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>
30 #include <linux/debugfs.h>
32 #include <asm/tlbflush.h>
33 #include <asm/kvm_ppc.h>
34 #include <asm/kvm_book3s.h>
35 #include <asm/book3s/64/mmu-hash.h>
36 #include <asm/hvcall.h>
37 #include <asm/synch.h>
38 #include <asm/ppc-opcode.h>
39 #include <asm/cputable.h>
40 #include <asm/pte-walk.h>
44 //#define DEBUG_RESIZE_HPT 1
46 #ifdef DEBUG_RESIZE_HPT
47 #define resize_hpt_debug(resize, ...) \
49 printk(KERN_DEBUG "RESIZE HPT %p: ", resize); \
50 printk(__VA_ARGS__); \
53 #define resize_hpt_debug(resize, ...) \
57 static long kvmppc_virtmode_do_h_enter(struct kvm *kvm, unsigned long flags,
58 long pte_index, unsigned long pteh,
59 unsigned long ptel, unsigned long *pte_idx_ret);
61 struct kvm_resize_hpt {
62 /* These fields read-only after init */
64 struct work_struct work;
67 /* These fields protected by kvm->lock */
69 /* Possible values and their usage:
70 * <0 an error occurred during allocation,
71 * -EBUSY allocation is in the progress,
72 * 0 allocation made successfuly.
76 /* Private to the work thread, until error != -EBUSY,
77 * then protected by kvm->lock.
79 struct kvm_hpt_info hpt;
82 int kvmppc_allocate_hpt(struct kvm_hpt_info *info, u32 order)
84 unsigned long hpt = 0;
86 struct page *page = NULL;
87 struct revmap_entry *rev;
90 if ((order < PPC_MIN_HPT_ORDER) || (order > PPC_MAX_HPT_ORDER))
93 page = kvm_alloc_hpt_cma(1ul << (order - PAGE_SHIFT));
95 hpt = (unsigned long)pfn_to_kaddr(page_to_pfn(page));
96 memset((void *)hpt, 0, (1ul << order));
101 hpt = __get_free_pages(GFP_KERNEL|__GFP_ZERO|__GFP_RETRY_MAYFAIL
102 |__GFP_NOWARN, order - PAGE_SHIFT);
107 /* HPTEs are 2**4 bytes long */
108 npte = 1ul << (order - 4);
110 /* Allocate reverse map array */
111 rev = vmalloc(array_size(npte, sizeof(struct revmap_entry)));
114 kvm_free_hpt_cma(page, 1 << (order - PAGE_SHIFT));
116 free_pages(hpt, order - PAGE_SHIFT);
128 void kvmppc_set_hpt(struct kvm *kvm, struct kvm_hpt_info *info)
130 atomic64_set(&kvm->arch.mmio_update, 0);
131 kvm->arch.hpt = *info;
132 kvm->arch.sdr1 = __pa(info->virt) | (info->order - 18);
134 pr_debug("KVM guest htab at %lx (order %ld), LPID %x\n",
135 info->virt, (long)info->order, kvm->arch.lpid);
138 long kvmppc_alloc_reset_hpt(struct kvm *kvm, int order)
141 struct kvm_hpt_info info;
143 mutex_lock(&kvm->lock);
144 if (kvm->arch.mmu_ready) {
145 kvm->arch.mmu_ready = 0;
146 /* order mmu_ready vs. vcpus_running */
148 if (atomic_read(&kvm->arch.vcpus_running)) {
149 kvm->arch.mmu_ready = 1;
153 if (kvm_is_radix(kvm)) {
154 err = kvmppc_switch_mmu_to_hpt(kvm);
159 if (kvm->arch.hpt.order == order) {
160 /* We already have a suitable HPT */
162 /* Set the entire HPT to 0, i.e. invalid HPTEs */
163 memset((void *)kvm->arch.hpt.virt, 0, 1ul << order);
165 * Reset all the reverse-mapping chains for all memslots
167 kvmppc_rmap_reset(kvm);
172 if (kvm->arch.hpt.virt) {
173 kvmppc_free_hpt(&kvm->arch.hpt);
174 kvmppc_rmap_reset(kvm);
177 err = kvmppc_allocate_hpt(&info, order);
180 kvmppc_set_hpt(kvm, &info);
184 /* Ensure that each vcpu will flush its TLB on next entry. */
185 cpumask_setall(&kvm->arch.need_tlb_flush);
187 mutex_unlock(&kvm->lock);
191 void kvmppc_free_hpt(struct kvm_hpt_info *info)
196 kvm_free_hpt_cma(virt_to_page(info->virt),
197 1 << (info->order - PAGE_SHIFT));
199 free_pages(info->virt, info->order - PAGE_SHIFT);
204 /* Bits in first HPTE dword for pagesize 4k, 64k or 16M */
205 static inline unsigned long hpte0_pgsize_encoding(unsigned long pgsize)
207 return (pgsize > 0x1000) ? HPTE_V_LARGE : 0;
210 /* Bits in second HPTE dword for pagesize 4k, 64k or 16M */
211 static inline unsigned long hpte1_pgsize_encoding(unsigned long pgsize)
213 return (pgsize == 0x10000) ? 0x1000 : 0;
216 void kvmppc_map_vrma(struct kvm_vcpu *vcpu, struct kvm_memory_slot *memslot,
217 unsigned long porder)
220 unsigned long npages;
221 unsigned long hp_v, hp_r;
222 unsigned long addr, hash;
224 unsigned long hp0, hp1;
225 unsigned long idx_ret;
227 struct kvm *kvm = vcpu->kvm;
229 psize = 1ul << porder;
230 npages = memslot->npages >> (porder - PAGE_SHIFT);
232 /* VRMA can't be > 1TB */
233 if (npages > 1ul << (40 - porder))
234 npages = 1ul << (40 - porder);
235 /* Can't use more than 1 HPTE per HPTEG */
236 if (npages > kvmppc_hpt_mask(&kvm->arch.hpt) + 1)
237 npages = kvmppc_hpt_mask(&kvm->arch.hpt) + 1;
239 hp0 = HPTE_V_1TB_SEG | (VRMA_VSID << (40 - 16)) |
240 HPTE_V_BOLTED | hpte0_pgsize_encoding(psize);
241 hp1 = hpte1_pgsize_encoding(psize) |
242 HPTE_R_R | HPTE_R_C | HPTE_R_M | PP_RWXX;
244 for (i = 0; i < npages; ++i) {
246 /* can't use hpt_hash since va > 64 bits */
247 hash = (i ^ (VRMA_VSID ^ (VRMA_VSID << 25)))
248 & kvmppc_hpt_mask(&kvm->arch.hpt);
250 * We assume that the hash table is empty and no
251 * vcpus are using it at this stage. Since we create
252 * at most one HPTE per HPTEG, we just assume entry 7
253 * is available and use it.
