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>
42 /* Power architecture requires HPT is at least 256kB */
43 #define PPC_MIN_HPT_ORDER 18
45 static long kvmppc_virtmode_do_h_enter(struct kvm *kvm, unsigned long flags,
46 long pte_index, unsigned long pteh,
47 unsigned long ptel, unsigned long *pte_idx_ret);
48 static void kvmppc_rmap_reset(struct kvm *kvm);
50 long kvmppc_alloc_hpt(struct kvm *kvm, u32 *htab_orderp)
52 unsigned long hpt = 0;
53 struct revmap_entry *rev;
54 struct page *page = NULL;
55 long order = KVM_DEFAULT_HPT_ORDER;
59 if (order < PPC_MIN_HPT_ORDER)
60 order = PPC_MIN_HPT_ORDER;
63 kvm->arch.hpt_cma_alloc = 0;
64 page = kvm_alloc_hpt(1ul << (order - PAGE_SHIFT));
66 hpt = (unsigned long)pfn_to_kaddr(page_to_pfn(page));
67 memset((void *)hpt, 0, (1ul << order));
68 kvm->arch.hpt_cma_alloc = 1;
71 /* Lastly try successively smaller sizes from the page allocator */
72 while (!hpt && order > PPC_MIN_HPT_ORDER) {
73 hpt = __get_free_pages(GFP_KERNEL|__GFP_ZERO|__GFP_REPEAT|
74 __GFP_NOWARN, order - PAGE_SHIFT);
82 kvm->arch.hpt_virt = hpt;
83 kvm->arch.hpt_order = order;
84 /* HPTEs are 2**4 bytes long */
85 kvm->arch.hpt_npte = 1ul << (order - 4);
86 /* 128 (2**7) bytes in each HPTEG */
87 kvm->arch.hpt_mask = (1ul << (order - 7)) - 1;
89 /* Allocate reverse map array */
90 rev = vmalloc(sizeof(struct revmap_entry) * kvm->arch.hpt_npte);
92 pr_err("kvmppc_alloc_hpt: Couldn't alloc reverse map array\n");
95 kvm->arch.revmap = rev;
96 kvm->arch.sdr1 = __pa(hpt) | (order - 18);
98 pr_info("KVM guest htab at %lx (order %ld), LPID %x\n",
99 hpt, order, kvm->arch.lpid);
102 *htab_orderp = order;
106 if (kvm->arch.hpt_cma_alloc)
107 kvm_release_hpt(page, 1 << (order - PAGE_SHIFT));
109 free_pages(hpt, order - PAGE_SHIFT);
113 long kvmppc_alloc_reset_hpt(struct kvm *kvm, u32 *htab_orderp)
118 mutex_lock(&kvm->lock);
119 if (kvm->arch.rma_setup_done) {
120 kvm->arch.rma_setup_done = 0;
121 /* order rma_setup_done vs. vcpus_running */
123 if (atomic_read(&kvm->arch.vcpus_running)) {
124 kvm->arch.rma_setup_done = 1;
128 if (kvm->arch.hpt_virt) {
129 order = kvm->arch.hpt_order;
130 /* Set the entire HPT to 0, i.e. invalid HPTEs */
131 memset((void *)kvm->arch.hpt_virt, 0, 1ul << order);
133 * Reset all the reverse-mapping chains for all memslots
135 kvmppc_rmap_reset(kvm);
136 /* Ensure that each vcpu will flush its TLB on next entry. */
137 cpumask_setall(&kvm->arch.need_tlb_flush);
138 *htab_orderp = order;
141 err = kvmppc_alloc_hpt(kvm, htab_orderp);
142 order = *htab_orderp;
145 mutex_unlock(&kvm->lock);
149 void kvmppc_free_hpt(struct kvm *kvm)
151 kvmppc_free_lpid(kvm->arch.lpid);
152 vfree(kvm->arch.revmap);
153 if (kvm->arch.hpt_cma_alloc)
154 kvm_release_hpt(virt_to_page(kvm->arch.hpt_virt),
155 1 << (kvm->arch.hpt_order - PAGE_SHIFT));
157 free_pages(kvm->arch.hpt_virt,
158 kvm->arch.hpt_order - PAGE_SHIFT);
161 /* Bits in first HPTE dword for pagesize 4k, 64k or 16M */
162 static inline unsigned long hpte0_pgsize_encoding(unsigned long pgsize)
164 return (pgsize > 0x1000) ? HPTE_V_LARGE : 0;
167 /* Bits in second HPTE dword for pagesize 4k, 64k or 16M */
168 static inline unsigned long hpte1_pgsize_encoding(unsigned long pgsize)
170 return (pgsize == 0x10000) ? 0x1000 : 0;
173 void kvmppc_map_vrma(struct kvm_vcpu *vcpu, struct kvm_memory_slot *memslot,
174 unsigned long porder)
177 unsigned long npages;
178 unsigned long hp_v, hp_r;
179 unsigned long addr, hash;
181 unsigned long hp0, hp1;
182 unsigned long idx_ret;
184 struct kvm *kvm = vcpu->kvm;
186 psize = 1ul << porder;
187 npages = memslot->npages >> (porder - PAGE_SHIFT);
189 /* VRMA can't be > 1TB */
190 if (npages > 1ul << (40 - porder))
191 npages = 1ul << (40 - porder);
192 /* Can't use more than 1 HPTE per HPTEG */
193 if (npages > kvm->arch.hpt_mask + 1)
194 npages = kvm->arch.hpt_mask + 1;
196 hp0 = HPTE_V_1TB_SEG | (VRMA_VSID << (40 - 16)) |
197 HPTE_V_BOLTED | hpte0_pgsize_encoding(psize);
198 hp1 = hpte1_pgsize_encoding(psize) |
199 HPTE_R_R | HPTE_R_C | HPTE_R_M | PP_RWXX;
201 for (i = 0; i < npages; ++i) {
203 /* can't use hpt_hash since va > 64 bits */
204 hash = (i ^ (VRMA_VSID ^ (VRMA_VSID << 25))) & kvm->arch.hpt_mask;
206 * We assume that the hash table is empty and no
207 * vcpus are using it at this stage. Since we create
208 * at most one HPTE per HPTEG, we just assume entry 7
209 * is available and use it.
