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.
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.
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.
15 * Copyright 2010 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
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>
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>
40 #include "book3s_hv_cma.h"
42 /* POWER7 has 10-bit LPIDs, PPC970 has 6-bit LPIDs */
43 #define MAX_LPID_970 63
45 /* Power architecture requires HPT is at least 256kB */
46 #define PPC_MIN_HPT_ORDER 18
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);
53 long kvmppc_alloc_hpt(struct kvm *kvm, u32 *htab_orderp)
55 unsigned long hpt = 0;
56 struct revmap_entry *rev;
57 struct page *page = NULL;
58 long order = KVM_DEFAULT_HPT_ORDER;
62 if (order < PPC_MIN_HPT_ORDER)
63 order = PPC_MIN_HPT_ORDER;
66 kvm->arch.hpt_cma_alloc = 0;
67 VM_BUG_ON(order < KVM_CMA_CHUNK_ORDER);
68 page = kvm_alloc_hpt(1 << (order - PAGE_SHIFT));
70 hpt = (unsigned long)pfn_to_kaddr(page_to_pfn(page));
71 kvm->arch.hpt_cma_alloc = 1;
74 /* Lastly try successively smaller sizes from the page allocator */
75 while (!hpt && order > PPC_MIN_HPT_ORDER) {
76 hpt = __get_free_pages(GFP_KERNEL|__GFP_ZERO|__GFP_REPEAT|
77 __GFP_NOWARN, order - PAGE_SHIFT);
85 kvm->arch.hpt_virt = hpt;
86 kvm->arch.hpt_order = order;
87 /* HPTEs are 2**4 bytes long */
88 kvm->arch.hpt_npte = 1ul << (order - 4);
89 /* 128 (2**7) bytes in each HPTEG */
90 kvm->arch.hpt_mask = (1ul << (order - 7)) - 1;
92 /* Allocate reverse map array */
93 rev = vmalloc(sizeof(struct revmap_entry) * kvm->arch.hpt_npte);
95 pr_err("kvmppc_alloc_hpt: Couldn't alloc reverse map array\n");
98 kvm->arch.revmap = rev;
99 kvm->arch.sdr1 = __pa(hpt) | (order - 18);
101 pr_info("KVM guest htab at %lx (order %ld), LPID %x\n",
102 hpt, order, kvm->arch.lpid);
105 *htab_orderp = order;
109 if (kvm->arch.hpt_cma_alloc)
110 kvm_release_hpt(page, 1 << (order - PAGE_SHIFT));
112 free_pages(hpt, order - PAGE_SHIFT);
116 long kvmppc_alloc_reset_hpt(struct kvm *kvm, u32 *htab_orderp)
121 mutex_lock(&kvm->lock);
122 if (kvm->arch.rma_setup_done) {
123 kvm->arch.rma_setup_done = 0;
124 /* order rma_setup_done vs. vcpus_running */
126 if (atomic_read(&kvm->arch.vcpus_running)) {
127 kvm->arch.rma_setup_done = 1;
131 if (kvm->arch.hpt_virt) {
132 order = kvm->arch.hpt_order;
133 /* Set the entire HPT to 0, i.e. invalid HPTEs */
134 memset((void *)kvm->arch.hpt_virt, 0, 1ul << order);
136 * Reset all the reverse-mapping chains for all memslots
138 kvmppc_rmap_reset(kvm);
139 /* Ensure that each vcpu will flush its TLB on next entry. */
140 cpumask_setall(&kvm->arch.need_tlb_flush);
141 *htab_orderp = order;
144 err = kvmppc_alloc_hpt(kvm, htab_orderp);
145 order = *htab_orderp;
148 mutex_unlock(&kvm->lock);
152 void kvmppc_free_hpt(struct kvm *kvm)
154 kvmppc_free_lpid(kvm->arch.lpid);
155 vfree(kvm->arch.revmap);
156 if (kvm->arch.hpt_cma_alloc)
157 kvm_release_hpt(virt_to_page(kvm->arch.hpt_virt),
158 1 << (kvm->arch.hpt_order - PAGE_SHIFT));
160 free_pages(kvm->arch.hpt_virt,
161 kvm->arch.hpt_order - PAGE_SHIFT);
164 /* Bits in first HPTE dword for pagesize 4k, 64k or 16M */
165 static inline unsigned long hpte0_pgsize_encoding(unsigned long pgsize)
167 return (pgsize > 0x1000) ? HPTE_V_LARGE : 0;
170 /* Bits in second HPTE dword for pagesize 4k, 64k or 16M */
171 static inline unsigned long hpte1_pgsize_encoding(unsigned long pgsize)
173 return (pgsize == 0x10000) ? 0x1000 : 0;
176 void kvmppc_map_vrma(struct kvm_vcpu *vcpu, struct kvm_memory_slot *memslot,
177 unsigned long porder)
180 unsigned long npages;
181 unsigned long hp_v, hp_r;
182 unsigned long addr, hash;
184 unsigned long hp0, hp1;
185 unsigned long idx_ret;
187 struct kvm *kvm = vcpu->kvm;
189 psize = 1ul << porder;
190 npages = memslot->npages >> (porder - PAGE_SHIFT);
192 /* VRMA can't be > 1TB */
193 if (npages > 1ul << (40 - porder))
194 npages = 1ul << (40 - porder);
195 /* Can't use more than 1 HPTE per HPTEG */
196 if (npages > kvm->arch.hpt_mask + 1)
197 npages = kvm->arch.hpt_mask + 1;
199 hp0 = HPTE_V_1TB_SEG | (VRMA_VSID << (40 - 16)) |
200 HPTE_V_BOLTED | hpte0_pgsize_encoding(psize);
201 hp1 = hpte1_pgsize_encoding(psize) |
202 HPTE_R_R | HPTE_R_C | HPTE_R_M | PP_RWXX;
204 for (i = 0; i < npages; ++i) {
206 /* can't use hpt_hash since va > 64 bits */
207 hash = (i ^ (VRMA_VSID ^ (VRMA_VSID << 25))) & kvm->arch.hpt_mask;
209 * We assume that the hash table is empty and no
210 * vcpus are using it at this stage. Since we create
211 * at most one HPTE per HPTEG, we just assume entry 7
212 * is available and use it.
