1 // SPDX-License-Identifier: GPL-2.0-only
4 * Copyright 2010 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
7 #include <linux/types.h>
8 #include <linux/string.h>
10 #include <linux/kvm_host.h>
11 #include <linux/highmem.h>
12 #include <linux/gfp.h>
13 #include <linux/slab.h>
14 #include <linux/hugetlb.h>
15 #include <linux/vmalloc.h>
16 #include <linux/srcu.h>
17 #include <linux/anon_inodes.h>
18 #include <linux/file.h>
19 #include <linux/debugfs.h>
21 #include <asm/kvm_ppc.h>
22 #include <asm/kvm_book3s.h>
23 #include <asm/book3s/64/mmu-hash.h>
24 #include <asm/hvcall.h>
25 #include <asm/synch.h>
26 #include <asm/ppc-opcode.h>
27 #include <asm/cputable.h>
28 #include <asm/pte-walk.h>
33 //#define DEBUG_RESIZE_HPT 1
35 #ifdef DEBUG_RESIZE_HPT
36 #define resize_hpt_debug(resize, ...) \
38 printk(KERN_DEBUG "RESIZE HPT %p: ", resize); \
39 printk(__VA_ARGS__); \
42 #define resize_hpt_debug(resize, ...) \
46 static long kvmppc_virtmode_do_h_enter(struct kvm *kvm, unsigned long flags,
47 long pte_index, unsigned long pteh,
48 unsigned long ptel, unsigned long *pte_idx_ret);
50 struct kvm_resize_hpt {
51 /* These fields read-only after init */
53 struct work_struct work;
56 /* These fields protected by kvm->arch.mmu_setup_lock */
58 /* Possible values and their usage:
59 * <0 an error occurred during allocation,
60 * -EBUSY allocation is in the progress,
61 * 0 allocation made successfully.
65 /* Private to the work thread, until error != -EBUSY,
66 * then protected by kvm->arch.mmu_setup_lock.
68 struct kvm_hpt_info hpt;
71 int kvmppc_allocate_hpt(struct kvm_hpt_info *info, u32 order)
73 unsigned long hpt = 0;
75 struct page *page = NULL;
76 struct revmap_entry *rev;
79 if ((order < PPC_MIN_HPT_ORDER) || (order > PPC_MAX_HPT_ORDER))
82 page = kvm_alloc_hpt_cma(1ul << (order - PAGE_SHIFT));
84 hpt = (unsigned long)pfn_to_kaddr(page_to_pfn(page));
85 memset((void *)hpt, 0, (1ul << order));
90 hpt = __get_free_pages(GFP_KERNEL|__GFP_ZERO|__GFP_RETRY_MAYFAIL
91 |__GFP_NOWARN, order - PAGE_SHIFT);
96 /* HPTEs are 2**4 bytes long */
97 npte = 1ul << (order - 4);
99 /* Allocate reverse map array */
100 rev = vmalloc(array_size(npte, sizeof(struct revmap_entry)));
103 kvm_free_hpt_cma(page, 1 << (order - PAGE_SHIFT));
105 free_pages(hpt, order - PAGE_SHIFT);
117 void kvmppc_set_hpt(struct kvm *kvm, struct kvm_hpt_info *info)
119 atomic64_set(&kvm->arch.mmio_update, 0);
120 kvm->arch.hpt = *info;
121 kvm->arch.sdr1 = __pa(info->virt) | (info->order - 18);
123 pr_debug("KVM guest htab at %lx (order %ld), LPID %x\n",
124 info->virt, (long)info->order, kvm->arch.lpid);
127 int kvmppc_alloc_reset_hpt(struct kvm *kvm, int order)
130 struct kvm_hpt_info info;
132 mutex_lock(&kvm->arch.mmu_setup_lock);
133 if (kvm->arch.mmu_ready) {
134 kvm->arch.mmu_ready = 0;
135 /* order mmu_ready vs. vcpus_running */
137 if (atomic_read(&kvm->arch.vcpus_running)) {
138 kvm->arch.mmu_ready = 1;
142 if (kvm_is_radix(kvm)) {
143 err = kvmppc_switch_mmu_to_hpt(kvm);
148 if (kvm->arch.hpt.order == order) {
149 /* We already have a suitable HPT */
151 /* Set the entire HPT to 0, i.e. invalid HPTEs */
152 memset((void *)kvm->arch.hpt.virt, 0, 1ul << order);
154 * Reset all the reverse-mapping chains for all memslots
156 kvmppc_rmap_reset(kvm);
161 if (kvm->arch.hpt.virt) {
162 kvmppc_free_hpt(&kvm->arch.hpt);
163 kvmppc_rmap_reset(kvm);
166 err = kvmppc_allocate_hpt(&info, order);
169 kvmppc_set_hpt(kvm, &info);
173 /* Ensure that each vcpu will flush its TLB on next entry. */
174 cpumask_setall(&kvm->arch.need_tlb_flush);
176 mutex_unlock(&kvm->arch.mmu_setup_lock);
180 void kvmppc_free_hpt(struct kvm_hpt_info *info)
185 kvm_free_hpt_cma(virt_to_page(info->virt),
186 1 << (info->order - PAGE_SHIFT));
188 free_pages(info->virt, info->order - PAGE_SHIFT);
193 /* Bits in first HPTE dword for pagesize 4k, 64k or 16M */
194 static inline unsigned long hpte0_pgsize_encoding(unsigned long pgsize)
196 return (pgsize > 0x1000) ? HPTE_V_LARGE : 0;
199 /* Bits in second HPTE dword for pagesize 4k, 64k or 16M */
200 static inline unsigned long hpte1_pgsize_encoding(unsigned long pgsize)
202 return (pgsize == 0x10000) ? 0x1000 : 0;
205 void kvmppc_map_vrma(struct kvm_vcpu *vcpu, struct kvm_memory_slot *memslot,
206 unsigned long porder)
209 unsigned long npages;
210 unsigned long hp_v, hp_r;
211 unsigned long addr, hash;
213 unsigned long hp0, hp1;
214 unsigned long idx_ret;
216 struct kvm *kvm = vcpu->kvm;
218 psize = 1ul << porder;
219 npages = memslot->npages >> (porder - PAGE_SHIFT);
221 /* VRMA can't be > 1TB */
222 if (npages > 1ul << (40 - porder))
223 npages = 1ul << (40 - porder);
224 /* Can't use more than 1 HPTE per HPTEG */
225 if (npages > kvmppc_hpt_mask(&kvm->arch.hpt) + 1)
226 npages = kvmppc_hpt_mask(&kvm->arch.hpt) + 1;
228 hp0 = HPTE_V_1TB_SEG | (VRMA_VSID << (40 - 16)) |
229 HPTE_V_BOLTED | hpte0_pgsize_encoding(psize);
230 hp1 = hpte1_pgsize_encoding(psize) |
231 HPTE_R_R | HPTE_R_C | HPTE_R_M | PP_RWXX;
233 for (i = 0; i < npages; ++i) {
235 /* can't use hpt_hash since va > 64 bits */
236 hash = (i ^ (VRMA_VSID ^ (VRMA_VSID << 25)))
237 & kvmppc_hpt_mask(&kvm->arch.hpt);
239 * We assume that the hash table is empty and no
240 * vcpus are using it at this stage. Since we create
241 * at most one HPTE per HPTEG, we just assume entry 7
242 * is available and use it.