211 hash = (hash << 3) + 7;
212 hp_v = hp0 | ((addr >> 16) & ~0x7fUL);
214 ret = kvmppc_virtmode_do_h_enter(kvm, H_EXACT, hash, hp_v, hp_r,
216 if (ret != H_SUCCESS) {
217 pr_err("KVM: map_vrma at %lx failed, ret=%ld\n",
224 int kvmppc_mmu_hv_init(void)
226 unsigned long host_lpid, rsvd_lpid;
228 if (!cpu_has_feature(CPU_FTR_HVMODE))
231 /* POWER7 has 10-bit LPIDs (12-bit in POWER8) */
232 host_lpid = mfspr(SPRN_LPID);
233 rsvd_lpid = LPID_RSVD;
235 kvmppc_init_lpid(rsvd_lpid + 1);
237 kvmppc_claim_lpid(host_lpid);
238 /* rsvd_lpid is reserved for use in partition switching */
239 kvmppc_claim_lpid(rsvd_lpid);
244 static void kvmppc_mmu_book3s_64_hv_reset_msr(struct kvm_vcpu *vcpu)
246 unsigned long msr = vcpu->arch.intr_msr;
248 /* If transactional, change to suspend mode on IRQ delivery */
249 if (MSR_TM_TRANSACTIONAL(vcpu->arch.shregs.msr))
252 msr |= vcpu->arch.shregs.msr & MSR_TS_MASK;
253 kvmppc_set_msr(vcpu, msr);
256 long kvmppc_virtmode_do_h_enter(struct kvm *kvm, unsigned long flags,
257 long pte_index, unsigned long pteh,
258 unsigned long ptel, unsigned long *pte_idx_ret)
262 /* Protect linux PTE lookup from page table destruction */
263 rcu_read_lock_sched(); /* this disables preemption too */
264 ret = kvmppc_do_h_enter(kvm, flags, pte_index, pteh, ptel,
265 current->mm->pgd, false, pte_idx_ret);
266 rcu_read_unlock_sched();
267 if (ret == H_TOO_HARD) {
268 /* this can't happen */
269 pr_err("KVM: Oops, kvmppc_h_enter returned too hard!\n");
270 ret = H_RESOURCE; /* or something */
276 static struct kvmppc_slb *kvmppc_mmu_book3s_hv_find_slbe(struct kvm_vcpu *vcpu,
282 for (i = 0; i < vcpu->arch.slb_nr; i++) {
283 if (!(vcpu->arch.slb[i].orige & SLB_ESID_V))
286 if (vcpu->arch.slb[i].origv & SLB_VSID_B_1T)
291 if (((vcpu->arch.slb[i].orige ^ eaddr) & mask) == 0)
292 return &vcpu->arch.slb[i];
297 static unsigned long kvmppc_mmu_get_real_addr(unsigned long v, unsigned long r,
300 unsigned long ra_mask;
302 ra_mask = hpte_page_size(v, r) - 1;
303 return (r & HPTE_R_RPN & ~ra_mask) | (ea & ra_mask);
306 static int kvmppc_mmu_book3s_64_hv_xlate(struct kvm_vcpu *vcpu, gva_t eaddr,
307 struct kvmppc_pte *gpte, bool data, bool iswrite)
309 struct kvm *kvm = vcpu->kvm;
310 struct kvmppc_slb *slbe;
312 unsigned long pp, key;
316 int virtmode = vcpu->arch.shregs.msr & (data ? MSR_DR : MSR_IR);
320 slbe = kvmppc_mmu_book3s_hv_find_slbe(vcpu, eaddr);
325 /* real mode access */
326 slb_v = vcpu->kvm->arch.vrma_slb_v;
330 /* Find the HPTE in the hash table */
331 index = kvmppc_hv_find_lock_hpte(kvm, eaddr, slb_v,
332 HPTE_V_VALID | HPTE_V_ABSENT);
337 hptep = (__be64 *)(kvm->arch.hpt_virt + (index << 4));
338 v = be64_to_cpu(hptep[0]) & ~HPTE_V_HVLOCK;
339 gr = kvm->arch.revmap[index].guest_rpte;
341 /* Unlock the HPTE */
342 asm volatile("lwsync" : : : "memory");
343 hptep[0] = cpu_to_be64(v);
347 gpte->vpage = ((v & HPTE_V_AVPN) << 4) | ((eaddr >> 12) & 0xfff);
349 /* Get PP bits and key for permission check */
350 pp = gr & (HPTE_R_PP0 | HPTE_R_PP);
351 key = (vcpu->arch.shregs.msr & MSR_PR) ? SLB_VSID_KP : SLB_VSID_KS;
354 /* Calculate permissions */
355 gpte->may_read = hpte_read_permission(pp, key);
356 gpte->may_write = hpte_write_permission(pp, key);
357 gpte->may_execute = gpte->may_read && !(gr & (HPTE_R_N | HPTE_R_G));
359 /* Storage key permission check for POWER7 */
360 if (data && virtmode) {
361 int amrfield = hpte_get_skey_perm(gr, vcpu->arch.amr);
368 /* Get the guest physical address */
369 gpte->raddr = kvmppc_mmu_get_real_addr(v, gr, eaddr);
374 * Quick test for whether an instruction is a load or a store.