214 hash = (hash << 3) + 7;
215 hp_v = hp0 | ((addr >> 16) & ~0x7fUL);
217 ret = kvmppc_virtmode_do_h_enter(kvm, H_EXACT, hash, hp_v, hp_r,
219 if (ret != H_SUCCESS) {
220 pr_err("KVM: map_vrma at %lx failed, ret=%ld\n",
227 int kvmppc_mmu_hv_init(void)
229 unsigned long host_lpid, rsvd_lpid;
231 if (!cpu_has_feature(CPU_FTR_HVMODE))
234 /* POWER7 has 10-bit LPIDs, PPC970 and e500mc have 6-bit LPIDs */
235 if (cpu_has_feature(CPU_FTR_ARCH_206)) {
236 host_lpid = mfspr(SPRN_LPID); /* POWER7 */
237 rsvd_lpid = LPID_RSVD;
239 host_lpid = 0; /* PPC970 */
240 rsvd_lpid = MAX_LPID_970;
243 kvmppc_init_lpid(rsvd_lpid + 1);
245 kvmppc_claim_lpid(host_lpid);
246 /* rsvd_lpid is reserved for use in partition switching */
247 kvmppc_claim_lpid(rsvd_lpid);
252 static void kvmppc_mmu_book3s_64_hv_reset_msr(struct kvm_vcpu *vcpu)
254 unsigned long msr = vcpu->arch.intr_msr;
256 /* If transactional, change to suspend mode on IRQ delivery */
257 if (MSR_TM_TRANSACTIONAL(vcpu->arch.shregs.msr))
260 msr |= vcpu->arch.shregs.msr & MSR_TS_MASK;
261 kvmppc_set_msr(vcpu, msr);
265 * This is called to get a reference to a guest page if there isn't
266 * one already in the memslot->arch.slot_phys[] array.
268 static long kvmppc_get_guest_page(struct kvm *kvm, unsigned long gfn,
269 struct kvm_memory_slot *memslot,
274 struct page *page, *hpage, *pages[1];
275 unsigned long s, pgsize;
276 unsigned long *physp;
277 unsigned int is_io, got, pgorder;
278 struct vm_area_struct *vma;
279 unsigned long pfn, i, npages;
281 physp = memslot->arch.slot_phys;
284 if (physp[gfn - memslot->base_gfn])
292 start = gfn_to_hva_memslot(memslot, gfn);
294 /* Instantiate and get the page we want access to */
295 np = get_user_pages_fast(start, 1, 1, pages);
297 /* Look up the vma for the page */
298 down_read(¤t->mm->mmap_sem);
299 vma = find_vma(current->mm, start);
300 if (!vma || vma->vm_start > start ||
301 start + psize > vma->vm_end ||
302 !(vma->vm_flags & VM_PFNMAP))
304 is_io = hpte_cache_bits(pgprot_val(vma->vm_page_prot));
305 pfn = vma->vm_pgoff + ((start - vma->vm_start) >> PAGE_SHIFT);
306 /* check alignment of pfn vs. requested page size */
307 if (psize > PAGE_SIZE && (pfn & ((psize >> PAGE_SHIFT) - 1)))
309 up_read(¤t->mm->mmap_sem);
313 got = KVMPPC_GOT_PAGE;
315 /* See if this is a large page */
317 if (PageHuge(page)) {
318 hpage = compound_head(page);
319 s <<= compound_order(hpage);
320 /* Get the whole large page if slot alignment is ok */
321 if (s > psize && slot_is_aligned(memslot, s) &&
322 !(memslot->userspace_addr & (s - 1))) {
332 pfn = page_to_pfn(page);
335 npages = pgsize >> PAGE_SHIFT;
336 pgorder = __ilog2(npages);
337 physp += (gfn - memslot->base_gfn) & ~(npages - 1);
338 spin_lock(&kvm->arch.slot_phys_lock);
339 for (i = 0; i < npages; ++i) {
341 physp[i] = ((pfn + i) << PAGE_SHIFT) +
342 got + is_io + pgorder;
346 spin_unlock(&kvm->arch.slot_phys_lock);
355 up_read(¤t->mm->mmap_sem);
359 long kvmppc_virtmode_do_h_enter(struct kvm *kvm, unsigned long flags,
360 long pte_index, unsigned long pteh,
361 unsigned long ptel, unsigned long *pte_idx_ret)
363 unsigned long psize, gpa, gfn;
364 struct kvm_memory_slot *memslot;
367 if (kvm->arch.using_mmu_notifiers)
370 psize = hpte_page_size(pteh, ptel);
374 pteh &= ~(HPTE_V_HVLOCK | HPTE_V_ABSENT | HPTE_V_VALID);
376 /* Find the memslot (if any) for this address */
377 gpa = (ptel & HPTE_R_RPN) & ~(psize - 1);
378 gfn = gpa >> PAGE_SHIFT;
379 memslot = gfn_to_memslot(kvm, gfn);
380 if (memslot && !(memslot->flags & KVM_MEMSLOT_INVALID)) {
381 if (!slot_is_aligned(memslot, psize))
383 if (kvmppc_get_guest_page(kvm, gfn, memslot, psize) < 0)
388 /* Protect linux PTE lookup from page table destruction */
389 rcu_read_lock_sched(); /* this disables preemption too */
390 ret = kvmppc_do_h_enter(kvm, flags, pte_index, pteh, ptel,
391 current->mm->pgd, false, pte_idx_ret);
392 rcu_read_unlock_sched();
393 if (ret == H_TOO_HARD) {
394 /* this can't happen */
395 pr_err("KVM: Oops, kvmppc_h_enter returned too hard!\n");
396 ret = H_RESOURCE; /* or something */
403 * We come here on a H_ENTER call from the guest when we are not
404 * using mmu notifiers and we don't have the requested page pinned
407 long kvmppc_virtmode_h_enter(struct kvm_vcpu *vcpu, unsigned long flags,
408 long pte_index, unsigned long pteh,
411 return kvmppc_virtmode_do_h_enter(vcpu->kvm, flags, pte_index,