255 hash = (hash << 3) + 7;
256 hp_v = hp0 | ((addr >> 16) & ~0x7fUL);
258 ret = kvmppc_virtmode_do_h_enter(kvm, H_EXACT, hash, hp_v, hp_r,
260 if (ret != H_SUCCESS) {
261 pr_err("KVM: map_vrma at %lx failed, ret=%ld\n",
268 int kvmppc_mmu_hv_init(void)
270 unsigned long host_lpid, rsvd_lpid;
272 if (!cpu_has_feature(CPU_FTR_HVMODE))
275 if (!mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE))
278 /* POWER7 has 10-bit LPIDs (12-bit in POWER8) */
279 host_lpid = mfspr(SPRN_LPID);
280 rsvd_lpid = LPID_RSVD;
282 kvmppc_init_lpid(rsvd_lpid + 1);
284 kvmppc_claim_lpid(host_lpid);
285 /* rsvd_lpid is reserved for use in partition switching */
286 kvmppc_claim_lpid(rsvd_lpid);
291 static void kvmppc_mmu_book3s_64_hv_reset_msr(struct kvm_vcpu *vcpu)
293 unsigned long msr = vcpu->arch.intr_msr;
295 /* If transactional, change to suspend mode on IRQ delivery */
296 if (MSR_TM_TRANSACTIONAL(vcpu->arch.shregs.msr))
299 msr |= vcpu->arch.shregs.msr & MSR_TS_MASK;
300 kvmppc_set_msr(vcpu, msr);
303 static long kvmppc_virtmode_do_h_enter(struct kvm *kvm, unsigned long flags,
304 long pte_index, unsigned long pteh,
305 unsigned long ptel, unsigned long *pte_idx_ret)
309 /* Protect linux PTE lookup from page table destruction */
310 rcu_read_lock_sched(); /* this disables preemption too */
311 ret = kvmppc_do_h_enter(kvm, flags, pte_index, pteh, ptel,
312 current->mm->pgd, false, pte_idx_ret);
313 rcu_read_unlock_sched();
314 if (ret == H_TOO_HARD) {
315 /* this can't happen */
316 pr_err("KVM: Oops, kvmppc_h_enter returned too hard!\n");
317 ret = H_RESOURCE; /* or something */
323 static struct kvmppc_slb *kvmppc_mmu_book3s_hv_find_slbe(struct kvm_vcpu *vcpu,
329 for (i = 0; i < vcpu->arch.slb_nr; i++) {
330 if (!(vcpu->arch.slb[i].orige & SLB_ESID_V))
333 if (vcpu->arch.slb[i].origv & SLB_VSID_B_1T)
338 if (((vcpu->arch.slb[i].orige ^ eaddr) & mask) == 0)
339 return &vcpu->arch.slb[i];
344 static unsigned long kvmppc_mmu_get_real_addr(unsigned long v, unsigned long r,
347 unsigned long ra_mask;
349 ra_mask = kvmppc_actual_pgsz(v, r) - 1;
350 return (r & HPTE_R_RPN & ~ra_mask) | (ea & ra_mask);
353 static int kvmppc_mmu_book3s_64_hv_xlate(struct kvm_vcpu *vcpu, gva_t eaddr,
354 struct kvmppc_pte *gpte, bool data, bool iswrite)
356 struct kvm *kvm = vcpu->kvm;
357 struct kvmppc_slb *slbe;
359 unsigned long pp, key;
360 unsigned long v, orig_v, gr;
363 int virtmode = vcpu->arch.shregs.msr & (data ? MSR_DR : MSR_IR);
365 if (kvm_is_radix(vcpu->kvm))
366 return kvmppc_mmu_radix_xlate(vcpu, eaddr, gpte, data, iswrite);
370 slbe = kvmppc_mmu_book3s_hv_find_slbe(vcpu, eaddr);
375 /* real mode access */
376 slb_v = vcpu->kvm->arch.vrma_slb_v;
380 /* Find the HPTE in the hash table */
381 index = kvmppc_hv_find_lock_hpte(kvm, eaddr, slb_v,
382 HPTE_V_VALID | HPTE_V_ABSENT);
387 hptep = (__be64 *)(kvm->arch.hpt.virt + (index << 4));
388 v = orig_v = be64_to_cpu(hptep[0]) & ~HPTE_V_HVLOCK;
389 if (cpu_has_feature(CPU_FTR_ARCH_300))
390 v = hpte_new_to_old_v(v, be64_to_cpu(hptep[1]));
391 gr = kvm->arch.hpt.rev[index].guest_rpte;
393 unlock_hpte(hptep, orig_v);
397 gpte->vpage = ((v & HPTE_V_AVPN) << 4) | ((eaddr >> 12) & 0xfff);
399 /* Get PP bits and key for permission check */
400 pp = gr & (HPTE_R_PP0 | HPTE_R_PP);
401 key = (vcpu->arch.shregs.msr & MSR_PR) ? SLB_VSID_KP : SLB_VSID_KS;
404 /* Calculate permissions */
405 gpte->may_read = hpte_read_permission(pp, key);
406 gpte->may_write = hpte_write_permission(pp, key);
407 gpte->may_execute = gpte->may_read && !(gr & (HPTE_R_N | HPTE_R_G));
409 /* Storage key permission check for POWER7 */
410 if (data && virtmode) {
411 int amrfield = hpte_get_skey_perm(gr, vcpu->arch.amr);
418 /* Get the guest physical address */
419 gpte->raddr = kvmppc_mmu_get_real_addr(v, gr, eaddr);
424 * Quick test for whether an instruction is a load or a store.
425 * If the instruction is a load or a store, then this will indicate
426 * which it is, at least on server processors. (Embedded processors
427 * have some external PID instructions that don't follow the rule
428 * embodied here.) If the instruction isn't a load or store, then
429 * this doesn't return anything useful.
431 static int instruction_is_store(unsigned int instr)
436 if ((instr & 0xfc000000) == 0x7c000000)
437 mask = 0x100; /* major opcode 31 */
438 return (instr & mask) != 0;
441 int kvmppc_hv_emulate_mmio(struct kvm_run *run, struct kvm_vcpu *vcpu,
442 unsigned long gpa, gva_t ea, int is_store)
447 * If we fail, we just return to the guest and try executing it again.
449 if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
454 * WARNING: We do not know for sure whether the instruction we just
455 * read from memory is the same that caused the fault in the first
456 * place. If the instruction we read is neither an load or a store,
457 * then it can't access memory, so we don't need to worry about
458 * enforcing access permissions. So, assuming it is a load or
459 * store, we just check that its direction (load or store) is
460 * consistent with the original fault, since that's what we
461 * checked the access permissions against. If there is a mismatch
462 * we just return and retry the instruction.
465 if (instruction_is_store(last_inst) != !!is_store)
469 * Emulated accesses are emulated by looking at the hash for
470 * translation once, then performing the access later. The
471 * translation could be invalidated in the meantime in which
472 * point performing the subsequent memory access on the old
473 * physical address could possibly be a security hole for the
474 * guest (but not the host).
476 * This is less of an issue for MMIO stores since they aren't
477 * globally visible. It could be an issue for MMIO loads to
478 * a certain extent but we'll ignore it for now.
481 vcpu->arch.paddr_accessed = gpa;
482 vcpu->arch.vaddr_accessed = ea;
483 return kvmppc_emulate_mmio(run, vcpu);
486 int kvmppc_book3s_hv_page_fault(struct kvm_run *run, struct kvm_vcpu *vcpu,
487 unsigned long ea, unsigned long dsisr)
489 struct kvm *kvm = vcpu->kvm;
490 unsigned long hpte[3], r;
491 unsigned long hnow_v, hnow_r;
493 unsigned long mmu_seq, psize, pte_size;
494 unsigned long gpa_base, gfn_base;
495 unsigned long gpa, gfn, hva, pfn;
496 struct kvm_memory_slot *memslot;
498 struct revmap_entry *rev;
499 struct page *page, *pages[1];
500 long index, ret, npages;
502 unsigned int writing, write_ok;
503 struct vm_area_struct *vma;
504 unsigned long rcbits;
507 if (kvm_is_radix(kvm))
508 return kvmppc_book3s_radix_page_fault(run, vcpu, ea, dsisr);
511 * Real-mode code has already searched the HPT and found the
512 * entry we're interested in. Lock the entry and check that
513 * it hasn't changed. If it has, just return and re-execute the
516 if (ea != vcpu->arch.pgfault_addr)
519 if (vcpu->arch.pgfault_cache) {
520 mmio_update = atomic64_read(&kvm->arch.mmio_update);
521 if (mmio_update == vcpu->arch.pgfault_cache->mmio_update) {
522 r = vcpu->arch.pgfault_cache->rpte;
523 psize = kvmppc_actual_pgsz(vcpu->arch.pgfault_hpte[0],
525 gpa_base = r & HPTE_R_RPN & ~(psize - 1);
526 gfn_base = gpa_base >> PAGE_SHIFT;
527 gpa = gpa_base | (ea & (psize - 1));
528 return kvmppc_hv_emulate_mmio(run, vcpu, gpa, ea,
529 dsisr & DSISR_ISSTORE);
532 index = vcpu->arch.pgfault_index;
533 hptep = (__be64 *)(kvm->arch.hpt.virt + (index << 4));
534 rev = &kvm->arch.hpt.rev[index];
536 while (!try_lock_hpte(hptep, HPTE_V_HVLOCK))
538 hpte[0] = be64_to_cpu(hptep[0]) & ~HPTE_V_HVLOCK;
539 hpte[1] = be64_to_cpu(hptep[1]);
540 hpte[2] = r = rev->guest_rpte;
541 unlock_hpte(hptep, hpte[0]);
544 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
545 hpte[0] = hpte_new_to_old_v(hpte[0], hpte[1]);
546 hpte[1] = hpte_new_to_old_r(hpte[1]);
548 if (hpte[0] != vcpu->arch.pgfault_hpte[0] ||
549 hpte[1] != vcpu->arch.pgfault_hpte[1])
552 /* Translate the logical address and get the page */
553 psize = kvmppc_actual_pgsz(hpte[0], r);
554 gpa_base = r & HPTE_R_RPN & ~(psize - 1);
555 gfn_base = gpa_base >> PAGE_SHIFT;
556 gpa = gpa_base | (ea & (psize - 1));
557 gfn = gpa >> PAGE_SHIFT;
558 memslot = gfn_to_memslot(kvm, gfn);
560 trace_kvm_page_fault_enter(vcpu, hpte, memslot, ea, dsisr);
562 /* No memslot means it's an emulated MMIO region */
563 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
564 return kvmppc_hv_emulate_mmio(run, vcpu, gpa, ea,
565 dsisr & DSISR_ISSTORE);
568 * This should never happen, because of the slot_is_aligned()
569 * check in kvmppc_do_h_enter().