375 * If the instruction is a load or a store, then this will indicate
376 * which it is, at least on server processors. (Embedded processors
377 * have some external PID instructions that don't follow the rule
378 * embodied here.) If the instruction isn't a load or store, then
379 * this doesn't return anything useful.
381 static int instruction_is_store(unsigned int instr)
386 if ((instr & 0xfc000000) == 0x7c000000)
387 mask = 0x100; /* major opcode 31 */
388 return (instr & mask) != 0;
391 static int kvmppc_hv_emulate_mmio(struct kvm_run *run, struct kvm_vcpu *vcpu,
392 unsigned long gpa, gva_t ea, int is_store)
397 * If we fail, we just return to the guest and try executing it again.
399 if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
404 * WARNING: We do not know for sure whether the instruction we just
405 * read from memory is the same that caused the fault in the first
406 * place. If the instruction we read is neither an load or a store,
407 * then it can't access memory, so we don't need to worry about
408 * enforcing access permissions. So, assuming it is a load or
409 * store, we just check that its direction (load or store) is
410 * consistent with the original fault, since that's what we
411 * checked the access permissions against. If there is a mismatch
412 * we just return and retry the instruction.
415 if (instruction_is_store(last_inst) != !!is_store)
419 * Emulated accesses are emulated by looking at the hash for
420 * translation once, then performing the access later. The
421 * translation could be invalidated in the meantime in which
422 * point performing the subsequent memory access on the old
423 * physical address could possibly be a security hole for the
424 * guest (but not the host).
426 * This is less of an issue for MMIO stores since they aren't
427 * globally visible. It could be an issue for MMIO loads to
428 * a certain extent but we'll ignore it for now.
431 vcpu->arch.paddr_accessed = gpa;
432 vcpu->arch.vaddr_accessed = ea;
433 return kvmppc_emulate_mmio(run, vcpu);
436 int kvmppc_book3s_hv_page_fault(struct kvm_run *run, struct kvm_vcpu *vcpu,
437 unsigned long ea, unsigned long dsisr)
439 struct kvm *kvm = vcpu->kvm;
440 unsigned long hpte[3], r;
442 unsigned long mmu_seq, psize, pte_size;
443 unsigned long gpa_base, gfn_base;
444 unsigned long gpa, gfn, hva, pfn;
445 struct kvm_memory_slot *memslot;
447 struct revmap_entry *rev;
448 struct page *page, *pages[1];
449 long index, ret, npages;
451 unsigned int writing, write_ok;
452 struct vm_area_struct *vma;
453 unsigned long rcbits;
456 * Real-mode code has already searched the HPT and found the
457 * entry we're interested in. Lock the entry and check that
458 * it hasn't changed. If it has, just return and re-execute the
461 if (ea != vcpu->arch.pgfault_addr)
463 index = vcpu->arch.pgfault_index;
464 hptep = (__be64 *)(kvm->arch.hpt_virt + (index << 4));
465 rev = &kvm->arch.revmap[index];
467 while (!try_lock_hpte(hptep, HPTE_V_HVLOCK))
469 hpte[0] = be64_to_cpu(hptep[0]) & ~HPTE_V_HVLOCK;
470 hpte[1] = be64_to_cpu(hptep[1]);
471 hpte[2] = r = rev->guest_rpte;
472 asm volatile("lwsync" : : : "memory");
473 hptep[0] = cpu_to_be64(hpte[0]);
476 if (hpte[0] != vcpu->arch.pgfault_hpte[0] ||
477 hpte[1] != vcpu->arch.pgfault_hpte[1])
480 /* Translate the logical address and get the page */
481 psize = hpte_page_size(hpte[0], r);
482 gpa_base = r & HPTE_R_RPN & ~(psize - 1);
483 gfn_base = gpa_base >> PAGE_SHIFT;
484 gpa = gpa_base | (ea & (psize - 1));
485 gfn = gpa >> PAGE_SHIFT;
486 memslot = gfn_to_memslot(kvm, gfn);
488 trace_kvm_page_fault_enter(vcpu, hpte, memslot, ea, dsisr);
490 /* No memslot means it's an emulated MMIO region */
491 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
492 return kvmppc_hv_emulate_mmio(run, vcpu, gpa, ea,
493 dsisr & DSISR_ISSTORE);
496 * This should never happen, because of the slot_is_aligned()
497 * check in kvmppc_do_h_enter().
499 if (gfn_base < memslot->base_gfn)
502 /* used to check for invalidations in progress */
503 mmu_seq = kvm->mmu_notifier_seq;
510 pte_size = PAGE_SIZE;
511 writing = (dsisr & DSISR_ISSTORE) != 0;
512 /* If writing != 0, then the HPTE must allow writing, if we get here */
514 hva = gfn_to_hva_memslot(memslot, gfn);
515 npages = get_user_pages_fast(hva, 1, writing, pages);
517 /* Check if it's an I/O mapping */
518 down_read(¤t->mm->mmap_sem);
519 vma = find_vma(current->mm, hva);
520 if (vma && vma->vm_start <= hva && hva + psize <= vma->vm_end &&
521 (vma->vm_flags & VM_PFNMAP)) {
522 pfn = vma->vm_pgoff +
523 ((hva - vma->vm_start) >> PAGE_SHIFT);
525 is_io = hpte_cache_bits(pgprot_val(vma->vm_page_prot));
526 write_ok = vma->vm_flags & VM_WRITE;
528 up_read(¤t->mm->mmap_sem);
533 pfn = page_to_pfn(page);
534 if (PageHuge(page)) {
535 page = compound_head(page);
536 pte_size <<= compound_order(page);
538 /* if the guest wants write access, see if that is OK */
539 if (!writing && hpte_is_writable(r)) {
540 unsigned int hugepage_shift;
544 * We need to protect against page table destruction
545 * while looking up and updating the pte.