412 pteh, ptel, &vcpu->arch.gpr[4]);
415 static struct kvmppc_slb *kvmppc_mmu_book3s_hv_find_slbe(struct kvm_vcpu *vcpu,
421 for (i = 0; i < vcpu->arch.slb_nr; i++) {
422 if (!(vcpu->arch.slb[i].orige & SLB_ESID_V))
425 if (vcpu->arch.slb[i].origv & SLB_VSID_B_1T)
430 if (((vcpu->arch.slb[i].orige ^ eaddr) & mask) == 0)
431 return &vcpu->arch.slb[i];
436 static unsigned long kvmppc_mmu_get_real_addr(unsigned long v, unsigned long r,
439 unsigned long ra_mask;
441 ra_mask = hpte_page_size(v, r) - 1;
442 return (r & HPTE_R_RPN & ~ra_mask) | (ea & ra_mask);
445 static int kvmppc_mmu_book3s_64_hv_xlate(struct kvm_vcpu *vcpu, gva_t eaddr,
446 struct kvmppc_pte *gpte, bool data, bool iswrite)
448 struct kvm *kvm = vcpu->kvm;
449 struct kvmppc_slb *slbe;
451 unsigned long pp, key;
453 unsigned long *hptep;
455 int virtmode = vcpu->arch.shregs.msr & (data ? MSR_DR : MSR_IR);
459 slbe = kvmppc_mmu_book3s_hv_find_slbe(vcpu, eaddr);
464 /* real mode access */
465 slb_v = vcpu->kvm->arch.vrma_slb_v;
469 /* Find the HPTE in the hash table */
470 index = kvmppc_hv_find_lock_hpte(kvm, eaddr, slb_v,
471 HPTE_V_VALID | HPTE_V_ABSENT);
476 hptep = (unsigned long *)(kvm->arch.hpt_virt + (index << 4));
477 v = hptep[0] & ~HPTE_V_HVLOCK;
478 gr = kvm->arch.revmap[index].guest_rpte;
480 /* Unlock the HPTE */
481 asm volatile("lwsync" : : : "memory");
486 gpte->vpage = ((v & HPTE_V_AVPN) << 4) | ((eaddr >> 12) & 0xfff);
488 /* Get PP bits and key for permission check */
489 pp = gr & (HPTE_R_PP0 | HPTE_R_PP);
490 key = (vcpu->arch.shregs.msr & MSR_PR) ? SLB_VSID_KP : SLB_VSID_KS;
493 /* Calculate permissions */
494 gpte->may_read = hpte_read_permission(pp, key);
495 gpte->may_write = hpte_write_permission(pp, key);
496 gpte->may_execute = gpte->may_read && !(gr & (HPTE_R_N | HPTE_R_G));
498 /* Storage key permission check for POWER7 */
499 if (data && virtmode && cpu_has_feature(CPU_FTR_ARCH_206)) {
500 int amrfield = hpte_get_skey_perm(gr, vcpu->arch.amr);
507 /* Get the guest physical address */
508 gpte->raddr = kvmppc_mmu_get_real_addr(v, gr, eaddr);
513 * Quick test for whether an instruction is a load or a store.
514 * If the instruction is a load or a store, then this will indicate
515 * which it is, at least on server processors. (Embedded processors
516 * have some external PID instructions that don't follow the rule
517 * embodied here.) If the instruction isn't a load or store, then
518 * this doesn't return anything useful.
520 static int instruction_is_store(unsigned int instr)
525 if ((instr & 0xfc000000) == 0x7c000000)
526 mask = 0x100; /* major opcode 31 */
527 return (instr & mask) != 0;
530 static int kvmppc_hv_emulate_mmio(struct kvm_run *run, struct kvm_vcpu *vcpu,
531 unsigned long gpa, gva_t ea, int is_store)
535 unsigned long srr0 = kvmppc_get_pc(vcpu);
537 /* We try to load the last instruction. We don't let
538 * emulate_instruction do it as it doesn't check what
540 * If we fail, we just return to the guest and try executing it again.
542 if (vcpu->arch.last_inst == KVM_INST_FETCH_FAILED) {
543 ret = kvmppc_ld(vcpu, &srr0, sizeof(u32), &last_inst, false);
544 if (ret != EMULATE_DONE || last_inst == KVM_INST_FETCH_FAILED)
546 vcpu->arch.last_inst = last_inst;
550 * WARNING: We do not know for sure whether the instruction we just
551 * read from memory is the same that caused the fault in the first
552 * place. If the instruction we read is neither an load or a store,
553 * then it can't access memory, so we don't need to worry about
554 * enforcing access permissions. So, assuming it is a load or
555 * store, we just check that its direction (load or store) is
556 * consistent with the original fault, since that's what we
557 * checked the access permissions against. If there is a mismatch
558 * we just return and retry the instruction.
561 if (instruction_is_store(kvmppc_get_last_inst(vcpu)) != !!is_store)
565 * Emulated accesses are emulated by looking at the hash for
566 * translation once, then performing the access later. The
567 * translation could be invalidated in the meantime in which
568 * point performing the subsequent memory access on the old
569 * physical address could possibly be a security hole for the
570 * guest (but not the host).
572 * This is less of an issue for MMIO stores since they aren't
573 * globally visible. It could be an issue for MMIO loads to
574 * a certain extent but we'll ignore it for now.