244 hash = (hash << 3) + 7;
245 hp_v = hp0 | ((addr >> 16) & ~0x7fUL);
247 ret = kvmppc_virtmode_do_h_enter(kvm, H_EXACT, hash, hp_v, hp_r,
249 if (ret != H_SUCCESS) {
250 pr_err("KVM: map_vrma at %lx failed, ret=%ld\n",
257 int kvmppc_mmu_hv_init(void)
259 unsigned long nr_lpids;
261 if (!mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE))
264 if (cpu_has_feature(CPU_FTR_HVMODE)) {
265 if (WARN_ON(mfspr(SPRN_LPID) != 0))
267 nr_lpids = 1UL << mmu_lpid_bits;
269 nr_lpids = 1UL << KVM_MAX_NESTED_GUESTS_SHIFT;
272 if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
273 /* POWER7 has 10-bit LPIDs, POWER8 has 12-bit LPIDs */
274 if (cpu_has_feature(CPU_FTR_ARCH_207S))
275 WARN_ON(nr_lpids != 1UL << 12);
277 WARN_ON(nr_lpids != 1UL << 10);
280 * Reserve the last implemented LPID use in partition
281 * switching for POWER7 and POWER8.
286 kvmppc_init_lpid(nr_lpids);
291 static long kvmppc_virtmode_do_h_enter(struct kvm *kvm, unsigned long flags,
292 long pte_index, unsigned long pteh,
293 unsigned long ptel, unsigned long *pte_idx_ret)
298 ret = kvmppc_do_h_enter(kvm, flags, pte_index, pteh, ptel,
299 kvm->mm->pgd, false, pte_idx_ret);
301 if (ret == H_TOO_HARD) {
302 /* this can't happen */
303 pr_err("KVM: Oops, kvmppc_h_enter returned too hard!\n");
304 ret = H_RESOURCE; /* or something */
310 static struct kvmppc_slb *kvmppc_mmu_book3s_hv_find_slbe(struct kvm_vcpu *vcpu,
316 for (i = 0; i < vcpu->arch.slb_nr; i++) {
317 if (!(vcpu->arch.slb[i].orige & SLB_ESID_V))
320 if (vcpu->arch.slb[i].origv & SLB_VSID_B_1T)
325 if (((vcpu->arch.slb[i].orige ^ eaddr) & mask) == 0)
326 return &vcpu->arch.slb[i];
331 static unsigned long kvmppc_mmu_get_real_addr(unsigned long v, unsigned long r,
334 unsigned long ra_mask;
336 ra_mask = kvmppc_actual_pgsz(v, r) - 1;
337 return (r & HPTE_R_RPN & ~ra_mask) | (ea & ra_mask);
340 static int kvmppc_mmu_book3s_64_hv_xlate(struct kvm_vcpu *vcpu, gva_t eaddr,
341 struct kvmppc_pte *gpte, bool data, bool iswrite)
343 struct kvm *kvm = vcpu->kvm;
344 struct kvmppc_slb *slbe;
346 unsigned long pp, key;
347 unsigned long v, orig_v, gr;
350 int virtmode = vcpu->arch.shregs.msr & (data ? MSR_DR : MSR_IR);
352 if (kvm_is_radix(vcpu->kvm))
353 return kvmppc_mmu_radix_xlate(vcpu, eaddr, gpte, data, iswrite);
357 slbe = kvmppc_mmu_book3s_hv_find_slbe(vcpu, eaddr);
362 /* real mode access */
363 slb_v = vcpu->kvm->arch.vrma_slb_v;
367 /* Find the HPTE in the hash table */
368 index = kvmppc_hv_find_lock_hpte(kvm, eaddr, slb_v,
369 HPTE_V_VALID | HPTE_V_ABSENT);
374 hptep = (__be64 *)(kvm->arch.hpt.virt + (index << 4));
375 v = orig_v = be64_to_cpu(hptep[0]) & ~HPTE_V_HVLOCK;
376 if (cpu_has_feature(CPU_FTR_ARCH_300))
377 v = hpte_new_to_old_v(v, be64_to_cpu(hptep[1]));
378 gr = kvm->arch.hpt.rev[index].guest_rpte;
380 unlock_hpte(hptep, orig_v);
384 gpte->vpage = ((v & HPTE_V_AVPN) << 4) | ((eaddr >> 12) & 0xfff);
386 /* Get PP bits and key for permission check */
387 pp = gr & (HPTE_R_PP0 | HPTE_R_PP);
388 key = (vcpu->arch.shregs.msr & MSR_PR) ? SLB_VSID_KP : SLB_VSID_KS;
391 /* Calculate permissions */
392 gpte->may_read = hpte_read_permission(pp, key);
393 gpte->may_write = hpte_write_permission(pp, key);
394 gpte->may_execute = gpte->may_read && !(gr & (HPTE_R_N | HPTE_R_G));
396 /* Storage key permission check for POWER7 */
397 if (data && virtmode) {
398 int amrfield = hpte_get_skey_perm(gr, vcpu->arch.amr);
405 /* Get the guest physical address */
406 gpte->raddr = kvmppc_mmu_get_real_addr(v, gr, eaddr);
411 * Quick test for whether an instruction is a load or a store.