571 if (gfn_base < memslot->base_gfn)
574 /* used to check for invalidations in progress */
575 mmu_seq = kvm->mmu_notifier_seq;
582 pte_size = PAGE_SIZE;
583 writing = (dsisr & DSISR_ISSTORE) != 0;
584 /* If writing != 0, then the HPTE must allow writing, if we get here */
586 hva = gfn_to_hva_memslot(memslot, gfn);
587 npages = get_user_pages_fast(hva, 1, writing, pages);
589 /* Check if it's an I/O mapping */
590 down_read(¤t->mm->mmap_sem);
591 vma = find_vma(current->mm, hva);
592 if (vma && vma->vm_start <= hva && hva + psize <= vma->vm_end &&
593 (vma->vm_flags & VM_PFNMAP)) {
594 pfn = vma->vm_pgoff +
595 ((hva - vma->vm_start) >> PAGE_SHIFT);
597 is_ci = pte_ci(__pte((pgprot_val(vma->vm_page_prot))));
598 write_ok = vma->vm_flags & VM_WRITE;
600 up_read(¤t->mm->mmap_sem);
605 pfn = page_to_pfn(page);
606 if (PageHuge(page)) {
607 page = compound_head(page);
608 pte_size <<= compound_order(page);
610 /* if the guest wants write access, see if that is OK */
611 if (!writing && hpte_is_writable(r)) {
615 * We need to protect against page table destruction
616 * hugepage split and collapse.
618 local_irq_save(flags);
619 ptep = find_current_mm_pte(current->mm->pgd,
622 pte = kvmppc_read_update_linux_pte(ptep, 1);
623 if (__pte_write(pte))
626 local_irq_restore(flags);
630 if (psize > pte_size)
633 /* Check WIMG vs. the actual page we're accessing */
634 if (!hpte_cache_flags_ok(r, is_ci)) {
638 * Allow guest to map emulated device memory as
639 * uncacheable, but actually make it cacheable.
641 r = (r & ~(HPTE_R_W|HPTE_R_I|HPTE_R_G)) | HPTE_R_M;
645 * Set the HPTE to point to pfn.
646 * Since the pfn is at PAGE_SIZE granularity, make sure we
647 * don't mask out lower-order bits if psize < PAGE_SIZE.
649 if (psize < PAGE_SIZE)
651 r = (r & HPTE_R_KEY_HI) | (r & ~(HPTE_R_PP0 - psize)) |
652 ((pfn << PAGE_SHIFT) & ~(psize - 1));
653 if (hpte_is_writable(r) && !write_ok)
654 r = hpte_make_readonly(r);
657 while (!try_lock_hpte(hptep, HPTE_V_HVLOCK))
659 hnow_v = be64_to_cpu(hptep[0]);
660 hnow_r = be64_to_cpu(hptep[1]);
661 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
662 hnow_v = hpte_new_to_old_v(hnow_v, hnow_r);
663 hnow_r = hpte_new_to_old_r(hnow_r);
667 * If the HPT is being resized, don't update the HPTE,
668 * instead let the guest retry after the resize operation is complete.
669 * The synchronization for mmu_ready test vs. set is provided
672 if (!kvm->arch.mmu_ready)
675 if ((hnow_v & ~HPTE_V_HVLOCK) != hpte[0] || hnow_r != hpte[1] ||
676 rev->guest_rpte != hpte[2])
677 /* HPTE has been changed under us; let the guest retry */
679 hpte[0] = (hpte[0] & ~HPTE_V_ABSENT) | HPTE_V_VALID;
681 /* Always put the HPTE in the rmap chain for the page base address */
682 rmap = &memslot->arch.rmap[gfn_base - memslot->base_gfn];
685 /* Check if we might have been invalidated; let the guest retry if so */
687 if (mmu_notifier_retry(vcpu->kvm, mmu_seq)) {
692 /* Only set R/C in real HPTE if set in both *rmap and guest_rpte */
693 rcbits = *rmap >> KVMPPC_RMAP_RC_SHIFT;
694 r &= rcbits | ~(HPTE_R_R | HPTE_R_C);
696 if (be64_to_cpu(hptep[0]) & HPTE_V_VALID) {
697 /* HPTE was previously valid, so we need to invalidate it */
699 hptep[0] |= cpu_to_be64(HPTE_V_ABSENT);
700 kvmppc_invalidate_hpte(kvm, hptep, index);
701 /* don't lose previous R and C bits */
702 r |= be64_to_cpu(hptep[1]) & (HPTE_R_R | HPTE_R_C);
704 kvmppc_add_revmap_chain(kvm, rev, rmap, index, 0);
707 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
708 r = hpte_old_to_new_r(hpte[0], r);
709 hpte[0] = hpte_old_to_new_v(hpte[0]);
711 hptep[1] = cpu_to_be64(r);
713 __unlock_hpte(hptep, hpte[0]);
714 asm volatile("ptesync" : : : "memory");
716 if (page && hpte_is_writable(r))
720 trace_kvm_page_fault_exit(vcpu, hpte, ret);
724 * We drop pages[0] here, not page because page might
725 * have been set to the head page of a compound, but
726 * we have to drop the reference on the correct tail
727 * page to match the get inside gup()
734 __unlock_hpte(hptep, be64_to_cpu(hptep[0]));
739 void kvmppc_rmap_reset(struct kvm *kvm)
741 struct kvm_memslots *slots;
742 struct kvm_memory_slot *memslot;
745 srcu_idx = srcu_read_lock(&kvm->srcu);
746 slots = kvm_memslots(kvm);
747 kvm_for_each_memslot(memslot, slots) {
749 * This assumes it is acceptable to lose reference and
750 * change bits across a reset.
752 memset(memslot->arch.rmap, 0,
753 memslot->npages * sizeof(*memslot->arch.rmap));
755 srcu_read_unlock(&kvm->srcu, srcu_idx);
758 typedef int (*hva_handler_fn)(struct kvm *kvm, struct kvm_memory_slot *memslot,
761 static int kvm_handle_hva_range(struct kvm *kvm,
764 hva_handler_fn handler)
768 struct kvm_memslots *slots;
769 struct kvm_memory_slot *memslot;
771 slots = kvm_memslots(kvm);
772 kvm_for_each_memslot(memslot, slots) {
773 unsigned long hva_start, hva_end;
776 hva_start = max(start, memslot->userspace_addr);
777 hva_end = min(end, memslot->userspace_addr +
778 (memslot->npages << PAGE_SHIFT));
779 if (hva_start >= hva_end)
782 * {gfn(page) | page intersects with [hva_start, hva_end)} =
783 * {gfn, gfn+1, ..., gfn_end-1}.