577 vcpu->arch.paddr_accessed = gpa;
578 vcpu->arch.vaddr_accessed = ea;
579 return kvmppc_emulate_mmio(run, vcpu);
582 int kvmppc_book3s_hv_page_fault(struct kvm_run *run, struct kvm_vcpu *vcpu,
583 unsigned long ea, unsigned long dsisr)
585 struct kvm *kvm = vcpu->kvm;
586 unsigned long *hptep, hpte[3], r;
587 unsigned long mmu_seq, psize, pte_size;
588 unsigned long gpa_base, gfn_base;
589 unsigned long gpa, gfn, hva, pfn;
590 struct kvm_memory_slot *memslot;
592 struct revmap_entry *rev;
593 struct page *page, *pages[1];
594 long index, ret, npages;
596 unsigned int writing, write_ok;
597 struct vm_area_struct *vma;
598 unsigned long rcbits;
601 * Real-mode code has already searched the HPT and found the
602 * entry we're interested in. Lock the entry and check that
603 * it hasn't changed. If it has, just return and re-execute the
606 if (ea != vcpu->arch.pgfault_addr)
608 index = vcpu->arch.pgfault_index;
609 hptep = (unsigned long *)(kvm->arch.hpt_virt + (index << 4));
610 rev = &kvm->arch.revmap[index];
612 while (!try_lock_hpte(hptep, HPTE_V_HVLOCK))
614 hpte[0] = hptep[0] & ~HPTE_V_HVLOCK;
616 hpte[2] = r = rev->guest_rpte;
617 asm volatile("lwsync" : : : "memory");
621 if (hpte[0] != vcpu->arch.pgfault_hpte[0] ||
622 hpte[1] != vcpu->arch.pgfault_hpte[1])
625 /* Translate the logical address and get the page */
626 psize = hpte_page_size(hpte[0], r);
627 gpa_base = r & HPTE_R_RPN & ~(psize - 1);
628 gfn_base = gpa_base >> PAGE_SHIFT;
629 gpa = gpa_base | (ea & (psize - 1));
630 gfn = gpa >> PAGE_SHIFT;
631 memslot = gfn_to_memslot(kvm, gfn);
633 /* No memslot means it's an emulated MMIO region */
634 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
635 return kvmppc_hv_emulate_mmio(run, vcpu, gpa, ea,
636 dsisr & DSISR_ISSTORE);
638 if (!kvm->arch.using_mmu_notifiers)
639 return -EFAULT; /* should never get here */
642 * This should never happen, because of the slot_is_aligned()
643 * check in kvmppc_do_h_enter().
645 if (gfn_base < memslot->base_gfn)
648 /* used to check for invalidations in progress */
649 mmu_seq = kvm->mmu_notifier_seq;
655 pte_size = PAGE_SIZE;
656 writing = (dsisr & DSISR_ISSTORE) != 0;
657 /* If writing != 0, then the HPTE must allow writing, if we get here */
659 hva = gfn_to_hva_memslot(memslot, gfn);
660 npages = get_user_pages_fast(hva, 1, writing, pages);
662 /* Check if it's an I/O mapping */
663 down_read(¤t->mm->mmap_sem);
664 vma = find_vma(current->mm, hva);
665 if (vma && vma->vm_start <= hva && hva + psize <= vma->vm_end &&
666 (vma->vm_flags & VM_PFNMAP)) {
667 pfn = vma->vm_pgoff +
668 ((hva - vma->vm_start) >> PAGE_SHIFT);
670 is_io = hpte_cache_bits(pgprot_val(vma->vm_page_prot));
671 write_ok = vma->vm_flags & VM_WRITE;
673 up_read(¤t->mm->mmap_sem);
678 pfn = page_to_pfn(page);
679 if (PageHuge(page)) {
680 page = compound_head(page);
681 pte_size <<= compound_order(page);
683 /* if the guest wants write access, see if that is OK */
684 if (!writing && hpte_is_writable(r)) {
685 unsigned int hugepage_shift;
689 * We need to protect against page table destruction
690 * while looking up and updating the pte.
692 rcu_read_lock_sched();
693 ptep = find_linux_pte_or_hugepte(current->mm->pgd,
694 hva, &hugepage_shift);
696 pte = kvmppc_read_update_linux_pte(ptep, 1,
701 rcu_read_unlock_sched();
706 if (psize > pte_size)
709 /* Check WIMG vs. the actual page we're accessing */
710 if (!hpte_cache_flags_ok(r, is_io)) {
714 * Allow guest to map emulated device memory as
715 * uncacheable, but actually make it cacheable.
717 r = (r & ~(HPTE_R_W|HPTE_R_I|HPTE_R_G)) | HPTE_R_M;
721 * Set the HPTE to point to pfn.
722 * Since the pfn is at PAGE_SIZE granularity, make sure we
723 * don't mask out lower-order bits if psize < PAGE_SIZE.
725 if (psize < PAGE_SIZE)
727 r = (r & ~(HPTE_R_PP0 - psize)) | ((pfn << PAGE_SHIFT) & ~(psize - 1));
728 if (hpte_is_writable(r) && !write_ok)
729 r = hpte_make_readonly(r);
732 while (!try_lock_hpte(hptep, HPTE_V_HVLOCK))
734 if ((hptep[0] & ~HPTE_V_HVLOCK) != hpte[0] || hptep[1] != hpte[1] ||
735 rev->guest_rpte != hpte[2])
736 /* HPTE has been changed under us; let the guest retry */
738 hpte[0] = (hpte[0] & ~HPTE_V_ABSENT) | HPTE_V_VALID;
740 /* Always put the HPTE in the rmap chain for the page base address */
741 rmap = &memslot->arch.rmap[gfn_base - memslot->base_gfn];
744 /* Check if we might have been invalidated; let the guest retry if so */
746 if (mmu_notifier_retry(vcpu->kvm, mmu_seq)) {
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);
755 if (hptep[0] & HPTE_V_VALID) {
756 /* HPTE was previously valid, so we need to invalidate it */
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);
763 kvmppc_add_revmap_chain(kvm, rev, rmap, index, 0);
769 asm volatile("ptesync" : : : "memory");
771 if (page && hpte_is_writable(r))
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()
787 hptep[0] &= ~HPTE_V_HVLOCK;
792 static void kvmppc_rmap_reset(struct kvm *kvm)
794 struct kvm_memslots *slots;
795 struct kvm_memory_slot *memslot;
798 srcu_idx = srcu_read_lock(&kvm->srcu);
799 slots = kvm->memslots;
800 kvm_for_each_memslot(memslot, slots) {
802 * This assumes it is acceptable to lose reference and
803 * change bits across a reset.