547 rcu_read_lock_sched();
548 ptep = find_linux_pte_or_hugepte(current->mm->pgd,
549 hva, &hugepage_shift);
551 pte = kvmppc_read_update_linux_pte(ptep, 1,
556 rcu_read_unlock_sched();
560 if (psize > pte_size)
563 /* Check WIMG vs. the actual page we're accessing */
564 if (!hpte_cache_flags_ok(r, is_io)) {
569 * Allow guest to map emulated device memory as
570 * uncacheable, but actually make it cacheable.
572 r = (r & ~(HPTE_R_W|HPTE_R_I|HPTE_R_G)) | HPTE_R_M;
576 * Set the HPTE to point to pfn.
577 * Since the pfn is at PAGE_SIZE granularity, make sure we
578 * don't mask out lower-order bits if psize < PAGE_SIZE.
580 if (psize < PAGE_SIZE)
582 r = (r & ~(HPTE_R_PP0 - psize)) | ((pfn << PAGE_SHIFT) & ~(psize - 1));
583 if (hpte_is_writable(r) && !write_ok)
584 r = hpte_make_readonly(r);
587 while (!try_lock_hpte(hptep, HPTE_V_HVLOCK))
589 if ((be64_to_cpu(hptep[0]) & ~HPTE_V_HVLOCK) != hpte[0] ||
590 be64_to_cpu(hptep[1]) != hpte[1] ||
591 rev->guest_rpte != hpte[2])
592 /* HPTE has been changed under us; let the guest retry */
594 hpte[0] = (hpte[0] & ~HPTE_V_ABSENT) | HPTE_V_VALID;
596 /* Always put the HPTE in the rmap chain for the page base address */
597 rmap = &memslot->arch.rmap[gfn_base - memslot->base_gfn];
600 /* Check if we might have been invalidated; let the guest retry if so */
602 if (mmu_notifier_retry(vcpu->kvm, mmu_seq)) {
607 /* Only set R/C in real HPTE if set in both *rmap and guest_rpte */
608 rcbits = *rmap >> KVMPPC_RMAP_RC_SHIFT;
609 r &= rcbits | ~(HPTE_R_R | HPTE_R_C);
611 if (be64_to_cpu(hptep[0]) & HPTE_V_VALID) {
612 /* HPTE was previously valid, so we need to invalidate it */
614 hptep[0] |= cpu_to_be64(HPTE_V_ABSENT);
615 kvmppc_invalidate_hpte(kvm, hptep, index);
616 /* don't lose previous R and C bits */
617 r |= be64_to_cpu(hptep[1]) & (HPTE_R_R | HPTE_R_C);
619 kvmppc_add_revmap_chain(kvm, rev, rmap, index, 0);
622 hptep[1] = cpu_to_be64(r);
624 hptep[0] = cpu_to_be64(hpte[0]);
625 asm volatile("ptesync" : : : "memory");
627 if (page && hpte_is_writable(r))
631 trace_kvm_page_fault_exit(vcpu, hpte, ret);
635 * We drop pages[0] here, not page because page might
636 * have been set to the head page of a compound, but
637 * we have to drop the reference on the correct tail
638 * page to match the get inside gup()
645 hptep[0] &= ~cpu_to_be64(HPTE_V_HVLOCK);
650 static void kvmppc_rmap_reset(struct kvm *kvm)
652 struct kvm_memslots *slots;
653 struct kvm_memory_slot *memslot;
656 srcu_idx = srcu_read_lock(&kvm->srcu);
657 slots = kvm->memslots;
658 kvm_for_each_memslot(memslot, slots) {
660 * This assumes it is acceptable to lose reference and
661 * change bits across a reset.
663 memset(memslot->arch.rmap, 0,
664 memslot->npages * sizeof(*memslot->arch.rmap));
666 srcu_read_unlock(&kvm->srcu, srcu_idx);
669 static int kvm_handle_hva_range(struct kvm *kvm,
672 int (*handler)(struct kvm *kvm,
673 unsigned long *rmapp,
678 struct kvm_memslots *slots;
679 struct kvm_memory_slot *memslot;
681 slots = kvm_memslots(kvm);
682 kvm_for_each_memslot(memslot, slots) {
683 unsigned long hva_start, hva_end;
686 hva_start = max(start, memslot->userspace_addr);
687 hva_end = min(end, memslot->userspace_addr +
688 (memslot->npages << PAGE_SHIFT));
689 if (hva_start >= hva_end)
692 * {gfn(page) | page intersects with [hva_start, hva_end)} =
693 * {gfn, gfn+1, ..., gfn_end-1}.