412 * If the instruction is a load or a store, then this will indicate
413 * which it is, at least on server processors. (Embedded processors
414 * have some external PID instructions that don't follow the rule
415 * embodied here.) If the instruction isn't a load or store, then
416 * this doesn't return anything useful.
418 static int instruction_is_store(ppc_inst_t instr)
424 suffix = ppc_inst_val(instr);
425 if (ppc_inst_prefixed(instr))
426 suffix = ppc_inst_suffix(instr);
427 else if ((suffix & 0xfc000000) == 0x7c000000)
428 mask = 0x100; /* major opcode 31 */
429 return (suffix & mask) != 0;
432 int kvmppc_hv_emulate_mmio(struct kvm_vcpu *vcpu,
433 unsigned long gpa, gva_t ea, int is_store)
435 ppc_inst_t last_inst;
436 bool is_prefixed = !!(kvmppc_get_msr(vcpu) & SRR1_PREFIXED);
439 * Fast path - check if the guest physical address corresponds to a
440 * device on the FAST_MMIO_BUS, if so we can avoid loading the
441 * instruction all together, then we can just handle it and return.
446 idx = srcu_read_lock(&vcpu->kvm->srcu);
447 ret = kvm_io_bus_write(vcpu, KVM_FAST_MMIO_BUS, (gpa_t) gpa, 0,
449 srcu_read_unlock(&vcpu->kvm->srcu, idx);
451 kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + (is_prefixed ? 8 : 4));
457 * If we fail, we just return to the guest and try executing it again.
459 if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
464 * WARNING: We do not know for sure whether the instruction we just
465 * read from memory is the same that caused the fault in the first
468 * If the fault is prefixed but the instruction is not or vice
469 * versa, try again so that we don't advance pc the wrong amount.
471 if (ppc_inst_prefixed(last_inst) != is_prefixed)
475 * If the instruction we read is neither an load or a store,
476 * then it can't access memory, so we don't need to worry about
477 * enforcing access permissions. So, assuming it is a load or
478 * store, we just check that its direction (load or store) is
479 * consistent with the original fault, since that's what we
480 * checked the access permissions against. If there is a mismatch
481 * we just return and retry the instruction.
484 if (instruction_is_store(last_inst) != !!is_store)
488 * Emulated accesses are emulated by looking at the hash for
489 * translation once, then performing the access later. The
490 * translation could be invalidated in the meantime in which
491 * point performing the subsequent memory access on the old
492 * physical address could possibly be a security hole for the
493 * guest (but not the host).
495 * This is less of an issue for MMIO stores since they aren't
496 * globally visible. It could be an issue for MMIO loads to
497 * a certain extent but we'll ignore it for now.
500 vcpu->arch.paddr_accessed = gpa;
501 vcpu->arch.vaddr_accessed = ea;
502 return kvmppc_emulate_mmio(vcpu);
505 int kvmppc_book3s_hv_page_fault(struct kvm_vcpu *vcpu,
506 unsigned long ea, unsigned long dsisr)
508 struct kvm *kvm = vcpu->kvm;
509 unsigned long hpte[3], r;
510 unsigned long hnow_v, hnow_r;
512 unsigned long mmu_seq, psize, pte_size;
513 unsigned long gpa_base, gfn_base;
514 unsigned long gpa, gfn, hva, pfn, hpa;
515 struct kvm_memory_slot *memslot;
517 struct revmap_entry *rev;
521 bool writing, write_ok;
523 unsigned long rcbits;
527 if (kvm_is_radix(kvm))
528 return kvmppc_book3s_radix_page_fault(vcpu, ea, dsisr);
531 * Real-mode code has already searched the HPT and found the
532 * entry we're interested in. Lock the entry and check that
533 * it hasn't changed. If it has, just return and re-execute the
536 if (ea != vcpu->arch.pgfault_addr)
539 if (vcpu->arch.pgfault_cache) {
540 mmio_update = atomic64_read(&kvm->arch.mmio_update);
541 if (mmio_update == vcpu->arch.pgfault_cache->mmio_update) {
542 r = vcpu->arch.pgfault_cache->rpte;
543 psize = kvmppc_actual_pgsz(vcpu->arch.pgfault_hpte[0],
545 gpa_base = r & HPTE_R_RPN & ~(psize - 1);
546 gfn_base = gpa_base >> PAGE_SHIFT;
547 gpa = gpa_base | (ea & (psize - 1));
548 return kvmppc_hv_emulate_mmio(vcpu, gpa, ea,
549 dsisr & DSISR_ISSTORE);
552 index = vcpu->arch.pgfault_index;
553 hptep = (__be64 *)(kvm->arch.hpt.virt + (index << 4));
554 rev = &kvm->arch.hpt.rev[index];
556 while (!try_lock_hpte(hptep, HPTE_V_HVLOCK))
558 hpte[0] = be64_to_cpu(hptep[0]) & ~HPTE_V_HVLOCK;
559 hpte[1] = be64_to_cpu(hptep[1]);
560 hpte[2] = r = rev->guest_rpte;
561 unlock_hpte(hptep, hpte[0]);
564 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
565 hpte[0] = hpte_new_to_old_v(hpte[0], hpte[1]);
566 hpte[1] = hpte_new_to_old_r(hpte[1]);
568 if (hpte[0] != vcpu->arch.pgfault_hpte[0] ||
569 hpte[1] != vcpu->arch.pgfault_hpte[1])
572 /* Translate the logical address and get the page */
573 psize = kvmppc_actual_pgsz(hpte[0], r);
574 gpa_base = r & HPTE_R_RPN & ~(psize - 1);
575 gfn_base = gpa_base >> PAGE_SHIFT;
576 gpa = gpa_base | (ea & (psize - 1));
577 gfn = gpa >> PAGE_SHIFT;
578 memslot = gfn_to_memslot(kvm, gfn);
580 trace_kvm_page_fault_enter(vcpu, hpte, memslot, ea, dsisr);
582 /* No memslot means it's an emulated MMIO region */
583 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
584 return kvmppc_hv_emulate_mmio(vcpu, gpa, ea,
585 dsisr & DSISR_ISSTORE);
588 * This should never happen, because of the slot_is_aligned()
589 * check in kvmppc_do_h_enter().