785 gfn = hva_to_gfn_memslot(hva_start, memslot);
786 gfn_end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, memslot);
788 for (; gfn < gfn_end; ++gfn) {
789 ret = handler(kvm, memslot, gfn);
797 static int kvm_handle_hva(struct kvm *kvm, unsigned long hva,
798 hva_handler_fn handler)
800 return kvm_handle_hva_range(kvm, hva, hva + 1, handler);
803 /* Must be called with both HPTE and rmap locked */
804 static void kvmppc_unmap_hpte(struct kvm *kvm, unsigned long i,
805 struct kvm_memory_slot *memslot,
806 unsigned long *rmapp, unsigned long gfn)
808 __be64 *hptep = (__be64 *) (kvm->arch.hpt.virt + (i << 4));
809 struct revmap_entry *rev = kvm->arch.hpt.rev;
811 unsigned long ptel, psize, rcbits;
815 /* chain is now empty */
816 *rmapp &= ~(KVMPPC_RMAP_PRESENT | KVMPPC_RMAP_INDEX);
818 /* remove i from chain */
822 rev[i].forw = rev[i].back = i;
823 *rmapp = (*rmapp & ~KVMPPC_RMAP_INDEX) | j;
826 /* Now check and modify the HPTE */
827 ptel = rev[i].guest_rpte;
828 psize = kvmppc_actual_pgsz(be64_to_cpu(hptep[0]), ptel);
829 if ((be64_to_cpu(hptep[0]) & HPTE_V_VALID) &&
830 hpte_rpn(ptel, psize) == gfn) {
831 hptep[0] |= cpu_to_be64(HPTE_V_ABSENT);
832 kvmppc_invalidate_hpte(kvm, hptep, i);
833 hptep[1] &= ~cpu_to_be64(HPTE_R_KEY_HI | HPTE_R_KEY_LO);
834 /* Harvest R and C */
835 rcbits = be64_to_cpu(hptep[1]) & (HPTE_R_R | HPTE_R_C);
836 *rmapp |= rcbits << KVMPPC_RMAP_RC_SHIFT;
837 if ((rcbits & HPTE_R_C) && memslot->dirty_bitmap)
838 kvmppc_update_dirty_map(memslot, gfn, psize);
839 if (rcbits & ~rev[i].guest_rpte) {
840 rev[i].guest_rpte = ptel | rcbits;
841 note_hpte_modification(kvm, &rev[i]);
846 static int kvm_unmap_rmapp(struct kvm *kvm, struct kvm_memory_slot *memslot,
851 unsigned long *rmapp;
853 rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn];
856 if (!(*rmapp & KVMPPC_RMAP_PRESENT)) {
862 * To avoid an ABBA deadlock with the HPTE lock bit,
863 * we can't spin on the HPTE lock while holding the
866 i = *rmapp & KVMPPC_RMAP_INDEX;
867 hptep = (__be64 *) (kvm->arch.hpt.virt + (i << 4));
868 if (!try_lock_hpte(hptep, HPTE_V_HVLOCK)) {
869 /* unlock rmap before spinning on the HPTE lock */
871 while (be64_to_cpu(hptep[0]) & HPTE_V_HVLOCK)
876 kvmppc_unmap_hpte(kvm, i, memslot, rmapp, gfn);
878 __unlock_hpte(hptep, be64_to_cpu(hptep[0]));
883 int kvm_unmap_hva_range_hv(struct kvm *kvm, unsigned long start, unsigned long end)
885 hva_handler_fn handler;
887 handler = kvm_is_radix(kvm) ? kvm_unmap_radix : kvm_unmap_rmapp;
888 kvm_handle_hva_range(kvm, start, end, handler);
892 void kvmppc_core_flush_memslot_hv(struct kvm *kvm,
893 struct kvm_memory_slot *memslot)
897 unsigned long *rmapp;
899 gfn = memslot->base_gfn;
900 rmapp = memslot->arch.rmap;
901 for (n = memslot->npages; n; --n, ++gfn) {
902 if (kvm_is_radix(kvm)) {
903 kvm_unmap_radix(kvm, memslot, gfn);
907 * Testing the present bit without locking is OK because
908 * the memslot has been marked invalid already, and hence
909 * no new HPTEs referencing this page can be created,
910 * thus the present bit can't go from 0 to 1.
912 if (*rmapp & KVMPPC_RMAP_PRESENT)
913 kvm_unmap_rmapp(kvm, memslot, gfn);
918 static int kvm_age_rmapp(struct kvm *kvm, struct kvm_memory_slot *memslot,
921 struct revmap_entry *rev = kvm->arch.hpt.rev;
922 unsigned long head, i, j;
925 unsigned long *rmapp;
927 rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn];
930 if (*rmapp & KVMPPC_RMAP_REFERENCED) {
931 *rmapp &= ~KVMPPC_RMAP_REFERENCED;
934 if (!(*rmapp & KVMPPC_RMAP_PRESENT)) {
939 i = head = *rmapp & KVMPPC_RMAP_INDEX;
941 hptep = (__be64 *) (kvm->arch.hpt.virt + (i << 4));
944 /* If this HPTE isn't referenced, ignore it */
945 if (!(be64_to_cpu(hptep[1]) & HPTE_R_R))
948 if (!try_lock_hpte(hptep, HPTE_V_HVLOCK)) {
949 /* unlock rmap before spinning on the HPTE lock */
951 while (be64_to_cpu(hptep[0]) & HPTE_V_HVLOCK)
956 /* Now check and modify the HPTE */
957 if ((be64_to_cpu(hptep[0]) & HPTE_V_VALID) &&
958 (be64_to_cpu(hptep[1]) & HPTE_R_R)) {
959 kvmppc_clear_ref_hpte(kvm, hptep, i);
960 if (!(rev[i].guest_rpte & HPTE_R_R)) {
961 rev[i].guest_rpte |= HPTE_R_R;
962 note_hpte_modification(kvm, &rev[i]);
966 __unlock_hpte(hptep, be64_to_cpu(hptep[0]));
967 } while ((i = j) != head);
973 int kvm_age_hva_hv(struct kvm *kvm, unsigned long start, unsigned long end)
975 hva_handler_fn handler;
977 handler = kvm_is_radix(kvm) ? kvm_age_radix : kvm_age_rmapp;
978 return kvm_handle_hva_range(kvm, start, end, handler);
981 static int kvm_test_age_rmapp(struct kvm *kvm, struct kvm_memory_slot *memslot,
984 struct revmap_entry *rev = kvm->arch.hpt.rev;
985 unsigned long head, i, j;
988 unsigned long *rmapp;
990 rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn];
991 if (*rmapp & KVMPPC_RMAP_REFERENCED)
995 if (*rmapp & KVMPPC_RMAP_REFERENCED)
998 if (*rmapp & KVMPPC_RMAP_PRESENT) {
999 i = head = *rmapp & KVMPPC_RMAP_INDEX;
1001 hp = (unsigned long *)(kvm->arch.hpt.virt + (i << 4));
1003 if (be64_to_cpu(hp[1]) & HPTE_R_R)
1005 } while ((i = j) != head);
1014 int kvm_test_age_hva_hv(struct kvm *kvm, unsigned long hva)
1016 hva_handler_fn handler;
1018 handler = kvm_is_radix(kvm) ? kvm_test_age_radix : kvm_test_age_rmapp;
1019 return kvm_handle_hva(kvm, hva, handler);
1022 void kvm_set_spte_hva_hv(struct kvm *kvm, unsigned long hva, pte_t pte)
1024 hva_handler_fn handler;
1026 handler = kvm_is_radix(kvm) ? kvm_unmap_radix : kvm_unmap_rmapp;
1027 kvm_handle_hva(kvm, hva, handler);
1030 static int vcpus_running(struct kvm *kvm)
1032 return atomic_read(&kvm->arch.vcpus_running) != 0;
1036 * Returns the number of system pages that are dirty.