695 gfn = hva_to_gfn_memslot(hva_start, memslot);
696 gfn_end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, memslot);
698 for (; gfn < gfn_end; ++gfn) {
699 gfn_t gfn_offset = gfn - memslot->base_gfn;
701 ret = handler(kvm, &memslot->arch.rmap[gfn_offset], gfn);
709 static int kvm_handle_hva(struct kvm *kvm, unsigned long hva,
710 int (*handler)(struct kvm *kvm, unsigned long *rmapp,
713 return kvm_handle_hva_range(kvm, hva, hva + 1, handler);
716 static int kvm_unmap_rmapp(struct kvm *kvm, unsigned long *rmapp,
719 struct revmap_entry *rev = kvm->arch.revmap;
720 unsigned long h, i, j;
722 unsigned long ptel, psize, rcbits;
726 if (!(*rmapp & KVMPPC_RMAP_PRESENT)) {
732 * To avoid an ABBA deadlock with the HPTE lock bit,
733 * we can't spin on the HPTE lock while holding the
736 i = *rmapp & KVMPPC_RMAP_INDEX;
737 hptep = (__be64 *) (kvm->arch.hpt_virt + (i << 4));
738 if (!try_lock_hpte(hptep, HPTE_V_HVLOCK)) {
739 /* unlock rmap before spinning on the HPTE lock */
741 while (be64_to_cpu(hptep[0]) & HPTE_V_HVLOCK)
747 /* chain is now empty */
748 *rmapp &= ~(KVMPPC_RMAP_PRESENT | KVMPPC_RMAP_INDEX);
750 /* remove i from chain */
754 rev[i].forw = rev[i].back = i;
755 *rmapp = (*rmapp & ~KVMPPC_RMAP_INDEX) | j;
758 /* Now check and modify the HPTE */
759 ptel = rev[i].guest_rpte;
760 psize = hpte_page_size(be64_to_cpu(hptep[0]), ptel);
761 if ((be64_to_cpu(hptep[0]) & HPTE_V_VALID) &&
762 hpte_rpn(ptel, psize) == gfn) {
763 hptep[0] |= cpu_to_be64(HPTE_V_ABSENT);
764 kvmppc_invalidate_hpte(kvm, hptep, i);
765 /* Harvest R and C */
766 rcbits = be64_to_cpu(hptep[1]) & (HPTE_R_R | HPTE_R_C);
767 *rmapp |= rcbits << KVMPPC_RMAP_RC_SHIFT;
768 if (rcbits & ~rev[i].guest_rpte) {
769 rev[i].guest_rpte = ptel | rcbits;
770 note_hpte_modification(kvm, &rev[i]);
774 hptep[0] &= ~cpu_to_be64(HPTE_V_HVLOCK);
779 int kvm_unmap_hva_hv(struct kvm *kvm, unsigned long hva)
781 kvm_handle_hva(kvm, hva, kvm_unmap_rmapp);
785 int kvm_unmap_hva_range_hv(struct kvm *kvm, unsigned long start, unsigned long end)
787 kvm_handle_hva_range(kvm, start, end, kvm_unmap_rmapp);
791 void kvmppc_core_flush_memslot_hv(struct kvm *kvm,
792 struct kvm_memory_slot *memslot)
794 unsigned long *rmapp;
798 rmapp = memslot->arch.rmap;
799 gfn = memslot->base_gfn;
800 for (n = memslot->npages; n; --n) {
802 * Testing the present bit without locking is OK because
803 * the memslot has been marked invalid already, and hence
804 * no new HPTEs referencing this page can be created,
805 * thus the present bit can't go from 0 to 1.
807 if (*rmapp & KVMPPC_RMAP_PRESENT)
808 kvm_unmap_rmapp(kvm, rmapp, gfn);
814 static int kvm_age_rmapp(struct kvm *kvm, unsigned long *rmapp,
817 struct revmap_entry *rev = kvm->arch.revmap;
818 unsigned long head, i, j;
824 if (*rmapp & KVMPPC_RMAP_REFERENCED) {
825 *rmapp &= ~KVMPPC_RMAP_REFERENCED;
828 if (!(*rmapp & KVMPPC_RMAP_PRESENT)) {
833 i = head = *rmapp & KVMPPC_RMAP_INDEX;
835 hptep = (__be64 *) (kvm->arch.hpt_virt + (i << 4));
838 /* If this HPTE isn't referenced, ignore it */
839 if (!(be64_to_cpu(hptep[1]) & HPTE_R_R))
842 if (!try_lock_hpte(hptep, HPTE_V_HVLOCK)) {
843 /* unlock rmap before spinning on the HPTE lock */
845 while (be64_to_cpu(hptep[0]) & HPTE_V_HVLOCK)
850 /* Now check and modify the HPTE */
851 if ((be64_to_cpu(hptep[0]) & HPTE_V_VALID) &&
852 (be64_to_cpu(hptep[1]) & HPTE_R_R)) {
853 kvmppc_clear_ref_hpte(kvm, hptep, i);
854 if (!(rev[i].guest_rpte & HPTE_R_R)) {
855 rev[i].guest_rpte |= HPTE_R_R;
856 note_hpte_modification(kvm, &rev[i]);
860 hptep[0] &= ~cpu_to_be64(HPTE_V_HVLOCK);
861 } while ((i = j) != head);
867 int kvm_age_hva_hv(struct kvm *kvm, unsigned long start, unsigned long end)
869 return kvm_handle_hva_range(kvm, start, end, kvm_age_rmapp);
872 static int kvm_test_age_rmapp(struct kvm *kvm, unsigned long *rmapp,
875 struct revmap_entry *rev = kvm->arch.revmap;
876 unsigned long head, i, j;
880 if (*rmapp & KVMPPC_RMAP_REFERENCED)
884 if (*rmapp & KVMPPC_RMAP_REFERENCED)
887 if (*rmapp & KVMPPC_RMAP_PRESENT) {
888 i = head = *rmapp & KVMPPC_RMAP_INDEX;
890 hp = (unsigned long *)(kvm->arch.hpt_virt + (i << 4));
892 if (be64_to_cpu(hp[1]) & HPTE_R_R)
894 } while ((i = j) != head);
903 int kvm_test_age_hva_hv(struct kvm *kvm, unsigned long hva)
905 return kvm_handle_hva(kvm, hva, kvm_test_age_rmapp);
908 void kvm_set_spte_hva_hv(struct kvm *kvm, unsigned long hva, pte_t pte)
910 kvm_handle_hva(kvm, hva, kvm_unmap_rmapp);
913 static int vcpus_running(struct kvm *kvm)
915 return atomic_read(&kvm->arch.vcpus_running) != 0;
919 * Returns the number of system pages that are dirty.
920 * This can be more than 1 if we find a huge-page HPTE.