591 if (gfn_base < memslot->base_gfn)
594 /* used to check for invalidations in progress */
595 mmu_seq = kvm->mmu_invalidate_seq;
600 writing = (dsisr & DSISR_ISSTORE) != 0;
601 /* If writing != 0, then the HPTE must allow writing, if we get here */
603 hva = gfn_to_hva_memslot(memslot, gfn);
606 * Do a fast check first, since __gfn_to_pfn_memslot doesn't
607 * do it with !atomic && !async, which is how we call it.
608 * We always ask for write permission since the common case
609 * is that the page is writable.
611 if (get_user_page_fast_only(hva, FOLL_WRITE, &page)) {
614 /* Call KVM generic code to do the slow-path check */
615 pfn = __gfn_to_pfn_memslot(memslot, gfn, false, false, NULL,
616 writing, &write_ok, NULL);
617 if (is_error_noslot_pfn(pfn))
620 if (pfn_valid(pfn)) {
621 page = pfn_to_page(pfn);
622 if (PageReserved(page))
628 * Read the PTE from the process' radix tree and use that
629 * so we get the shift and attribute bits.
631 spin_lock(&kvm->mmu_lock);
632 ptep = find_kvm_host_pte(kvm, mmu_seq, hva, &shift);
635 pte = READ_ONCE(*ptep);
636 spin_unlock(&kvm->mmu_lock);
638 * If the PTE disappeared temporarily due to a THP
639 * collapse, just return and let the guest try again.
641 if (!pte_present(pte)) {
646 hpa = pte_pfn(pte) << PAGE_SHIFT;
647 pte_size = PAGE_SIZE;
649 pte_size = 1ul << shift;
652 if (psize > pte_size)
654 if (pte_size > psize)
655 hpa |= hva & (pte_size - psize);
657 /* Check WIMG vs. the actual page we're accessing */
658 if (!hpte_cache_flags_ok(r, is_ci)) {
662 * Allow guest to map emulated device memory as
663 * uncacheable, but actually make it cacheable.
665 r = (r & ~(HPTE_R_W|HPTE_R_I|HPTE_R_G)) | HPTE_R_M;
669 * Set the HPTE to point to hpa.
670 * Since the hpa is at PAGE_SIZE granularity, make sure we
671 * don't mask out lower-order bits if psize < PAGE_SIZE.
673 if (psize < PAGE_SIZE)
675 r = (r & HPTE_R_KEY_HI) | (r & ~(HPTE_R_PP0 - psize)) | hpa;
676 if (hpte_is_writable(r) && !write_ok)
677 r = hpte_make_readonly(r);
680 while (!try_lock_hpte(hptep, HPTE_V_HVLOCK))
682 hnow_v = be64_to_cpu(hptep[0]);
683 hnow_r = be64_to_cpu(hptep[1]);
684 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
685 hnow_v = hpte_new_to_old_v(hnow_v, hnow_r);
686 hnow_r = hpte_new_to_old_r(hnow_r);
690 * If the HPT is being resized, don't update the HPTE,
691 * instead let the guest retry after the resize operation is complete.