805 memset(memslot->arch.rmap, 0,
806 memslot->npages * sizeof(*memslot->arch.rmap));
808 srcu_read_unlock(&kvm->srcu, srcu_idx);
811 static int kvm_handle_hva_range(struct kvm *kvm,
814 int (*handler)(struct kvm *kvm,
815 unsigned long *rmapp,
820 struct kvm_memslots *slots;
821 struct kvm_memory_slot *memslot;
823 slots = kvm_memslots(kvm);
824 kvm_for_each_memslot(memslot, slots) {
825 unsigned long hva_start, hva_end;
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)
834 * {gfn(page) | page intersects with [hva_start, hva_end)} =
835 * {gfn, gfn+1, ..., gfn_end-1}.
837 gfn = hva_to_gfn_memslot(hva_start, memslot);
838 gfn_end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, memslot);
840 for (; gfn < gfn_end; ++gfn) {
841 gfn_t gfn_offset = gfn - memslot->base_gfn;
843 ret = handler(kvm, &memslot->arch.rmap[gfn_offset], gfn);
851 static int kvm_handle_hva(struct kvm *kvm, unsigned long hva,
852 int (*handler)(struct kvm *kvm, unsigned long *rmapp,
855 return kvm_handle_hva_range(kvm, hva, hva + 1, handler);
858 static int kvm_unmap_rmapp(struct kvm *kvm, unsigned long *rmapp,
861 struct revmap_entry *rev = kvm->arch.revmap;
862 unsigned long h, i, j;
863 unsigned long *hptep;
864 unsigned long ptel, psize, rcbits;
868 if (!(*rmapp & KVMPPC_RMAP_PRESENT)) {
874 * To avoid an ABBA deadlock with the HPTE lock bit,
875 * we can't spin on the HPTE lock while holding the
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 */
883 while (hptep[0] & HPTE_V_HVLOCK)
889 /* chain is now empty */
890 *rmapp &= ~(KVMPPC_RMAP_PRESENT | KVMPPC_RMAP_INDEX);
892 /* remove i from chain */
896 rev[i].forw = rev[i].back = i;
897 *rmapp = (*rmapp & ~KVMPPC_RMAP_INDEX) | j;
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]);
917 hptep[0] &= ~HPTE_V_HVLOCK;
922 int kvm_unmap_hva_hv(struct kvm *kvm, unsigned long hva)
924 if (kvm->arch.using_mmu_notifiers)
925 kvm_handle_hva(kvm, hva, kvm_unmap_rmapp);
929 int kvm_unmap_hva_range_hv(struct kvm *kvm, unsigned long start, unsigned long end)
931 if (kvm->arch.using_mmu_notifiers)
932 kvm_handle_hva_range(kvm, start, end, kvm_unmap_rmapp);
936 void kvmppc_core_flush_memslot_hv(struct kvm *kvm,
937 struct kvm_memory_slot *memslot)
939 unsigned long *rmapp;
943 rmapp = memslot->arch.rmap;
944 gfn = memslot->base_gfn;
945 for (n = memslot->npages; n; --n) {
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.
952 if (*rmapp & KVMPPC_RMAP_PRESENT)
953 kvm_unmap_rmapp(kvm, rmapp, gfn);
959 static int kvm_age_rmapp(struct kvm *kvm, unsigned long *rmapp,
962 struct revmap_entry *rev = kvm->arch.revmap;
963 unsigned long head, i, j;
964 unsigned long *hptep;
969 if (*rmapp & KVMPPC_RMAP_REFERENCED) {
970 *rmapp &= ~KVMPPC_RMAP_REFERENCED;
973 if (!(*rmapp & KVMPPC_RMAP_PRESENT)) {
978 i = head = *rmapp & KVMPPC_RMAP_INDEX;
980 hptep = (unsigned long *) (kvm->arch.hpt_virt + (i << 4));
983 /* If this HPTE isn't referenced, ignore it */
984 if (!(hptep[1] & HPTE_R_R))
987 if (!try_lock_hpte(hptep, HPTE_V_HVLOCK)) {
988 /* unlock rmap before spinning on the HPTE lock */
990 while (hptep[0] & HPTE_V_HVLOCK)
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]);
1004 hptep[0] &= ~HPTE_V_HVLOCK;
1005 } while ((i = j) != head);
1011 int kvm_age_hva_hv(struct kvm *kvm, unsigned long hva)
1013 if (!kvm->arch.using_mmu_notifiers)
1015 return kvm_handle_hva(kvm, hva, kvm_age_rmapp);
1018 static int kvm_test_age_rmapp(struct kvm *kvm, unsigned long *rmapp,
1021 struct revmap_entry *rev = kvm->arch.revmap;
1022 unsigned long head, i, j;
1026 if (*rmapp & KVMPPC_RMAP_REFERENCED)
1030 if (*rmapp & KVMPPC_RMAP_REFERENCED)
1033 if (*rmapp & KVMPPC_RMAP_PRESENT) {
1034 i = head = *rmapp & KVMPPC_RMAP_INDEX;
1036 hp = (unsigned long *)(kvm->arch.hpt_virt + (i << 4));
1038 if (hp[1] & HPTE_R_R)
1040 } while ((i = j) != head);
1049 int kvm_test_age_hva_hv(struct kvm *kvm, unsigned long hva)
1051 if (!kvm->arch.using_mmu_notifiers)
1053 return kvm_handle_hva(kvm, hva, kvm_test_age_rmapp);
1056 void kvm_set_spte_hva_hv(struct kvm *kvm, unsigned long hva, pte_t pte)
1058 if (!kvm->arch.using_mmu_notifiers)
1060 kvm_handle_hva(kvm, hva, kvm_unmap_rmapp);
1064 * Returns the number of system pages that are dirty.