922 static int kvm_test_clear_dirty_npages(struct kvm *kvm, unsigned long *rmapp)
924 struct revmap_entry *rev = kvm->arch.revmap;
925 unsigned long head, i, j;
929 int npages_dirty = 0;
933 if (*rmapp & KVMPPC_RMAP_CHANGED) {
934 *rmapp &= ~KVMPPC_RMAP_CHANGED;
937 if (!(*rmapp & KVMPPC_RMAP_PRESENT)) {
942 i = head = *rmapp & KVMPPC_RMAP_INDEX;
944 unsigned long hptep1;
945 hptep = (__be64 *) (kvm->arch.hpt_virt + (i << 4));
949 * Checking the C (changed) bit here is racy since there
950 * is no guarantee about when the hardware writes it back.
951 * If the HPTE is not writable then it is stable since the
952 * page can't be written to, and we would have done a tlbie
953 * (which forces the hardware to complete any writeback)
954 * when making the HPTE read-only.
955 * If vcpus are running then this call is racy anyway
956 * since the page could get dirtied subsequently, so we
957 * expect there to be a further call which would pick up
958 * any delayed C bit writeback.
959 * Otherwise we need to do the tlbie even if C==0 in
960 * order to pick up any delayed writeback of C.
962 hptep1 = be64_to_cpu(hptep[1]);
963 if (!(hptep1 & HPTE_R_C) &&
964 (!hpte_is_writable(hptep1) || vcpus_running(kvm)))
967 if (!try_lock_hpte(hptep, HPTE_V_HVLOCK)) {
968 /* unlock rmap before spinning on the HPTE lock */
970 while (hptep[0] & cpu_to_be64(HPTE_V_HVLOCK))
975 /* Now check and modify the HPTE */
976 if (!(hptep[0] & cpu_to_be64(HPTE_V_VALID))) {
977 /* unlock and continue */
978 hptep[0] &= ~cpu_to_be64(HPTE_V_HVLOCK);
982 /* need to make it temporarily absent so C is stable */
983 hptep[0] |= cpu_to_be64(HPTE_V_ABSENT);
984 kvmppc_invalidate_hpte(kvm, hptep, i);
985 v = be64_to_cpu(hptep[0]);
986 r = be64_to_cpu(hptep[1]);
988 hptep[1] = cpu_to_be64(r & ~HPTE_R_C);
989 if (!(rev[i].guest_rpte & HPTE_R_C)) {
990 rev[i].guest_rpte |= HPTE_R_C;
991 note_hpte_modification(kvm, &rev[i]);
993 n = hpte_page_size(v, r);
994 n = (n + PAGE_SIZE - 1) >> PAGE_SHIFT;
995 if (n > npages_dirty)
999 v &= ~(HPTE_V_ABSENT | HPTE_V_HVLOCK);
1001 hptep[0] = cpu_to_be64(v);
1002 } while ((i = j) != head);
1005 return npages_dirty;
1008 static void harvest_vpa_dirty(struct kvmppc_vpa *vpa,
1009 struct kvm_memory_slot *memslot,
1014 if (!vpa->dirty || !vpa->pinned_addr)
1016 gfn = vpa->gpa >> PAGE_SHIFT;
1017 if (gfn < memslot->base_gfn ||
1018 gfn >= memslot->base_gfn + memslot->npages)
1023 __set_bit_le(gfn - memslot->base_gfn, map);
1026 long kvmppc_hv_get_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot,
1030 unsigned long *rmapp;
1031 struct kvm_vcpu *vcpu;
1034 rmapp = memslot->arch.rmap;
1035 for (i = 0; i < memslot->npages; ++i) {
1036 int npages = kvm_test_clear_dirty_npages(kvm, rmapp);
1038 * Note that if npages > 0 then i must be a multiple of npages,
1039 * since we always put huge-page HPTEs in the rmap chain
1040 * corresponding to their page base address.
1043 for (j = i; npages; ++j, --npages)
1044 __set_bit_le(j, map);
1048 /* Harvest dirty bits from VPA and DTL updates */
1049 /* Note: we never modify the SLB shadow buffer areas */
1050 kvm_for_each_vcpu(i, vcpu, kvm) {
1051 spin_lock(&vcpu->arch.vpa_update_lock);
1052 harvest_vpa_dirty(&vcpu->arch.vpa, memslot, map);
1053 harvest_vpa_dirty(&vcpu->arch.dtl, memslot, map);
1054 spin_unlock(&vcpu->arch.vpa_update_lock);
1060 void *kvmppc_pin_guest_page(struct kvm *kvm, unsigned long gpa,
1061 unsigned long *nb_ret)
1063 struct kvm_memory_slot *memslot;
1064 unsigned long gfn = gpa >> PAGE_SHIFT;
1065 struct page *page, *pages[1];
1067 unsigned long hva, offset;
1070 srcu_idx = srcu_read_lock(&kvm->srcu);
1071 memslot = gfn_to_memslot(kvm, gfn);
1072 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
1074 hva = gfn_to_hva_memslot(memslot, gfn);
1075 npages = get_user_pages_fast(hva, 1, 1, pages);
1079 srcu_read_unlock(&kvm->srcu, srcu_idx);
1081 offset = gpa & (PAGE_SIZE - 1);
1083 *nb_ret = PAGE_SIZE - offset;
1084 return page_address(page) + offset;
1087 srcu_read_unlock(&kvm->srcu, srcu_idx);
1091 void kvmppc_unpin_guest_page(struct kvm *kvm, void *va, unsigned long gpa,
1094 struct page *page = virt_to_page(va);
1095 struct kvm_memory_slot *memslot;
1097 unsigned long *rmap;
1105 /* We need to mark this page dirty in the rmap chain */
1106 gfn = gpa >> PAGE_SHIFT;
1107 srcu_idx = srcu_read_lock(&kvm->srcu);
1108 memslot = gfn_to_memslot(kvm, gfn);
1110 rmap = &memslot->arch.rmap[gfn - memslot->base_gfn];
1112 *rmap |= KVMPPC_RMAP_CHANGED;
1115 srcu_read_unlock(&kvm->srcu, srcu_idx);
1119 * Functions for reading and writing the hash table via reads and
1120 * writes on a file descriptor.