692 * The synchronization for mmu_ready test vs. set is provided
695 if (!kvm->arch.mmu_ready)
698 if ((hnow_v & ~HPTE_V_HVLOCK) != hpte[0] || hnow_r != hpte[1] ||
699 rev->guest_rpte != hpte[2])
700 /* HPTE has been changed under us; let the guest retry */
702 hpte[0] = (hpte[0] & ~HPTE_V_ABSENT) | HPTE_V_VALID;
704 /* Always put the HPTE in the rmap chain for the page base address */
705 rmap = &memslot->arch.rmap[gfn_base - memslot->base_gfn];
708 /* Check if we might have been invalidated; let the guest retry if so */
710 if (mmu_invalidate_retry(vcpu->kvm, mmu_seq)) {
715 /* Only set R/C in real HPTE if set in both *rmap and guest_rpte */
716 rcbits = *rmap >> KVMPPC_RMAP_RC_SHIFT;
717 r &= rcbits | ~(HPTE_R_R | HPTE_R_C);
719 if (be64_to_cpu(hptep[0]) & HPTE_V_VALID) {
720 /* HPTE was previously valid, so we need to invalidate it */
722 hptep[0] |= cpu_to_be64(HPTE_V_ABSENT);
723 kvmppc_invalidate_hpte(kvm, hptep, index);
724 /* don't lose previous R and C bits */
725 r |= be64_to_cpu(hptep[1]) & (HPTE_R_R | HPTE_R_C);
727 kvmppc_add_revmap_chain(kvm, rev, rmap, index, 0);
730 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
731 r = hpte_old_to_new_r(hpte[0], r);
732 hpte[0] = hpte_old_to_new_v(hpte[0]);
734 hptep[1] = cpu_to_be64(r);
736 __unlock_hpte(hptep, hpte[0]);
737 asm volatile("ptesync" : : : "memory");
739 if (page && hpte_is_writable(r))
740 set_page_dirty_lock(page);
743 trace_kvm_page_fault_exit(vcpu, hpte, ret);
750 __unlock_hpte(hptep, be64_to_cpu(hptep[0]));
755 void kvmppc_rmap_reset(struct kvm *kvm)
757 struct kvm_memslots *slots;
758 struct kvm_memory_slot *memslot;
761 srcu_idx = srcu_read_lock(&kvm->srcu);
762 slots = kvm_memslots(kvm);
763 kvm_for_each_memslot(memslot, bkt, slots) {
764 /* Mutual exclusion with kvm_unmap_hva_range etc. */
765 spin_lock(&kvm->mmu_lock);
767 * This assumes it is acceptable to lose reference and
768 * change bits across a reset.
770 memset(memslot->arch.rmap, 0,
771 memslot->npages * sizeof(*memslot->arch.rmap));
772 spin_unlock(&kvm->mmu_lock);
774 srcu_read_unlock(&kvm->srcu, srcu_idx);
777 /* Must be called with both HPTE and rmap locked */
778 static void kvmppc_unmap_hpte(struct kvm *kvm, unsigned long i,
779 struct kvm_memory_slot *memslot,
780 unsigned long *rmapp, unsigned long gfn)
782 __be64 *hptep = (__be64 *) (kvm->arch.hpt.virt + (i << 4));
783 struct revmap_entry *rev = kvm->arch.hpt.rev;
785 unsigned long ptel, psize, rcbits;
789 /* chain is now empty */
790 *rmapp &= ~(KVMPPC_RMAP_PRESENT | KVMPPC_RMAP_INDEX);
792 /* remove i from chain */
796 rev[i].forw = rev[i].back = i;
797 *rmapp = (*rmapp & ~KVMPPC_RMAP_INDEX) | j;
800 /* Now check and modify the HPTE */
801 ptel = rev[i].guest_rpte;
802 psize = kvmppc_actual_pgsz(be64_to_cpu(hptep[0]), ptel);
803 if ((be64_to_cpu(hptep[0]) & HPTE_V_VALID) &&
804 hpte_rpn(ptel, psize) == gfn) {
805 hptep[0] |= cpu_to_be64(HPTE_V_ABSENT);
806 kvmppc_invalidate_hpte(kvm, hptep, i);
807 hptep[1] &= ~cpu_to_be64(HPTE_R_KEY_HI | HPTE_R_KEY_LO);
808 /* Harvest R and C */
809 rcbits = be64_to_cpu(hptep[1]) & (HPTE_R_R | HPTE_R_C);
810 *rmapp |= rcbits << KVMPPC_RMAP_RC_SHIFT;
811 if ((rcbits & HPTE_R_C) && memslot->dirty_bitmap)
812 kvmppc_update_dirty_map(memslot, gfn, psize);
813 if (rcbits & ~rev[i].guest_rpte) {
814 rev[i].guest_rpte = ptel | rcbits;
815 note_hpte_modification(kvm, &rev[i]);
820 static void kvm_unmap_rmapp(struct kvm *kvm, struct kvm_memory_slot *memslot,
825 unsigned long *rmapp;
827 rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn];
830 if (!(*rmapp & KVMPPC_RMAP_PRESENT)) {
836 * To avoid an ABBA deadlock with the HPTE lock bit,
837 * we can't spin on the HPTE lock while holding the
840 i = *rmapp & KVMPPC_RMAP_INDEX;
841 hptep = (__be64 *) (kvm->arch.hpt.virt + (i << 4));
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 kvmppc_unmap_hpte(kvm, i, memslot, rmapp, gfn);
852 __unlock_hpte(hptep, be64_to_cpu(hptep[0]));
856 bool kvm_unmap_gfn_range_hv(struct kvm *kvm, struct kvm_gfn_range *range)
860 if (kvm_is_radix(kvm)) {
861 for (gfn = range->start; gfn < range->end; gfn++)
862 kvm_unmap_radix(kvm, range->slot, gfn);
864 for (gfn = range->start; gfn < range->end; gfn++)
865 kvm_unmap_rmapp(kvm, range->slot, gfn);
871 void kvmppc_core_flush_memslot_hv(struct kvm *kvm,
872 struct kvm_memory_slot *memslot)
876 unsigned long *rmapp;
878 gfn = memslot->base_gfn;
879 rmapp = memslot->arch.rmap;
880 if (kvm_is_radix(kvm)) {
881 kvmppc_radix_flush_memslot(kvm, memslot);
885 for (n = memslot->npages; n; --n, ++gfn) {
887 * Testing the present bit without locking is OK because
888 * the memslot has been marked invalid already, and hence
889 * no new HPTEs referencing this page can be created,
890 * thus the present bit can't go from 0 to 1.