1065 * This can be more than 1 if we find a huge-page HPTE.
1067 static int kvm_test_clear_dirty_npages(struct kvm *kvm, unsigned long *rmapp)
1069 struct revmap_entry *rev = kvm->arch.revmap;
1070 unsigned long head, i, j;
1072 unsigned long *hptep;
1073 int npages_dirty = 0;
1077 if (*rmapp & KVMPPC_RMAP_CHANGED) {
1078 *rmapp &= ~KVMPPC_RMAP_CHANGED;
1081 if (!(*rmapp & KVMPPC_RMAP_PRESENT)) {
1083 return npages_dirty;
1086 i = head = *rmapp & KVMPPC_RMAP_INDEX;
1088 hptep = (unsigned long *) (kvm->arch.hpt_virt + (i << 4));
1091 if (!(hptep[1] & HPTE_R_C))
1094 if (!try_lock_hpte(hptep, HPTE_V_HVLOCK)) {
1095 /* unlock rmap before spinning on the HPTE lock */
1097 while (hptep[0] & HPTE_V_HVLOCK)
1102 /* Now check and modify the HPTE */
1103 if ((hptep[0] & HPTE_V_VALID) && (hptep[1] & HPTE_R_C)) {
1104 /* need to make it temporarily absent to clear C */
1105 hptep[0] |= HPTE_V_ABSENT;
1106 kvmppc_invalidate_hpte(kvm, hptep, i);
1107 hptep[1] &= ~HPTE_R_C;
1109 hptep[0] = (hptep[0] & ~HPTE_V_ABSENT) | HPTE_V_VALID;
1110 if (!(rev[i].guest_rpte & HPTE_R_C)) {
1111 rev[i].guest_rpte |= HPTE_R_C;
1112 note_hpte_modification(kvm, &rev[i]);
1114 n = hpte_page_size(hptep[0], hptep[1]);
1115 n = (n + PAGE_SIZE - 1) >> PAGE_SHIFT;
1116 if (n > npages_dirty)
1119 hptep[0] &= ~HPTE_V_HVLOCK;
1120 } while ((i = j) != head);
1123 return npages_dirty;
1126 static void harvest_vpa_dirty(struct kvmppc_vpa *vpa,
1127 struct kvm_memory_slot *memslot,
1132 if (!vpa->dirty || !vpa->pinned_addr)
1134 gfn = vpa->gpa >> PAGE_SHIFT;
1135 if (gfn < memslot->base_gfn ||
1136 gfn >= memslot->base_gfn + memslot->npages)
1141 __set_bit_le(gfn - memslot->base_gfn, map);
1144 long kvmppc_hv_get_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot,
1148 unsigned long *rmapp;
1149 struct kvm_vcpu *vcpu;
1152 rmapp = memslot->arch.rmap;
1153 for (i = 0; i < memslot->npages; ++i) {
1154 int npages = kvm_test_clear_dirty_npages(kvm, rmapp);
1156 * Note that if npages > 0 then i must be a multiple of npages,
1157 * since we always put huge-page HPTEs in the rmap chain
1158 * corresponding to their page base address.
1161 for (j = i; npages; ++j, --npages)
1162 __set_bit_le(j, map);
1166 /* Harvest dirty bits from VPA and DTL updates */
1167 /* Note: we never modify the SLB shadow buffer areas */
1168 kvm_for_each_vcpu(i, vcpu, kvm) {
1169 spin_lock(&vcpu->arch.vpa_update_lock);
1170 harvest_vpa_dirty(&vcpu->arch.vpa, memslot, map);
1171 harvest_vpa_dirty(&vcpu->arch.dtl, memslot, map);
1172 spin_unlock(&vcpu->arch.vpa_update_lock);
1178 void *kvmppc_pin_guest_page(struct kvm *kvm, unsigned long gpa,
1179 unsigned long *nb_ret)
1181 struct kvm_memory_slot *memslot;
1182 unsigned long gfn = gpa >> PAGE_SHIFT;
1183 struct page *page, *pages[1];
1185 unsigned long hva, offset;
1187 unsigned long *physp;
1190 srcu_idx = srcu_read_lock(&kvm->srcu);
1191 memslot = gfn_to_memslot(kvm, gfn);
1192 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
1194 if (!kvm->arch.using_mmu_notifiers) {
1195 physp = memslot->arch.slot_phys;
1198 physp += gfn - memslot->base_gfn;
1201 if (kvmppc_get_guest_page(kvm, gfn, memslot,
1206 page = pfn_to_page(pa >> PAGE_SHIFT);
1209 hva = gfn_to_hva_memslot(memslot, gfn);
1210 npages = get_user_pages_fast(hva, 1, 1, pages);
1215 srcu_read_unlock(&kvm->srcu, srcu_idx);
1217 offset = gpa & (PAGE_SIZE - 1);
1219 *nb_ret = PAGE_SIZE - offset;
1220 return page_address(page) + offset;
1223 srcu_read_unlock(&kvm->srcu, srcu_idx);
1227 void kvmppc_unpin_guest_page(struct kvm *kvm, void *va, unsigned long gpa,
1230 struct page *page = virt_to_page(va);
1231 struct kvm_memory_slot *memslot;
1233 unsigned long *rmap;
1238 if (!dirty || !kvm->arch.using_mmu_notifiers)
1241 /* We need to mark this page dirty in the rmap chain */
1242 gfn = gpa >> PAGE_SHIFT;
1243 srcu_idx = srcu_read_lock(&kvm->srcu);
1244 memslot = gfn_to_memslot(kvm, gfn);
1246 rmap = &memslot->arch.rmap[gfn - memslot->base_gfn];
1248 *rmap |= KVMPPC_RMAP_CHANGED;
1251 srcu_read_unlock(&kvm->srcu, srcu_idx);
1255 * Functions for reading and writing the hash table via reads and
1256 * writes on a file descriptor.
