HPTE_R_R | HPTE_R_C); 1321 if (valid && (rcbits_unset & hptp[1])) { 1322 revp->guest_rpte |= (hptp[1] & (HPTE_R_R | HPTE_R_C)) | 1323 HPTE_GR_MODIFIED; 1324 dirty = 1; 1325 } 1326 1327 if (v & HPTE_V_ABSENT) { 1328 v &= ~HPTE_V_ABSENT; 1329 v |= HPTE_V_VALID; 1330 valid = 1; 1331 } 1332 if ((flags & KVM_GET_HTAB_BOLTED_ONLY) && !(v & HPTE_V_BOLTED)) 1333 valid = 0; 1334 1335 r = revp->guest_rpte; 1336 /* only clear modified if this is the right sort of entry */ 1337 if (valid == want_valid && dirty) { 1338 r &= ~HPTE_GR_MODIFIED; 1339 revp->guest_rpte = r; 1340 } 1341 asm volatile(PPC_RELEASE_BARRIER "" : : : "memory"); 1342 hptp[0] &= ~HPTE_V_HVLOCK; 1343 preempt_enable(); 1344 if (!(valid == want_valid && (first_pass || dirty))) 1345 ok = 0; 1346 } 1347 hpte[0] = v; 1348 hpte[1] = r; 1349 return ok; 1350 } 1351 1352 static ssize_t kvm_htab_read(struct file *file, char __user *buf, 1353 size_t count, loff_t *ppos) 1354 { 1355 struct kvm_htab_ctx *ctx = file->private_data; 1356 struct kvm *kvm = ctx->kvm; 1357 struct kvm_get_htab_header hdr; 1358 unsigned long *hptp; 1359 struct revmap_entry *revp; 1360 unsigned long i, nb, nw; 1361 unsigned long __user *lbuf; 1362 struct kvm_get_htab_header __user *hptr; 1363 unsigned long flags; 1364 int first_pass; 1365 unsigned long hpte[2]; 1366 1367 if (!access_ok(VERIFY_WRITE, buf, count)) 1368 return -EFAULT; 1369 1370 first_pass = ctx->first_pass; 1371 flags = ctx->flags; 1372 1373 i = ctx->index; 1374 hptp = (unsigned long *)(kvm->arch.hpt_virt + (i * HPTE_SIZE)); 1375 revp = kvm->arch.revmap + i; 1376 lbuf = (unsigned long __user *)buf; 1377 1378 nb = 0; 1379 while (nb + sizeof(hdr) + HPTE_SIZE < count) { 1380 /* Initialize header */ 1381 hptr = (struct kvm_get_htab_header __user *)buf; 1382 hdr.n_valid = 0; 1383 hdr.n_invalid = 0; 1384 nw = nb; 1385 nb += sizeof(hdr); 1386 lbuf = (unsigned long __user *)(buf + sizeof(hdr)); 1387 1388 /* Skip uninteresting entries, i.e. clean on not-first pass */ 1389 if (!first_pass) { 1390 while (i < kvm->arch.hpt_npte && 1391 !hpte_dirty(revp, hptp)) { 1392 ++i; 1393 hptp += 2; 1394 ++revp; 1395 } 1396 } 1397 hdr.index = i; 1398 1399 /* Grab a series of valid entries */ 1400 while (i < kvm->arch.hpt_npte && 1401 hdr.n_valid < 0xffff && 1402 nb + HPTE_SIZE < count && 1403 record_hpte(flags, hptp, hpte, revp, 1, first_pass)) { 1404 /* valid entry, write it out */ 1405 ++hdr.n_valid; 1406 if (__put_user(hpte[0], lbuf) || 1407 __put_user(hpte[1], lbuf + 1)) 1408 return -EFAULT; 1409 nb += HPTE_SIZE; 1410 lbuf += 2; 1411 ++i; 1412 hptp += 2; 1413 ++revp; 1414 } 1415 /* Now skip invalid entries while we can */ 1416 while (i < kvm->arch.hpt_npte && 1417 hdr.n_invalid < 0xffff && 1418 record_hpte(flags, hptp, hpte, revp, 