892 if (*rmapp & KVMPPC_RMAP_PRESENT)
893 kvm_unmap_rmapp(kvm, memslot, gfn);
898 static bool kvm_age_rmapp(struct kvm *kvm, struct kvm_memory_slot *memslot,
901 struct revmap_entry *rev = kvm->arch.hpt.rev;
902 unsigned long head, i, j;
905 unsigned long *rmapp;
907 rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn];
910 if (*rmapp & KVMPPC_RMAP_REFERENCED) {
911 *rmapp &= ~KVMPPC_RMAP_REFERENCED;
914 if (!(*rmapp & KVMPPC_RMAP_PRESENT)) {
919 i = head = *rmapp & KVMPPC_RMAP_INDEX;
921 hptep = (__be64 *) (kvm->arch.hpt.virt + (i << 4));
924 /* If this HPTE isn't referenced, ignore it */
925 if (!(be64_to_cpu(hptep[1]) & HPTE_R_R))
928 if (!try_lock_hpte(hptep, HPTE_V_HVLOCK)) {
929 /* unlock rmap before spinning on the HPTE lock */
931 while (be64_to_cpu(hptep[0]) & HPTE_V_HVLOCK)
936 /* Now check and modify the HPTE */
937 if ((be64_to_cpu(hptep[0]) & HPTE_V_VALID) &&
938 (be64_to_cpu(hptep[1]) & HPTE_R_R)) {
939 kvmppc_clear_ref_hpte(kvm, hptep, i);
940 if (!(rev[i].guest_rpte & HPTE_R_R)) {
941 rev[i].guest_rpte |= HPTE_R_R;
942 note_hpte_modification(kvm, &rev[i]);
946 __unlock_hpte(hptep, be64_to_cpu(hptep[0]));
947 } while ((i = j) != head);
953 bool kvm_age_gfn_hv(struct kvm *kvm, struct kvm_gfn_range *range)
958 if (kvm_is_radix(kvm)) {
959 for (gfn = range->start; gfn < range->end; gfn++)
960 ret |= kvm_age_radix(kvm, range->slot, gfn);
962 for (gfn = range->start; gfn < range->end; gfn++)
963 ret |= kvm_age_rmapp(kvm, range->slot, gfn);
969 static bool kvm_test_age_rmapp(struct kvm *kvm, struct kvm_memory_slot *memslot,
972 struct revmap_entry *rev = kvm->arch.hpt.rev;
973 unsigned long head, i, j;
976 unsigned long *rmapp;
978 rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn];
979 if (*rmapp & KVMPPC_RMAP_REFERENCED)
983 if (*rmapp & KVMPPC_RMAP_REFERENCED)
986 if (*rmapp & KVMPPC_RMAP_PRESENT) {
987 i = head = *rmapp & KVMPPC_RMAP_INDEX;
989 hp = (unsigned long *)(kvm->arch.hpt.virt + (i << 4));
991 if (be64_to_cpu(hp[1]) & HPTE_R_R)
993 } while ((i = j) != head);
1002 bool kvm_test_age_gfn_hv(struct kvm *kvm, struct kvm_gfn_range *range)
1004 WARN_ON(range->start + 1 != range->end);
1006 if (kvm_is_radix(kvm))
1007 return kvm_test_age_radix(kvm, range->slot, range->start);
1009 return kvm_test_age_rmapp(kvm, range->slot, range->start);
1012 bool kvm_set_spte_gfn_hv(struct kvm *kvm, struct kvm_gfn_range *range)
1014 WARN_ON(range->start + 1 != range->end);
1016 if (kvm_is_radix(kvm))
1017 kvm_unmap_radix(kvm, range->slot, range->start);
1019 kvm_unmap_rmapp(kvm, range->slot, range->start);
1024 static int vcpus_running(struct kvm *kvm)
1026 return atomic_read(&kvm->arch.vcpus_running) != 0;
1030 * Returns the number of system pages that are dirty.
1031 * This can be more than 1 if we find a huge-page HPTE.
1033 static int kvm_test_clear_dirty_npages(struct kvm *kvm, unsigned long *rmapp)
1035 struct revmap_entry *rev = kvm->arch.hpt.rev;
1036 unsigned long head, i, j;
1040 int npages_dirty = 0;
1044 if (!(*rmapp & KVMPPC_RMAP_PRESENT)) {
1046 return npages_dirty;
1049 i = head = *rmapp & KVMPPC_RMAP_INDEX;
1051 unsigned long hptep1;
1052 hptep = (__be64 *) (kvm->arch.hpt.virt + (i << 4));
1056 * Checking the C (changed) bit here is racy since there
1057 * is no guarantee about when the hardware writes it back.
1058 * If the HPTE is not writable then it is stable since the
1059 * page can't be written to, and we would have done a tlbie
1060 * (which forces the hardware to complete any writeback)
1061 * when making the HPTE read-only.
1062 * If vcpus are running then this call is racy anyway
1063 * since the page could get dirtied subsequently, so we
1064 * expect there to be a further call which would pick up
1065 * any delayed C bit writeback.
1066 * Otherwise we need to do the tlbie even if C==0 in
1067 * order to pick up any delayed writeback of C.