1258 * Reads return the guest view of the hash table, which has to be
1259 * pieced together from the real hash table and the guest_rpte
1260 * values in the revmap array.
1262 * On writes, each HPTE written is considered in turn, and if it
1263 * is valid, it is written to the HPT as if an H_ENTER with the
1264 * exact flag set was done. When the invalid count is non-zero
1265 * in the header written to the stream, the kernel will make
1266 * sure that that many HPTEs are invalid, and invalidate them
1270 struct kvm_htab_ctx {
1271 unsigned long index;
1272 unsigned long flags;
1277 #define HPTE_SIZE (2 * sizeof(unsigned long))
1280 * Returns 1 if this HPT entry has been modified or has pending
1283 static int hpte_dirty(struct revmap_entry *revp, unsigned long *hptp)
1285 unsigned long rcbits_unset;
1287 if (revp->guest_rpte & HPTE_GR_MODIFIED)
1290 /* Also need to consider changes in reference and changed bits */
1291 rcbits_unset = ~revp->guest_rpte & (HPTE_R_R | HPTE_R_C);
1292 if ((hptp[0] & HPTE_V_VALID) && (hptp[1] & rcbits_unset))
1298 static long record_hpte(unsigned long flags, unsigned long *hptp,
1299 unsigned long *hpte, struct revmap_entry *revp,
1300 int want_valid, int first_pass)
1303 unsigned long rcbits_unset;
1307 /* Unmodified entries are uninteresting except on the first pass */
1308 dirty = hpte_dirty(revp, hptp);
1309 if (!first_pass && !dirty)
1313 if (hptp[0] & (HPTE_V_VALID | HPTE_V_ABSENT)) {
1315 if ((flags & KVM_GET_HTAB_BOLTED_ONLY) &&
1316 !(hptp[0] & HPTE_V_BOLTED))
1319 if (valid != want_valid)
1323 if (valid || dirty) {
1324 /* lock the HPTE so it's stable and read it */
1326 while (!try_lock_hpte(hptp, HPTE_V_HVLOCK))
1330 /* re-evaluate valid and dirty from synchronized HPTE value */
1331 valid = !!(v & HPTE_V_VALID);
1332 dirty = !!(revp->guest_rpte & HPTE_GR_MODIFIED);
1334 /* Harvest R and C into guest view if necessary */
1335 rcbits_unset = ~revp->guest_rpte & (HPTE_R_R | HPTE_R_C);
1336 if (valid && (rcbits_unset & hptp[1])) {
1337 revp->guest_rpte |= (hptp[1] & (HPTE_R_R | HPTE_R_C)) |
1342 if (v & HPTE_V_ABSENT) {
1343 v &= ~HPTE_V_ABSENT;
1347 if ((flags & KVM_GET_HTAB_BOLTED_ONLY) && !(v & HPTE_V_BOLTED))
1350 r = revp->guest_rpte;
1351 /* only clear modified if this is the right sort of entry */
1352 if (valid == want_valid && dirty) {
1353 r &= ~HPTE_GR_MODIFIED;
1354 revp->guest_rpte = r;
1356 asm volatile(PPC_RELEASE_BARRIER "" : : : "memory");
1357 hptp[0] &= ~HPTE_V_HVLOCK;
1359 if (!(valid == want_valid && (first_pass || dirty)))
1367 static ssize_t kvm_htab_read(struct file *file, char __user *buf,
1368 size_t count, loff_t *ppos)
1370 struct kvm_htab_ctx *ctx = file->private_data;
1371 struct kvm *kvm = ctx->kvm;
1372 struct kvm_get_htab_header hdr;
1373 unsigned long *hptp;
1374 struct revmap_entry *revp;
1375 unsigned long i, nb, nw;
1376 unsigned long __user *lbuf;
1377 struct kvm_get_htab_header __user *hptr;
1378 unsigned long flags;
1380 unsigned long hpte[2];
1382 if (!access_ok(VERIFY_WRITE, buf, count))
1385 first_pass = ctx->first_pass;
1389 hptp = (unsigned long *)(kvm->arch.hpt_virt + (i * HPTE_SIZE));
1390 revp = kvm->arch.revmap + i;
1391 lbuf = (unsigned long __user *)buf;
1394 while (nb + sizeof(hdr) + HPTE_SIZE < count) {
1395 /* Initialize header */
1396 hptr = (struct kvm_get_htab_header __user *)buf;
1401 lbuf = (unsigned long __user *)(buf + sizeof(hdr));
1403 /* Skip uninteresting entries, i.e. clean on not-first pass */
1405 while (i < kvm->arch.hpt_npte &&
1406 !hpte_dirty(revp, hptp)) {
1414 /* Grab a series of valid entries */
1415 while (i < kvm->arch.hpt_npte &&
1416 hdr.n_valid < 0xffff &&
1417 nb + HPTE_SIZE < count &&
1418 record_hpte(flags, hptp, hpte, revp, 1, first_pass)) {
1419 /* valid entry, write it out */
1421 if (__put_user(hpte[0], lbuf) ||
1422 __put_user(hpte[1], lbuf + 1))
1430 /* Now skip invalid entries while we can */
1431 while (i < kvm->arch.hpt_npte &&
1432 hdr.n_invalid < 0xffff &&
1433 record_hpte(flags, hptp, hpte, revp, 0, first_pass)) {