0, first_pass)) { 1419 /* found an invalid entry */ 1420 ++hdr.n_invalid; 1421 ++i; 1422 hptp += 2; 1423 ++revp; 1424 } 1425 1426 if (hdr.n_valid || hdr.n_invalid) { 1427 /* write back the header */ 1428 if (__copy_to_user(hptr, &hdr, sizeof(hdr))) 1429 return -EFAULT; 1430 nw = nb; 1431 buf = (char __user *)lbuf; 1432 } else { 1433 nb = nw; 1434 } 1435 1436 /* Check if we've wrapped around the hash table */ 1437 if (i >= kvm->arch.hpt_npte) { 1438 i = 0; 1439 ctx->first_pass = 0; 1440 break; 1441 } 1442 } 1443 1444 ctx->index = i; 1445 1446 return nb; 1447 } 1448 1449 static ssize_t kvm_htab_write(struct file *file, const char __user *buf, 1450 size_t count, loff_t *ppos) 1451 { 1452 struct kvm_htab_ctx *ctx = file->private_data; 1453 struct kvm *kvm = ctx->kvm; 1454 struct kvm_get_htab_header hdr; 1455 unsigned long i, j; 1456 unsigned long v, r; 1457 unsigned long __user *lbuf; 1458 unsigned long *hptp; 1459 unsigned long tmp[2]; 1460 ssize_t nb; 1461 long int err, ret; 1462 int rma_setup; 1463 1464 if (!access_ok(VERIFY_READ, buf, count)) 1465 return -EFAULT; 1466 1467 /* lock out vcpus from running while we're doing this */ 1468 mutex_lock(&kvm->lock); 1469 rma_setup = kvm->arch.rma_setup_done; 1470 if (rma_setup) { 1471 kvm->arch.rma_setup_done = 0; /* temporarily */ 1472 /* order rma_setup_done vs. vcpus_running */ 1473 smp_mb(); 1474 if (atomic_read(&kvm->arch.vcpus_running)) { 1475 kvm->arch.rma_setup_done = 1; 1476 mutex_unlock(&kvm->lock); 1477 return -EBUSY; 1478 } 1479 } 1480 1481 err = 0; 1482 for (nb = 0; nb + sizeof(hdr) <= count; ) { 1483 err = -EFAULT; 1484 if (__copy_from_user(&hdr, buf, sizeof(hdr))) 1485 break; 1486 1487 err = 0; 1488 if (nb + hdr.n_valid * HPTE_SIZE > count) 1489 break; 1490 1491 nb += sizeof(hdr); 1492 buf += sizeof(hdr); 1493 1494 err = -EINVAL; 1495 i = hdr.index; 1496 if (i >= kvm->arch.hpt_npte || 1497 i + hdr.n_valid + hdr.n_invalid > kvm->arch.hpt_npte) 1498 break; 1499 1500 hptp = (unsigned long *)(kvm->arch.hpt_virt + (i * HPTE_SIZE)); 1501 lbuf = (unsigned long __user *)buf; 1502 for (j = 0; j < hdr.n_valid; ++j) { 1503 err = -EFAULT; 1504 if (__get_user(v, lbuf) || __get_user(r, lbuf + 1)) 1505 goto out; 1506 err = -EINVAL; 1507 if (!(v & HPTE_V_VALID)) 1508 goto out; 1509 lbuf += 2; 1510 nb += HPTE_SIZE; 1511 1512 if (hptp[0] & (HPTE_V_VALID | HPTE_V_ABSENT)) 1513 kvmppc_do_h_remove(kvm, 0, i, 0, tmp); 1514 err = -EIO; 1515 ret = kvmppc_virtmode_do_h_enter(kvm, H_EXACT, i, v, r, 1516 tmp); 1517 if (ret != H_SUCCESS) { 1518 pr_err("kvm_htab_write ret %ld i=%ld v=%lx " 1519 "r=%lx\n", ret, i, v, r); 1520 goto out; 1521 } 1522 if (!rma_setup && is_vrma_hpte(v)) { 1523 unsigned long psize = hpte_page_size(v, r); 1524 unsigned