1069 hptep1 = be64_to_cpu(hptep[1]);
1070 if (!(hptep1 & HPTE_R_C) &&
1071 (!hpte_is_writable(hptep1) || vcpus_running(kvm)))
1074 if (!try_lock_hpte(hptep, HPTE_V_HVLOCK)) {
1075 /* unlock rmap before spinning on the HPTE lock */
1077 while (hptep[0] & cpu_to_be64(HPTE_V_HVLOCK))
1082 /* Now check and modify the HPTE */
1083 if (!(hptep[0] & cpu_to_be64(HPTE_V_VALID))) {
1084 __unlock_hpte(hptep, be64_to_cpu(hptep[0]));
1088 /* need to make it temporarily absent so C is stable */
1089 hptep[0] |= cpu_to_be64(HPTE_V_ABSENT);
1090 kvmppc_invalidate_hpte(kvm, hptep, i);
1091 v = be64_to_cpu(hptep[0]);
1092 r = be64_to_cpu(hptep[1]);
1094 hptep[1] = cpu_to_be64(r & ~HPTE_R_C);
1095 if (!(rev[i].guest_rpte & HPTE_R_C)) {
1096 rev[i].guest_rpte |= HPTE_R_C;
1097 note_hpte_modification(kvm, &rev[i]);
1099 n = kvmppc_actual_pgsz(v, r);
1100 n = (n + PAGE_SIZE - 1) >> PAGE_SHIFT;
1101 if (n > npages_dirty)
1105 v &= ~HPTE_V_ABSENT;
1107 __unlock_hpte(hptep, v);
1108 } while ((i = j) != head);
1111 return npages_dirty;
1114 void kvmppc_harvest_vpa_dirty(struct kvmppc_vpa *vpa,
1115 struct kvm_memory_slot *memslot,
1120 if (!vpa->dirty || !vpa->pinned_addr)
1122 gfn = vpa->gpa >> PAGE_SHIFT;
1123 if (gfn < memslot->base_gfn ||
1124 gfn >= memslot->base_gfn + memslot->npages)
1129 __set_bit_le(gfn - memslot->base_gfn, map);
1132 long kvmppc_hv_get_dirty_log_hpt(struct kvm *kvm,
1133 struct kvm_memory_slot *memslot, unsigned long *map)
1136 unsigned long *rmapp;
1139 rmapp = memslot->arch.rmap;
1140 for (i = 0; i < memslot->npages; ++i) {
1141 int npages = kvm_test_clear_dirty_npages(kvm, rmapp);
1143 * Note that if npages > 0 then i must be a multiple of npages,
1144 * since we always put huge-page HPTEs in the rmap chain
1145 * corresponding to their page base address.
1148 set_dirty_bits(map, i, npages);
1155 void *kvmppc_pin_guest_page(struct kvm *kvm, unsigned long gpa,
1156 unsigned long *nb_ret)
1158 struct kvm_memory_slot *memslot;
1159 unsigned long gfn = gpa >> PAGE_SHIFT;
1160 struct page *page, *pages[1];
1162 unsigned long hva, offset;
1165 srcu_idx = srcu_read_lock(&kvm->srcu);
1166 memslot = gfn_to_memslot(kvm, gfn);
1167 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
1169 hva = gfn_to_hva_memslot(memslot, gfn);
1170 npages = get_user_pages_fast(hva, 1, FOLL_WRITE, pages);
1174 srcu_read_unlock(&kvm->srcu, srcu_idx);
1176 offset = gpa & (PAGE_SIZE - 1);
1178 *nb_ret = PAGE_SIZE - offset;
1179 return page_address(page) + offset;
1182 srcu_read_unlock(&kvm->srcu, srcu_idx);
1186 void kvmppc_unpin_guest_page(struct kvm *kvm, void *va, unsigned long gpa,
1189 struct page *page = virt_to_page(va);
1190 struct kvm_memory_slot *memslot;
1199 /* We need to mark this page dirty in the memslot dirty_bitmap, if any */
1200 gfn = gpa >> PAGE_SHIFT;
1201 srcu_idx = srcu_read_lock(&kvm->srcu);
1202 memslot = gfn_to_memslot(kvm, gfn);
1203 if (memslot && memslot->dirty_bitmap)
1204 set_bit_le(gfn - memslot->base_gfn, memslot->dirty_bitmap);
1205 srcu_read_unlock(&kvm->srcu, srcu_idx);
1211 static int resize_hpt_allocate(struct kvm_resize_hpt *resize)
1215 rc = kvmppc_allocate_hpt(&resize->hpt, resize->order);
1219 resize_hpt_debug(resize, "%s(): HPT @ 0x%lx\n", __func__,
1225 static unsigned long resize_hpt_rehash_hpte(struct kvm_resize_hpt *resize,
1228 struct kvm *kvm = resize->kvm;
1229 struct kvm_hpt_info *old = &kvm->arch.hpt;
1230 struct kvm_hpt_info *new = &resize->hpt;
1231 unsigned long old_hash_mask = (1ULL << (old->order - 7)) - 1;
1232 unsigned long new_hash_mask = (1ULL << (new->order - 7)) - 1;
1233 __be64 *hptep, *new_hptep;
1234 unsigned long vpte, rpte, guest_rpte;
1236 struct revmap_entry *rev;
1237 unsigned long apsize, avpn, pteg, hash;
1238 unsigned long new_idx, new_pteg, replace_vpte;
1241 hptep = (__be64 *)(old->virt + (idx << 4));
1243 /* Guest is stopped, so new HPTEs can't be added or faulted
1244 * in, only unmapped or altered by host actions. So, it's
1245 * safe to check this before we take the HPTE lock */
1246 vpte = be64_to_cpu(hptep[0]);
1247 if (!(vpte & HPTE_V_VALID) && !(vpte & HPTE_V_ABSENT))
1248 return 0; /* nothing to do */
1250 while (!try_lock_hpte(hptep, HPTE_V_HVLOCK))
1253 vpte = be64_to_cpu(hptep[0]);
1256 if (!(vpte & HPTE_V_VALID) && !(vpte & HPTE_V_ABSENT))
1260 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
1261 rpte = be64_to_cpu(hptep[1]);
1262 vpte = hpte_new_to_old_v(vpte, rpte);