1037 * This can be more than 1 if we find a huge-page HPTE.
1039 static int kvm_test_clear_dirty_npages(struct kvm *kvm, unsigned long *rmapp)
1041 struct revmap_entry *rev = kvm->arch.hpt.rev;
1042 unsigned long head, i, j;
1046 int npages_dirty = 0;
1050 if (!(*rmapp & KVMPPC_RMAP_PRESENT)) {
1052 return npages_dirty;
1055 i = head = *rmapp & KVMPPC_RMAP_INDEX;
1057 unsigned long hptep1;
1058 hptep = (__be64 *) (kvm->arch.hpt.virt + (i << 4));
1062 * Checking the C (changed) bit here is racy since there
1063 * is no guarantee about when the hardware writes it back.
1064 * If the HPTE is not writable then it is stable since the
1065 * page can't be written to, and we would have done a tlbie
1066 * (which forces the hardware to complete any writeback)
1067 * when making the HPTE read-only.
1068 * If vcpus are running then this call is racy anyway
1069 * since the page could get dirtied subsequently, so we
1070 * expect there to be a further call which would pick up
1071 * any delayed C bit writeback.
1072 * Otherwise we need to do the tlbie even if C==0 in
1073 * order to pick up any delayed writeback of C.
1075 hptep1 = be64_to_cpu(hptep[1]);
1076 if (!(hptep1 & HPTE_R_C) &&
1077 (!hpte_is_writable(hptep1) || vcpus_running(kvm)))
1080 if (!try_lock_hpte(hptep, HPTE_V_HVLOCK)) {
1081 /* unlock rmap before spinning on the HPTE lock */
1083 while (hptep[0] & cpu_to_be64(HPTE_V_HVLOCK))
1088 /* Now check and modify the HPTE */
1089 if (!(hptep[0] & cpu_to_be64(HPTE_V_VALID))) {
1090 __unlock_hpte(hptep, be64_to_cpu(hptep[0]));
1094 /* need to make it temporarily absent so C is stable */
1095 hptep[0] |= cpu_to_be64(HPTE_V_ABSENT);
1096 kvmppc_invalidate_hpte(kvm, hptep, i);
1097 v = be64_to_cpu(hptep[0]);
1098 r = be64_to_cpu(hptep[1]);
1100 hptep[1] = cpu_to_be64(r & ~HPTE_R_C);
1101 if (!(rev[i].guest_rpte & HPTE_R_C)) {
1102 rev[i].guest_rpte |= HPTE_R_C;
1103 note_hpte_modification(kvm, &rev[i]);
1105 n = kvmppc_actual_pgsz(v, r);
1106 n = (n + PAGE_SIZE - 1) >> PAGE_SHIFT;
1107 if (n > npages_dirty)
1111 v &= ~HPTE_V_ABSENT;
1113 __unlock_hpte(hptep, v);
1114 } while ((i = j) != head);
1117 return npages_dirty;
1120 void kvmppc_harvest_vpa_dirty(struct kvmppc_vpa *vpa,
1121 struct kvm_memory_slot *memslot,
1126 if (!vpa->dirty || !vpa->pinned_addr)
1128 gfn = vpa->gpa >> PAGE_SHIFT;
1129 if (gfn < memslot->base_gfn ||
1130 gfn >= memslot->base_gfn + memslot->npages)
1135 __set_bit_le(gfn - memslot->base_gfn, map);
1138 long kvmppc_hv_get_dirty_log_hpt(struct kvm *kvm,
1139 struct kvm_memory_slot *memslot, unsigned long *map)
1142 unsigned long *rmapp;
1145 rmapp = memslot->arch.rmap;
1146 for (i = 0; i < memslot->npages; ++i) {
1147 int npages = kvm_test_clear_dirty_npages(kvm, rmapp);
1149 * Note that if npages > 0 then i must be a multiple of npages,
1150 * since we always put huge-page HPTEs in the rmap chain
1151 * corresponding to their page base address.
1154 set_dirty_bits(map, i, npages);
1161 void *kvmppc_pin_guest_page(struct kvm *kvm, unsigned long gpa,
1162 unsigned long *nb_ret)
1164 struct kvm_memory_slot *memslot;
1165 unsigned long gfn = gpa >> PAGE_SHIFT;
1166 struct page *page, *pages[1];
1168 unsigned long hva, offset;
1171 srcu_idx = srcu_read_lock(&kvm->srcu);
1172 memslot = gfn_to_memslot(kvm, gfn);
1173 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
1175 hva = gfn_to_hva_memslot(memslot, gfn);
1176 npages = get_user_pages_fast(hva, 1, 1, pages);
1180 srcu_read_unlock(&kvm->srcu, srcu_idx);
1182 offset = gpa & (PAGE_SIZE - 1);
1184 *nb_ret = PAGE_SIZE - offset;
1185 return page_address(page) + offset;
1188 srcu_read_unlock(&kvm->srcu, srcu_idx);
1192 void kvmppc_unpin_guest_page(struct kvm *kvm, void *va, unsigned long gpa,
1195 struct page *page = virt_to_page(va);
1196 struct kvm_memory_slot *memslot;
1205 /* We need to mark this page dirty in the memslot dirty_bitmap, if any */
1206 gfn = gpa >> PAGE_SHIFT;
1207 srcu_idx = srcu_read_lock(&kvm->srcu);
1208 memslot = gfn_to_memslot(kvm, gfn);
1209 if (memslot && memslot->dirty_bitmap)
1210 set_bit_le(gfn - memslot->base_gfn, memslot->dirty_bitmap);
1211 srcu_read_unlock(&kvm->srcu, srcu_idx);
1217 static int resize_hpt_allocate(struct kvm_resize_hpt *resize)
1221 rc = kvmppc_allocate_hpt(&resize->hpt, resize->order);
1225 resize_hpt_debug(resize, "resize_hpt_allocate(): HPT @ 0x%lx\n",
1231 static unsigned long resize_hpt_rehash_hpte(struct kvm_resize_hpt *resize,
1234 struct kvm *kvm = resize->kvm;
1235 struct kvm_hpt_info *old = &kvm->arch.hpt;
1236 struct kvm_hpt_info *new = &resize->hpt;
1237 unsigned long old_hash_mask = (1ULL << (old->order - 7)) - 1;
1238 unsigned long new_hash_mask = (1ULL << (new->order - 7)) - 1;
1239 __be64 *hptep, *new_hptep;
1240 unsigned long vpte, rpte, guest_rpte;
1242 struct revmap_entry *rev;
1243 unsigned long apsize, avpn, pteg, hash;
1244 unsigned long new_idx, new_pteg, replace_vpte;
1247 hptep = (__be64 *)(old->virt + (idx << 4));
1249 /* Guest is stopped, so new HPTEs can't be added or faulted
1250 * in, only unmapped or altered by host actions. So, it's
1251 * safe to check this before we take the HPTE lock */
1252 vpte = be64_to_cpu(hptep[0]);
1253 if (!(vpte & HPTE_V_VALID) && !(vpte & HPTE_V_ABSENT))
1254 return 0; /* nothing to do */
1256 while (!try_lock_hpte(hptep, HPTE_V_HVLOCK))
1259 vpte = be64_to_cpu(hptep[0]);
1262 if (!(vpte & HPTE_V_VALID) && !(vpte & HPTE_V_ABSENT))
1266 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
1267 rpte = be64_to_cpu(hptep[1]);
1268 vpte = hpte_new_to_old_v(vpte, rpte);
1272 rev = &old->rev[idx];
1273 guest_rpte = rev->guest_rpte;
1276 apsize = kvmppc_actual_pgsz(vpte, guest_rpte);
1280 if (vpte & HPTE_V_VALID) {
1281 unsigned long gfn = hpte_rpn(guest_rpte, apsize);
1282 int srcu_idx = srcu_read_lock(&kvm->srcu);
1283 struct kvm_memory_slot *memslot =
1284 __gfn_to_memslot(kvm_memslots(kvm), gfn);
1287 unsigned long *rmapp;
1288 rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn];
1291 kvmppc_unmap_hpte(kvm, idx, memslot, rmapp, gfn);
1295 srcu_read_unlock(&kvm->srcu, srcu_idx);
1298 /* Reload PTE after unmap */
1299 vpte = be64_to_cpu(hptep[0]);
1300 BUG_ON(vpte & HPTE_V_VALID);
1301 BUG_ON(!(vpte & HPTE_V_ABSENT));
1304 if (!(vpte & HPTE_V_BOLTED))
1307 rpte = be64_to_cpu(hptep[1]);