1122 * Reads return the guest view of the hash table, which has to be
1123 * pieced together from the real hash table and the guest_rpte
1124 * values in the revmap array.
1126 * On writes, each HPTE written is considered in turn, and if it
1127 * is valid, it is written to the HPT as if an H_ENTER with the
1128 * exact flag set was done. When the invalid count is non-zero
1129 * in the header written to the stream, the kernel will make
1130 * sure that that many HPTEs are invalid, and invalidate them
1134 struct kvm_htab_ctx {
1135 unsigned long index;
1136 unsigned long flags;
1141 #define HPTE_SIZE (2 * sizeof(unsigned long))
1144 * Returns 1 if this HPT entry has been modified or has pending
1147 static int hpte_dirty(struct revmap_entry *revp, __be64 *hptp)
1149 unsigned long rcbits_unset;
1151 if (revp->guest_rpte & HPTE_GR_MODIFIED)
1154 /* Also need to consider changes in reference and changed bits */
1155 rcbits_unset = ~revp->guest_rpte & (HPTE_R_R | HPTE_R_C);
1156 if ((be64_to_cpu(hptp[0]) & HPTE_V_VALID) &&
1157 (be64_to_cpu(hptp[1]) & rcbits_unset))
1163 static long record_hpte(unsigned long flags, __be64 *hptp,
1164 unsigned long *hpte, struct revmap_entry *revp,
1165 int want_valid, int first_pass)
1168 unsigned long rcbits_unset;
1172 /* Unmodified entries are uninteresting except on the first pass */
1173 dirty = hpte_dirty(revp, hptp);
1174 if (!first_pass && !dirty)
1178 if (be64_to_cpu(hptp[0]) & (HPTE_V_VALID | HPTE_V_ABSENT)) {
1180 if ((flags & KVM_GET_HTAB_BOLTED_ONLY) &&
1181 !(be64_to_cpu(hptp[0]) & HPTE_V_BOLTED))
1184 if (valid != want_valid)
1188 if (valid || dirty) {
1189 /* lock the HPTE so it's stable and read it */
1191 while (!try_lock_hpte(hptp, HPTE_V_HVLOCK))
1193 v = be64_to_cpu(hptp[0]);
1195 /* re-evaluate valid and dirty from synchronized HPTE value */
1196 valid = !!(v & HPTE_V_VALID);
1197 dirty = !!(revp->guest_rpte & HPTE_GR_MODIFIED);
1199 /* Harvest R and C into guest view if necessary */
1200 rcbits_unset = ~revp->guest_rpte & (HPTE_R_R | HPTE_R_C);
1201 if (valid && (rcbits_unset & be64_to_cpu(hptp[1]))) {
1202 revp->guest_rpte |= (be64_to_cpu(hptp[1]) &
1203 (HPTE_R_R | HPTE_R_C)) | HPTE_GR_MODIFIED;
1207 if (v & HPTE_V_ABSENT) {
1208 v &= ~HPTE_V_ABSENT;
1212 if ((flags & KVM_GET_HTAB_BOLTED_ONLY) && !(v & HPTE_V_BOLTED))
1215 r = revp->guest_rpte;
1216 /* only clear modified if this is the right sort of entry */
1217 if (valid == want_valid && dirty) {
1218 r &= ~HPTE_GR_MODIFIED;
1219 revp->guest_rpte = r;
1221 asm volatile(PPC_RELEASE_BARRIER "" : : : "memory");
1222 hptp[0] &= ~cpu_to_be64(HPTE_V_HVLOCK);
1224 if (!(valid == want_valid && (first_pass || dirty)))
1227 hpte[0] = cpu_to_be64(v);
1228 hpte[1] = cpu_to_be64(r);
1232 static ssize_t kvm_htab_read(struct file *file, char __user *buf,
1233 size_t count, loff_t *ppos)
1235 struct kvm_htab_ctx *ctx = file->private_data;
1236 struct kvm *kvm = ctx->kvm;
1237 struct kvm_get_htab_header hdr;
1239 struct revmap_entry *revp;
1240 unsigned long i, nb, nw;
1241 unsigned long __user *lbuf;
1242 struct kvm_get_htab_header __user *hptr;
1243 unsigned long flags;
1245 unsigned long hpte[2];
1247 if (!access_ok(VERIFY_WRITE, buf, count))
1250 first_pass = ctx->first_pass;
1254 hptp = (__be64 *)(kvm->arch.hpt_virt + (i * HPTE_SIZE));
1255 revp = kvm->arch.revmap + i;
1256 lbuf = (unsigned long __user *)buf;
1259 while (nb + sizeof(hdr) + HPTE_SIZE < count) {
1260 /* Initialize header */
1261 hptr = (struct kvm_get_htab_header __user *)buf;
1266 lbuf = (unsigned long __user *)(buf + sizeof(hdr));
1268 /* Skip uninteresting entries, i.e. clean on not-first pass */
1270 while (i < kvm->arch.hpt_npte &&
1271 !hpte_dirty(revp, hptp)) {
1279 /* Grab a series of valid entries */
1280 while (i < kvm->arch.hpt_npte &&
1281 hdr.n_valid < 0xffff &&
1282 nb + HPTE_SIZE < count &&
1283 record_hpte(flags, hptp, hpte, revp, 1, first_pass)) {
1284 /* valid entry, write it out */
1286 if (__put_user(hpte[0], lbuf) ||
1287 __put_user(hpte[1], lbuf + 1))
1295 /* Now skip invalid entries while we can */
1296 while (i < kvm->arch.hpt_npte &&