long senc = slb_pgsize_encoding(psize); 1525 unsigned long lpcr; 1526 1527 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T | 1528 (VRMA_VSID << SLB_VSID_SHIFT_1T); 1529 lpcr = senc << (LPCR_VRMASD_SH - 4); 1530 kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD); 1531 rma_setup = 1; 1532 } 1533 ++i; 1534 hptp += 2; 1535 } 1536 1537 for (j = 0; j < hdr.n_invalid; ++j) { 1538 if (hptp[0] & (HPTE_V_VALID | HPTE_V_ABSENT)) 1539 kvmppc_do_h_remove(kvm, 0, i, 0, tmp); 1540 ++i; 1541 hptp += 2; 1542 } 1543 err = 0; 1544 } 1545 1546 out: 1547 /* Order HPTE updates vs. rma_setup_done */ 1548 smp_wmb(); 1549 kvm->arch.rma_setup_done = rma_setup; 1550 mutex_unlock(&kvm->lock); 1551 1552 if (err) 1553 return err; 1554 return nb; 1555 } 1556 1557 static int kvm_htab_release(struct inode *inode, struct file *filp) 1558 { 1559 struct kvm_htab_ctx *ctx = filp->private_data; 1560 1561 filp->private_data = NULL; 1562 if (!(ctx->flags & KVM_GET_HTAB_WRITE)) 1563 atomic_dec(&ctx->kvm->arch.hpte_mod_interest); 1564 kvm_put_kvm(ctx->kvm); 1565 kfree(ctx); 1566 return 0; 1567 } 1568 1569 static const struct file_operations kvm_htab_fops = { 1570 .read = kvm_htab_read, 1571 .write = kvm_htab_write, 1572 .llseek = default_llseek, 1573 .release = kvm_htab_release, 1574 }; 1575 1576 int kvm_vm_ioctl_get_htab_fd(struct kvm *kvm, struct kvm_get_htab_fd *ghf) 1577 { 1578 int ret; 1579 struct kvm_htab_ctx *ctx; 1580 int rwflag; 1581 1582 /* reject flags we don't recognize */ 1583 if (ghf->flags & ~(KVM_GET_HTAB_BOLTED_ONLY | KVM_GET_HTAB_WRITE)) 1584 return -EINVAL; 1585 ctx = kzalloc(sizeof(*ctx), GFP_KERNEL); 1586 if (!ctx) 1587 return -ENOMEM; 1588 kvm_get_kvm(kvm); 1589 ctx->kvm = kvm; 1590 ctx->index = ghf->start_index; 1591 ctx->flags = ghf->flags; 1592 ctx->first_pass = 1; 1593 1594 rwflag = (ghf->flags & KVM_GET_HTAB_WRITE) ? O_WRONLY : O_RDONLY; 1595 ret = anon_inode_getfd("kvm-htab", &kvm_htab_fops, ctx, rwflag | O_CLOEXEC); 1596 if (ret < 0) { 1597 kvm_put_kvm(kvm); 1598 return ret; 1599 } 1600 1601 if (rwflag == O_RDONLY) { 1602 mutex_lock(&kvm->slots_lock); 1603 atomic_inc(&kvm->arch.hpte_mod_interest); 1604 /* make sure kvmppc_do_h_enter etc. see the increment */ 1605 synchronize_srcu_expedited(&kvm->srcu); 1606 mutex_unlock(&kvm->slots_lock); 1607 } 1608 1609 return ret; 1610 } 1611 1612 void kvmppc_mmu_book3s_hv_init(struct kvm_vcpu *vcpu) 1613 { 1614 struct kvmppc_mmu *mmu = &vcpu->arch.mmu; 1615 1616 if (cpu_has_feature(CPU_FTR_ARCH_206)) 1617 vcpu->arch.slb_nr = 32; /* POWER7 */ 1618 else 1619 vcpu->arch.slb_nr = 64; 1620 1621 mmu->xlate = kvmppc_mmu_book3s_64_hv_xlate; 1622 mmu->reset_msr = kvmppc_mmu_book3s_64_hv_reset_msr; 1623 1624 vcpu->arch.hflags |= BOOK3S_HFLAG_SLB; 1625 } 1626