1309 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
1310 vpte = hpte_new_to_old_v(vpte, rpte);
1311 rpte = hpte_new_to_old_r(rpte);
1314 pshift = kvmppc_hpte_base_page_shift(vpte, rpte);
1315 avpn = HPTE_V_AVPN_VAL(vpte) & ~(((1ul << pshift) - 1) >> 23);
1316 pteg = idx / HPTES_PER_GROUP;
1317 if (vpte & HPTE_V_SECONDARY)
1320 if (!(vpte & HPTE_V_1TB_SEG)) {
1321 unsigned long offset, vsid;
1323 /* We only have 28 - 23 bits of offset in avpn */
1324 offset = (avpn & 0x1f) << 23;
1326 /* We can find more bits from the pteg value */
1328 offset |= ((vsid ^ pteg) & old_hash_mask) << pshift;
1330 hash = vsid ^ (offset >> pshift);
1332 unsigned long offset, vsid;
1334 /* We only have 40 - 23 bits of seg_off in avpn */
1335 offset = (avpn & 0x1ffff) << 23;
1338 offset |= ((vsid ^ (vsid << 25) ^ pteg) & old_hash_mask) << pshift;
1340 hash = vsid ^ (vsid << 25) ^ (offset >> pshift);
1343 new_pteg = hash & new_hash_mask;
1344 if (vpte & HPTE_V_SECONDARY)
1345 new_pteg = ~hash & new_hash_mask;
1347 new_idx = new_pteg * HPTES_PER_GROUP + (idx % HPTES_PER_GROUP);
1348 new_hptep = (__be64 *)(new->virt + (new_idx << 4));
1350 replace_vpte = be64_to_cpu(new_hptep[0]);
1351 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
1352 unsigned long replace_rpte = be64_to_cpu(new_hptep[1]);
1353 replace_vpte = hpte_new_to_old_v(replace_vpte, replace_rpte);
1356 if (replace_vpte & (HPTE_V_VALID | HPTE_V_ABSENT)) {
1357 BUG_ON(new->order >= old->order);
1359 if (replace_vpte & HPTE_V_BOLTED) {
1360 if (vpte & HPTE_V_BOLTED)
1361 /* Bolted collision, nothing we can do */
1363 /* Discard the new HPTE */
1367 /* Discard the previous HPTE */
1370 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
1371 rpte = hpte_old_to_new_r(vpte, rpte);
1372 vpte = hpte_old_to_new_v(vpte);
1375 new_hptep[1] = cpu_to_be64(rpte);
1376 new->rev[new_idx].guest_rpte = guest_rpte;
1377 /* No need for a barrier, since new HPT isn't active */
1378 new_hptep[0] = cpu_to_be64(vpte);
1379 unlock_hpte(new_hptep, vpte);
1382 unlock_hpte(hptep, vpte);
1386 static int resize_hpt_rehash(struct kvm_resize_hpt *resize)
1388 struct kvm *kvm = resize->kvm;
1392 for (i = 0; i < kvmppc_hpt_npte(&kvm->arch.hpt); i++) {
1393 rc = resize_hpt_rehash_hpte(resize, i);
1401 static void resize_hpt_pivot(struct kvm_resize_hpt *resize)
1403 struct kvm *kvm = resize->kvm;
1404 struct kvm_hpt_info hpt_tmp;
1406 /* Exchange the pending tables in the resize structure with
1407 * the active tables */
1409 resize_hpt_debug(resize, "resize_hpt_pivot()\n");
1411 spin_lock(&kvm->mmu_lock);
1412 asm volatile("ptesync" : : : "memory");
1414 hpt_tmp = kvm->arch.hpt;
1415 kvmppc_set_hpt(kvm, &resize->hpt);
1416 resize->hpt = hpt_tmp;
1418 spin_unlock(&kvm->mmu_lock);
1420 synchronize_srcu_expedited(&kvm->srcu);
1422 if (cpu_has_feature(CPU_FTR_ARCH_300))
1423 kvmppc_setup_partition_table(kvm);
1425 resize_hpt_debug(resize, "resize_hpt_pivot() done\n");
1428 static void resize_hpt_release(struct kvm *kvm, struct kvm_resize_hpt *resize)
1430 if (WARN_ON(!mutex_is_locked(&kvm->lock)))
1436 if (resize->error != -EBUSY) {
1437 if (resize->hpt.virt)
1438 kvmppc_free_hpt(&resize->hpt);
1442 if (kvm->arch.resize_hpt == resize)
1443 kvm->arch.resize_hpt = NULL;
1446 static void resize_hpt_prepare_work(struct work_struct *work)
1448 struct kvm_resize_hpt *resize = container_of(work,
1449 struct kvm_resize_hpt,
1451 struct kvm *kvm = resize->kvm;
1454 if (WARN_ON(resize->error != -EBUSY))
1457 mutex_lock(&kvm->lock);
1459 /* Request is still current? */
1460 if (kvm->arch.resize_hpt == resize) {
1461 /* We may request large allocations here:
1462 * do not sleep with kvm->lock held for a while.
1464 mutex_unlock(&kvm->lock);
1466 resize_hpt_debug(resize, "resize_hpt_prepare_work(): order = %d\n",
1469 err = resize_hpt_allocate(resize);
1471 /* We have strict assumption about -EBUSY
1472 * when preparing for HPT resize.
1474 if (WARN_ON(err == -EBUSY))
1477 mutex_lock(&kvm->lock);
1478 /* It is possible that kvm->arch.resize_hpt != resize
1479 * after we grab kvm->lock again.
1483 resize->error = err;
1485 if (kvm->arch.resize_hpt != resize)
1486 resize_hpt_release(kvm, resize);
1488 mutex_unlock(&kvm->lock);
1491 long kvm_vm_ioctl_resize_hpt_prepare(struct kvm *kvm,
1492 struct kvm_ppc_resize_hpt *rhpt)
1494 unsigned long flags = rhpt->flags;
1495 unsigned long shift = rhpt->shift;
1496 struct kvm_resize_hpt *resize;
1499 if (flags != 0 || kvm_is_radix(kvm))
1502 if (shift && ((shift < 18) || (shift > 46)))
1505 mutex_lock(&kvm->lock);
1507 resize = kvm->arch.resize_hpt;
1510 if (resize->order == shift) {
1511 /* Suitable resize in progress? */
1512 ret = resize->error;
1514 ret = 100; /* estimated time in ms */
1516 resize_hpt_release(kvm, resize);
1521 /* not suitable, cancel it */
1522 resize_hpt_release(kvm, resize);
1527 goto out; /* nothing to do */
1529 /* start new resize */
1531 resize = kzalloc(sizeof(*resize), GFP_KERNEL);
1537 resize->error = -EBUSY;
1538 resize->order = shift;
1540 INIT_WORK(&resize->work, resize_hpt_prepare_work);
1541 kvm->arch.resize_hpt = resize;
1543 schedule_work(&resize->work);
1545 ret = 100; /* estimated time in ms */
1548 mutex_unlock(&kvm->lock);
1552 static void resize_hpt_boot_vcpu(void *opaque)
1554 /* Nothing to do, just force a KVM exit */
1557 long kvm_vm_ioctl_resize_hpt_commit(struct kvm *kvm,
1558 struct kvm_ppc_resize_hpt *rhpt)
1560 unsigned long flags = rhpt->flags;
1561 unsigned long shift = rhpt->shift;
1562 struct kvm_resize_hpt *resize;
1565 if (flags != 0 || kvm_is_radix(kvm))
1568 if (shift && ((shift < 18) || (shift > 46)))
1571 mutex_lock(&kvm->lock);
1573 resize = kvm->arch.resize_hpt;
1575 /* This shouldn't be possible */
1577 if (WARN_ON(!kvm->arch.mmu_ready))
1580 /* Stop VCPUs from running while we mess with the HPT */
1581 kvm->arch.mmu_ready = 0;
1584 /* Boot all CPUs out of the guest so they re-read
1586 on_each_cpu(resize_hpt_boot_vcpu, NULL, 1);
1589 if (!resize || (resize->order != shift))
1592 ret = resize->error;
1596 ret = resize_hpt_rehash(resize);
1600 resize_hpt_pivot(resize);
1603 /* Let VCPUs run again */
1604 kvm->arch.mmu_ready = 1;
1607 resize_hpt_release(kvm, resize);
1608 mutex_unlock(&kvm->lock);
1613 * Functions for reading and writing the hash table via reads and
1614 * writes on a file descriptor.