1434 /* found an invalid entry */
1441 if (hdr.n_valid || hdr.n_invalid) {
1442 /* write back the header */
1443 if (__copy_to_user(hptr, &hdr, sizeof(hdr)))
1446 buf = (char __user *)lbuf;
1451 /* Check if we've wrapped around the hash table */
1452 if (i >= kvm->arch.hpt_npte) {
1454 ctx->first_pass = 0;
1464 static ssize_t kvm_htab_write(struct file *file, const char __user *buf,
1465 size_t count, loff_t *ppos)
1467 struct kvm_htab_ctx *ctx = file->private_data;
1468 struct kvm *kvm = ctx->kvm;
1469 struct kvm_get_htab_header hdr;
1472 unsigned long __user *lbuf;
1473 unsigned long *hptp;
1474 unsigned long tmp[2];
1479 if (!access_ok(VERIFY_READ, buf, count))
1482 /* lock out vcpus from running while we're doing this */
1483 mutex_lock(&kvm->lock);
1484 rma_setup = kvm->arch.rma_setup_done;
1486 kvm->arch.rma_setup_done = 0; /* temporarily */
1487 /* order rma_setup_done vs. vcpus_running */
1489 if (atomic_read(&kvm->arch.vcpus_running)) {
1490 kvm->arch.rma_setup_done = 1;
1491 mutex_unlock(&kvm->lock);
1497 for (nb = 0; nb + sizeof(hdr) <= count; ) {
1499 if (__copy_from_user(&hdr, buf, sizeof(hdr)))
1503 if (nb + hdr.n_valid * HPTE_SIZE > count)
1511 if (i >= kvm->arch.hpt_npte ||
1512 i + hdr.n_valid + hdr.n_invalid > kvm->arch.hpt_npte)
1515 hptp = (unsigned long *)(kvm->arch.hpt_virt + (i * HPTE_SIZE));
1516 lbuf = (unsigned long __user *)buf;
1517 for (j = 0; j < hdr.n_valid; ++j) {
1519 if (__get_user(v, lbuf) || __get_user(r, lbuf + 1))
1522 if (!(v & HPTE_V_VALID))
1527 if (hptp[0] & (HPTE_V_VALID | HPTE_V_ABSENT))
1528 kvmppc_do_h_remove(kvm, 0, i, 0, tmp);
1530 ret = kvmppc_virtmode_do_h_enter(kvm, H_EXACT, i, v, r,
1532 if (ret != H_SUCCESS) {
1533 pr_err("kvm_htab_write ret %ld i=%ld v=%lx "
1534 "r=%lx\n", ret, i, v, r);
1537 if (!rma_setup && is_vrma_hpte(v)) {
1538 unsigned long psize = hpte_page_size(v, r);
1539 unsigned long senc = slb_pgsize_encoding(psize);
1542 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
1543 (VRMA_VSID << SLB_VSID_SHIFT_1T);
1544 lpcr = senc << (LPCR_VRMASD_SH - 4);
1545 kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
1552 for (j = 0; j < hdr.n_invalid; ++j) {
1553 if (hptp[0] & (HPTE_V_VALID | HPTE_V_ABSENT))
1554 kvmppc_do_h_remove(kvm, 0, i, 0, tmp);
1562 /* Order HPTE updates vs. rma_setup_done */
1564 kvm->arch.rma_setup_done = rma_setup;
1565 mutex_unlock(&kvm->lock);
1572 static int kvm_htab_release(struct inode *inode, struct file *filp)
1574 struct kvm_htab_ctx *ctx = filp->private_data;
1576 filp->private_data = NULL;
1577 if (!(ctx->flags & KVM_GET_HTAB_WRITE))
1578 atomic_dec(&ctx->kvm->arch.hpte_mod_interest);
1579 kvm_put_kvm(ctx->kvm);
1584 static const struct file_operations kvm_htab_fops = {
1585 .read = kvm_htab_read,
1586 .write = kvm_htab_write,
1587 .llseek = default_llseek,
1588 .release = kvm_htab_release,
1591 int kvm_vm_ioctl_get_htab_fd(struct kvm *kvm, struct kvm_get_htab_fd *ghf)
1594 struct kvm_htab_ctx *ctx;
1597 /* reject flags we don't recognize */
1598 if (ghf->flags & ~(KVM_GET_HTAB_BOLTED_ONLY | KVM_GET_HTAB_WRITE))
1600 ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
1605 ctx->index = ghf->start_index;
1606 ctx->flags = ghf->flags;
1607 ctx->first_pass = 1;
1609 rwflag = (ghf->flags & KVM_GET_HTAB_WRITE) ? O_WRONLY : O_RDONLY;
1610 ret = anon_inode_getfd("kvm-htab", &kvm_htab_fops, ctx, rwflag | O_CLOEXEC);
1616 if (rwflag == O_RDONLY) {
1617 mutex_lock(&kvm->slots_lock);
1618 atomic_inc(&kvm->arch.hpte_mod_interest);
1619 /* make sure kvmppc_do_h_enter etc. see the increment */
1620 synchronize_srcu_expedited(&kvm->srcu);
1621 mutex_unlock(&kvm->slots_lock);
1627 void kvmppc_mmu_book3s_hv_init(struct kvm_vcpu *vcpu)
1629 struct kvmppc_mmu *mmu = &vcpu->arch.mmu;
1631 if (cpu_has_feature(CPU_FTR_ARCH_206))
1632 vcpu->arch.slb_nr = 32; /* POWER7 */
1634 vcpu->arch.slb_nr = 64;
1636 mmu->xlate = kvmppc_mmu_book3s_64_hv_xlate;
1637 mmu->reset_msr = kvmppc_mmu_book3s_64_hv_reset_msr;
1639 vcpu->arch.hflags |= BOOK3S_HFLAG_SLB;