1297 hdr.n_invalid < 0xffff &&
1298 record_hpte(flags, hptp, hpte, revp, 0, first_pass)) {
1299 /* found an invalid entry */
1306 if (hdr.n_valid || hdr.n_invalid) {
1307 /* write back the header */
1308 if (__copy_to_user(hptr, &hdr, sizeof(hdr)))
1311 buf = (char __user *)lbuf;
1316 /* Check if we've wrapped around the hash table */
1317 if (i >= kvm->arch.hpt_npte) {
1319 ctx->first_pass = 0;
1329 static ssize_t kvm_htab_write(struct file *file, const char __user *buf,
1330 size_t count, loff_t *ppos)
1332 struct kvm_htab_ctx *ctx = file->private_data;
1333 struct kvm *kvm = ctx->kvm;
1334 struct kvm_get_htab_header hdr;
1337 unsigned long __user *lbuf;
1339 unsigned long tmp[2];
1344 if (!access_ok(VERIFY_READ, buf, count))
1347 /* lock out vcpus from running while we're doing this */
1348 mutex_lock(&kvm->lock);
1349 rma_setup = kvm->arch.rma_setup_done;
1351 kvm->arch.rma_setup_done = 0; /* temporarily */
1352 /* order rma_setup_done vs. vcpus_running */
1354 if (atomic_read(&kvm->arch.vcpus_running)) {
1355 kvm->arch.rma_setup_done = 1;
1356 mutex_unlock(&kvm->lock);
1362 for (nb = 0; nb + sizeof(hdr) <= count; ) {
1364 if (__copy_from_user(&hdr, buf, sizeof(hdr)))
1368 if (nb + hdr.n_valid * HPTE_SIZE > count)
1376 if (i >= kvm->arch.hpt_npte ||
1377 i + hdr.n_valid + hdr.n_invalid > kvm->arch.hpt_npte)
1380 hptp = (__be64 *)(kvm->arch.hpt_virt + (i * HPTE_SIZE));
1381 lbuf = (unsigned long __user *)buf;
1382 for (j = 0; j < hdr.n_valid; ++j) {
1387 if (__get_user(hpte_v, lbuf) ||
1388 __get_user(hpte_r, lbuf + 1))
1390 v = be64_to_cpu(hpte_v);
1391 r = be64_to_cpu(hpte_r);
1393 if (!(v & HPTE_V_VALID))
1398 if (be64_to_cpu(hptp[0]) & (HPTE_V_VALID | HPTE_V_ABSENT))
1399 kvmppc_do_h_remove(kvm, 0, i, 0, tmp);
1401 ret = kvmppc_virtmode_do_h_enter(kvm, H_EXACT, i, v, r,
1403 if (ret != H_SUCCESS) {
1404 pr_err("kvm_htab_write ret %ld i=%ld v=%lx "
1405 "r=%lx\n", ret, i, v, r);
1408 if (!rma_setup && is_vrma_hpte(v)) {
1409 unsigned long psize = hpte_base_page_size(v, r);
1410 unsigned long senc = slb_pgsize_encoding(psize);
1413 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
1414 (VRMA_VSID << SLB_VSID_SHIFT_1T);
1415 lpcr = senc << (LPCR_VRMASD_SH - 4);
1416 kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
1423 for (j = 0; j < hdr.n_invalid; ++j) {
1424 if (be64_to_cpu(hptp[0]) & (HPTE_V_VALID | HPTE_V_ABSENT))
1425 kvmppc_do_h_remove(kvm, 0, i, 0, tmp);
1433 /* Order HPTE updates vs. rma_setup_done */
1435 kvm->arch.rma_setup_done = rma_setup;
1436 mutex_unlock(&kvm->lock);
1443 static int kvm_htab_release(struct inode *inode, struct file *filp)
1445 struct kvm_htab_ctx *ctx = filp->private_data;
1447 filp->private_data = NULL;
1448 if (!(ctx->flags & KVM_GET_HTAB_WRITE))
1449 atomic_dec(&ctx->kvm->arch.hpte_mod_interest);
1450 kvm_put_kvm(ctx->kvm);
1455 static const struct file_operations kvm_htab_fops = {
1456 .read = kvm_htab_read,
1457 .write = kvm_htab_write,
1458 .llseek = default_llseek,
1459 .release = kvm_htab_release,
1462 int kvm_vm_ioctl_get_htab_fd(struct kvm *kvm, struct kvm_get_htab_fd *ghf)
1465 struct kvm_htab_ctx *ctx;
1468 /* reject flags we don't recognize */
1469 if (ghf->flags & ~(KVM_GET_HTAB_BOLTED_ONLY | KVM_GET_HTAB_WRITE))
1471 ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
1476 ctx->index = ghf->start_index;
1477 ctx->flags = ghf->flags;
1478 ctx->first_pass = 1;
1480 rwflag = (ghf->flags & KVM_GET_HTAB_WRITE) ? O_WRONLY : O_RDONLY;
1481 ret = anon_inode_getfd("kvm-htab", &kvm_htab_fops, ctx, rwflag | O_CLOEXEC);
1487 if (rwflag == O_RDONLY) {
1488 mutex_lock(&kvm->slots_lock);
1489 atomic_inc(&kvm->arch.hpte_mod_interest);
1490 /* make sure kvmppc_do_h_enter etc. see the increment */
1491 synchronize_srcu_expedited(&kvm->srcu);
1492 mutex_unlock(&kvm->slots_lock);
1498 void kvmppc_mmu_book3s_hv_init(struct kvm_vcpu *vcpu)
1500 struct kvmppc_mmu *mmu = &vcpu->arch.mmu;
1502 vcpu->arch.slb_nr = 32; /* POWER7/POWER8 */
1504 mmu->xlate = kvmppc_mmu_book3s_64_hv_xlate;
1505 mmu->reset_msr = kvmppc_mmu_book3s_64_hv_reset_msr;
1507 vcpu->arch.hflags |= BOOK3S_HFLAG_SLB;