1616 * Reads return the guest view of the hash table, which has to be
1617 * pieced together from the real hash table and the guest_rpte
1618 * values in the revmap array.
1620 * On writes, each HPTE written is considered in turn, and if it
1621 * is valid, it is written to the HPT as if an H_ENTER with the
1622 * exact flag set was done. When the invalid count is non-zero
1623 * in the header written to the stream, the kernel will make
1624 * sure that that many HPTEs are invalid, and invalidate them
1628 struct kvm_htab_ctx {
1629 unsigned long index;
1630 unsigned long flags;
1635 #define HPTE_SIZE (2 * sizeof(unsigned long))
1638 * Returns 1 if this HPT entry has been modified or has pending
1641 static int hpte_dirty(struct revmap_entry *revp, __be64 *hptp)
1643 unsigned long rcbits_unset;
1645 if (revp->guest_rpte & HPTE_GR_MODIFIED)
1648 /* Also need to consider changes in reference and changed bits */
1649 rcbits_unset = ~revp->guest_rpte & (HPTE_R_R | HPTE_R_C);
1650 if ((be64_to_cpu(hptp[0]) & HPTE_V_VALID) &&
1651 (be64_to_cpu(hptp[1]) & rcbits_unset))
1657 static long record_hpte(unsigned long flags, __be64 *hptp,
1658 unsigned long *hpte, struct revmap_entry *revp,
1659 int want_valid, int first_pass)
1661 unsigned long v, r, hr;
1662 unsigned long rcbits_unset;
1666 /* Unmodified entries are uninteresting except on the first pass */
1667 dirty = hpte_dirty(revp, hptp);
1668 if (!first_pass && !dirty)
1672 if (be64_to_cpu(hptp[0]) & (HPTE_V_VALID | HPTE_V_ABSENT)) {
1674 if ((flags & KVM_GET_HTAB_BOLTED_ONLY) &&
1675 !(be64_to_cpu(hptp[0]) & HPTE_V_BOLTED))
1678 if (valid != want_valid)
1682 if (valid || dirty) {
1683 /* lock the HPTE so it's stable and read it */
1685 while (!try_lock_hpte(hptp, HPTE_V_HVLOCK))
1687 v = be64_to_cpu(hptp[0]);
1688 hr = be64_to_cpu(hptp[1]);
1689 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
1690 v = hpte_new_to_old_v(v, hr);
1691 hr = hpte_new_to_old_r(hr);
1694 /* re-evaluate valid and dirty from synchronized HPTE value */
1695 valid = !!(v & HPTE_V_VALID);
1696 dirty = !!(revp->guest_rpte & HPTE_GR_MODIFIED);
1698 /* Harvest R and C into guest view if necessary */
1699 rcbits_unset = ~revp->guest_rpte & (HPTE_R_R | HPTE_R_C);
1700 if (valid && (rcbits_unset & hr)) {
1701 revp->guest_rpte |= (hr &
1702 (HPTE_R_R | HPTE_R_C)) | HPTE_GR_MODIFIED;
1706 if (v & HPTE_V_ABSENT) {
1707 v &= ~HPTE_V_ABSENT;
1711 if ((flags & KVM_GET_HTAB_BOLTED_ONLY) && !(v & HPTE_V_BOLTED))
1714 r = revp->guest_rpte;
1715 /* only clear modified if this is the right sort of entry */
1716 if (valid == want_valid && dirty) {
1717 r &= ~HPTE_GR_MODIFIED;
1718 revp->guest_rpte = r;
1720 unlock_hpte(hptp, be64_to_cpu(hptp[0]));
1722 if (!(valid == want_valid && (first_pass || dirty)))
1725 hpte[0] = cpu_to_be64(v);
1726 hpte[1] = cpu_to_be64(r);
1730 static ssize_t kvm_htab_read(struct file *file, char __user *buf,
1731 size_t count, loff_t *ppos)
1733 struct kvm_htab_ctx *ctx = file->private_data;
1734 struct kvm *kvm = ctx->kvm;
1735 struct kvm_get_htab_header hdr;
1737 struct revmap_entry *revp;
1738 unsigned long i, nb, nw;
1739 unsigned long __user *lbuf;
1740 struct kvm_get_htab_header __user *hptr;
1741 unsigned long flags;
1743 unsigned long hpte[2];
1745 if (!access_ok(VERIFY_WRITE, buf, count))
1747 if (kvm_is_radix(kvm))
1750 first_pass = ctx->first_pass;
1754 hptp = (__be64 *)(kvm->arch.hpt.virt + (i * HPTE_SIZE));
1755 revp = kvm->arch.hpt.rev + i;
1756 lbuf = (unsigned long __user *)buf;
1759 while (nb + sizeof(hdr) + HPTE_SIZE < count) {
1760 /* Initialize header */
1761 hptr = (struct kvm_get_htab_header __user *)buf;
1766 lbuf = (unsigned long __user *)(buf + sizeof(hdr));
1768 /* Skip uninteresting entries, i.e. clean on not-first pass */
1770 while (i < kvmppc_hpt_npte(&kvm->arch.hpt) &&
1771 !hpte_dirty(revp, hptp)) {
1779 /* Grab a series of valid entries */
1780 while (i < kvmppc_hpt_npte(&kvm->arch.hpt) &&
1781 hdr.n_valid < 0xffff &&
1782 nb + HPTE_SIZE < count &&
1783 record_hpte(flags, hptp, hpte, revp, 1, first_pass)) {
1784 /* valid entry, write it out */
1786 if (__put_user(hpte[0], lbuf) ||
1787 __put_user(hpte[1], lbuf + 1))
1795 /* Now skip invalid entries while we can */
1796 while (i < kvmppc_hpt_npte(&kvm->arch.hpt) &&
1797 hdr.n_invalid < 0xffff &&
1798 record_hpte(flags, hptp, hpte, revp, 0, first_pass)) {
1799 /* found an invalid entry */
1806 if (hdr.n_valid || hdr.n_invalid) {
1807 /* write back the header */
1808 if (__copy_to_user(hptr, &hdr, sizeof(hdr)))
1811 buf = (char __user *)lbuf;
1816 /* Check if we've wrapped around the hash table */
1817 if (i >= kvmppc_hpt_npte(&kvm->arch.hpt)) {
1819 ctx->first_pass = 0;
1829 static ssize_t kvm_htab_write(struct file *file, const char __user *buf,
1830 size_t count, loff_t *ppos)
1832 struct kvm_htab_ctx *ctx = file->private_data;
1833 struct kvm *kvm = ctx->kvm;
1834 struct kvm_get_htab_header hdr;
1837 unsigned long __user *lbuf;
1839 unsigned long tmp[2];
1845 if (!access_ok(VERIFY_READ, buf, count))
1847 if (kvm_is_radix(kvm))
1850 /* lock out vcpus from running while we're doing this */
1851 mutex_lock(&kvm->lock);
1852 mmu_ready = kvm->arch.mmu_ready;
1854 kvm->arch.mmu_ready = 0; /* temporarily */
1855 /* order mmu_ready vs. vcpus_running */
1857 if (atomic_read(&kvm->arch.vcpus_running)) {
1858 kvm->arch.mmu_ready = 1;
1859 mutex_unlock(&kvm->lock);
1865 for (nb = 0; nb + sizeof(hdr) <= count; ) {
1867 if (__copy_from_user(&hdr, buf, sizeof(hdr)))
1871 if (nb + hdr.n_valid * HPTE_SIZE > count)
1879 if (i >= kvmppc_hpt_npte(&kvm->arch.hpt) ||
1880 i + hdr.n_valid + hdr.n_invalid > kvmppc_hpt_npte(&kvm->arch.hpt))
1883 hptp = (__be64 *)(kvm->arch.hpt.virt + (i * HPTE_SIZE));
1884 lbuf = (unsigned long __user *)buf;
1885 for (j = 0; j < hdr.n_valid; ++j) {
1890 if (__get_user(hpte_v, lbuf) ||
1891 __get_user(hpte_r, lbuf + 1))
1893 v = be64_to_cpu(hpte_v);
1894 r = be64_to_cpu(hpte_r);
1896 if (!(v & HPTE_V_VALID))
1898 pshift = kvmppc_hpte_base_page_shift(v, r);
1904 if (be64_to_cpu(hptp[0]) & (HPTE_V_VALID | HPTE_V_ABSENT))
1905 kvmppc_do_h_remove(kvm, 0, i, 0, tmp);
1907 ret = kvmppc_virtmode_do_h_enter(kvm, H_EXACT, i, v, r,
1909 if (ret != H_SUCCESS) {
1910 pr_err("kvm_htab_write ret %ld i=%ld v=%lx "
1911 "r=%lx\n", ret, i, v, r);
1914 if (!mmu_ready && is_vrma_hpte(v)) {
1915 unsigned long senc, lpcr;
1917 senc = slb_pgsize_encoding(1ul << pshift);
1918 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
1919 (VRMA_VSID << SLB_VSID_SHIFT_1T);
1920 if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
1921 lpcr = senc << (LPCR_VRMASD_SH - 4);
1922 kvmppc_update_lpcr(kvm, lpcr,
1925 kvmppc_setup_partition_table(kvm);
1933 for (j = 0; j < hdr.n_invalid; ++j) {
1934 if (be64_to_cpu(hptp[0]) & (HPTE_V_VALID | HPTE_V_ABSENT))
1935 kvmppc_do_h_remove(kvm, 0, i, 0, tmp);
1943 /* Order HPTE updates vs. mmu_ready */
1945 kvm->arch.mmu_ready = mmu_ready;
1946 mutex_unlock(&kvm->lock);
1953 static int kvm_htab_release(struct inode *inode, struct file *filp)
1955 struct kvm_htab_ctx *ctx = filp->private_data;
1957 filp->private_data = NULL;
1958 if (!(ctx->flags & KVM_GET_HTAB_WRITE))
1959 atomic_dec(&ctx->kvm->arch.hpte_mod_interest);
1960 kvm_put_kvm(ctx->kvm);
1965 static const struct file_operations kvm_htab_fops = {
1966 .read = kvm_htab_read,
1967 .write = kvm_htab_write,
1968 .llseek = default_llseek,
1969 .release = kvm_htab_release,
1972 int kvm_vm_ioctl_get_htab_fd(struct kvm *kvm, struct kvm_get_htab_fd *ghf)
1975 struct kvm_htab_ctx *ctx;
1978 /* reject flags we don't recognize */
1979 if (ghf->flags & ~(KVM_GET_HTAB_BOLTED_ONLY | KVM_GET_HTAB_WRITE))
1981 ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
1986 ctx->index = ghf->start_index;
1987 ctx->flags = ghf->flags;
1988 ctx->first_pass = 1;
1990 rwflag = (ghf->flags & KVM_GET_HTAB_WRITE) ? O_WRONLY : O_RDONLY;
1991 ret = anon_inode_getfd("kvm-htab", &kvm_htab_fops, ctx, rwflag | O_CLOEXEC);
1998 if (rwflag == O_RDONLY) {
1999 mutex_lock(&kvm->slots_lock);
2000 atomic_inc(&kvm->arch.hpte_mod_interest);
2001 /* make sure kvmppc_do_h_enter etc. see the increment */
2002 synchronize_srcu_expedited(&kvm->srcu);
2003 mutex_unlock(&kvm->slots_lock);
2009 struct debugfs_htab_state {
2012 unsigned long hpt_index;
2018 static int debugfs_htab_open(struct inode *inode, struct file *file)
2020 struct kvm *kvm = inode->i_private;
2021 struct debugfs_htab_state *p;
2023 p = kzalloc(sizeof(*p), GFP_KERNEL);
2029 mutex_init(&p->mutex);
2030 file->private_data = p;
2032 return nonseekable_open(inode, file);
2035 static int debugfs_htab_release(struct inode *inode, struct file *file)
2037 struct debugfs_htab_state *p = file->private_data;
2039 kvm_put_kvm(p->kvm);
2044 static ssize_t debugfs_htab_read(struct file *file, char __user *buf,
2045 size_t len, loff_t *ppos)
2047 struct debugfs_htab_state *p = file->private_data;
2050 unsigned long v, hr, gr;
2055 if (kvm_is_radix(kvm))
2058 ret = mutex_lock_interruptible(&p->mutex);
2062 if (p->chars_left) {
2066 r = copy_to_user(buf, p->buf + p->buf_index, n);
2081 hptp = (__be64 *)(kvm->arch.hpt.virt + (i * HPTE_SIZE));
2082 for (; len != 0 && i < kvmppc_hpt_npte(&kvm->arch.hpt);
2084 if (!(be64_to_cpu(hptp[0]) & (HPTE_V_VALID | HPTE_V_ABSENT)))
2087 /* lock the HPTE so it's stable and read it */
2089 while (!try_lock_hpte(hptp, HPTE_V_HVLOCK))
2091 v = be64_to_cpu(hptp[0]) & ~HPTE_V_HVLOCK;
2092 hr = be64_to_cpu(hptp[1]);
2093 gr = kvm->arch.hpt.rev[i].guest_rpte;
2094 unlock_hpte(hptp, v);
2097 if (!(v & (HPTE_V_VALID | HPTE_V_ABSENT)))
2100 n = scnprintf(p->buf, sizeof(p->buf),
2101 "%6lx %.16lx %.16lx %.16lx\n",
2106 r = copy_to_user(buf, p->buf, n);
2122 mutex_unlock(&p->mutex);
2126 static ssize_t debugfs_htab_write(struct file *file, const char __user *buf,
2127 size_t len, loff_t *ppos)
2132 static const struct file_operations debugfs_htab_fops = {
2133 .owner = THIS_MODULE,
2134 .open = debugfs_htab_open,
2135 .release = debugfs_htab_release,
2136 .read = debugfs_htab_read,
2137 .write = debugfs_htab_write,
2138 .llseek = generic_file_llseek,
2141 void kvmppc_mmu_debugfs_init(struct kvm *kvm)
2143 kvm->arch.htab_dentry = debugfs_create_file("htab", 0400,
2144 kvm->arch.debugfs_dir, kvm,
2145 &debugfs_htab_fops);
2148 void kvmppc_mmu_book3s_hv_init(struct kvm_vcpu *vcpu)
2150 struct kvmppc_mmu *mmu = &vcpu->arch.mmu;
2152 vcpu->arch.slb_nr = 32; /* POWER7/POWER8 */
2154 mmu->xlate = kvmppc_mmu_book3s_64_hv_xlate;
2155 mmu->reset_msr = kvmppc_mmu_book3s_64_hv_reset_msr;
2157 vcpu->arch.hflags |= BOOK3S_HFLAG_SLB;