1266 rev = &old->rev[idx];
1267 guest_rpte = rev->guest_rpte;
1270 apsize = kvmppc_actual_pgsz(vpte, guest_rpte);
1274 if (vpte & HPTE_V_VALID) {
1275 unsigned long gfn = hpte_rpn(guest_rpte, apsize);
1276 int srcu_idx = srcu_read_lock(&kvm->srcu);
1277 struct kvm_memory_slot *memslot =
1278 __gfn_to_memslot(kvm_memslots(kvm), gfn);
1281 unsigned long *rmapp;
1282 rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn];
1285 kvmppc_unmap_hpte(kvm, idx, memslot, rmapp, gfn);
1289 srcu_read_unlock(&kvm->srcu, srcu_idx);
1292 /* Reload PTE after unmap */
1293 vpte = be64_to_cpu(hptep[0]);
1294 BUG_ON(vpte & HPTE_V_VALID);
1295 BUG_ON(!(vpte & HPTE_V_ABSENT));
1298 if (!(vpte & HPTE_V_BOLTED))
1301 rpte = be64_to_cpu(hptep[1]);
1303 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
1304 vpte = hpte_new_to_old_v(vpte, rpte);
1305 rpte = hpte_new_to_old_r(rpte);
1308 pshift = kvmppc_hpte_base_page_shift(vpte, rpte);
1309 avpn = HPTE_V_AVPN_VAL(vpte) & ~(((1ul << pshift) - 1) >> 23);
1310 pteg = idx / HPTES_PER_GROUP;
1311 if (vpte & HPTE_V_SECONDARY)
1314 if (!(vpte & HPTE_V_1TB_SEG)) {
1315 unsigned long offset, vsid;
1317 /* We only have 28 - 23 bits of offset in avpn */
1318 offset = (avpn & 0x1f) << 23;
1320 /* We can find more bits from the pteg value */
1322 offset |= ((vsid ^ pteg) & old_hash_mask) << pshift;
1324 hash = vsid ^ (offset >> pshift);
1326 unsigned long offset, vsid;
1328 /* We only have 40 - 23 bits of seg_off in avpn */
1329 offset = (avpn & 0x1ffff) << 23;
1332 offset |= ((vsid ^ (vsid << 25) ^ pteg) & old_hash_mask) << pshift;
1334 hash = vsid ^ (vsid << 25) ^ (offset >> pshift);
1337 new_pteg = hash & new_hash_mask;
1338 if (vpte & HPTE_V_SECONDARY)
1339 new_pteg = ~hash & new_hash_mask;
1341 new_idx = new_pteg * HPTES_PER_GROUP + (idx % HPTES_PER_GROUP);
1342 new_hptep = (__be64 *)(new->virt + (new_idx << 4));
1344 replace_vpte = be64_to_cpu(new_hptep[0]);
1345 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
1346 unsigned long replace_rpte = be64_to_cpu(new_hptep[1]);
1347 replace_vpte = hpte_new_to_old_v(replace_vpte, replace_rpte);
1350 if (replace_vpte & (HPTE_V_VALID | HPTE_V_ABSENT)) {
1351 BUG_ON(new->order >= old->order);
1353 if (replace_vpte & HPTE_V_BOLTED) {
1354 if (vpte & HPTE_V_BOLTED)
1355 /* Bolted collision, nothing we can do */
1357 /* Discard the new HPTE */
1361 /* Discard the previous HPTE */
1364 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
1365 rpte = hpte_old_to_new_r(vpte, rpte);
1366 vpte = hpte_old_to_new_v(vpte);
1369 new_hptep[1] = cpu_to_be64(rpte);
1370 new->rev[new_idx].guest_rpte = guest_rpte;
1371 /* No need for a barrier, since new HPT isn't active */
1372 new_hptep[0] = cpu_to_be64(vpte);
1373 unlock_hpte(new_hptep, vpte);
1376 unlock_hpte(hptep, vpte);
1380 static int resize_hpt_rehash(struct kvm_resize_hpt *resize)
1382 struct kvm *kvm = resize->kvm;
1386 for (i = 0; i < kvmppc_hpt_npte(&kvm->arch.hpt); i++) {
1387 rc = resize_hpt_rehash_hpte(resize, i);
1395 static void resize_hpt_pivot(struct kvm_resize_hpt *resize)
1397 struct kvm *kvm = resize->kvm;
1398 struct kvm_hpt_info hpt_tmp;
1400 /* Exchange the pending tables in the resize structure with
1401 * the active tables */
1403 resize_hpt_debug(resize, "resize_hpt_pivot()\n");
1405 spin_lock(&kvm->mmu_lock);
1406 asm volatile("ptesync" : : : "memory");
1408 hpt_tmp = kvm->arch.hpt;
1409 kvmppc_set_hpt(kvm, &resize->hpt);
1410 resize->hpt = hpt_tmp;
1412 spin_unlock(&kvm->mmu_lock);
1414 synchronize_srcu_expedited(&kvm->srcu);
1416 if (cpu_has_feature(CPU_FTR_ARCH_300))
1417 kvmppc_setup_partition_table(kvm);
1419 resize_hpt_debug(resize, "resize_hpt_pivot() done\n");
1422 static void resize_hpt_release(struct kvm *kvm, struct kvm_resize_hpt *resize)
1424 if (WARN_ON(!mutex_is_locked(&kvm->arch.mmu_setup_lock)))
1430 if (resize->error != -EBUSY) {
1431 if (resize->hpt.virt)
1432 kvmppc_free_hpt(&resize->hpt);
1436 if (kvm->arch.resize_hpt == resize)
1437 kvm->arch.resize_hpt = NULL;
1440 static void resize_hpt_prepare_work(struct work_struct *work)
1442 struct kvm_resize_hpt *resize = container_of(work,
1443 struct kvm_resize_hpt,
1445 struct kvm *kvm = resize->kvm;
1448 if (WARN_ON(resize->error != -EBUSY))
1451 mutex_lock(&kvm->arch.mmu_setup_lock);
1453 /* Request is still current? */
1454 if (kvm->arch.resize_hpt == resize) {
1455 /* We may request large allocations here:
1456 * do not sleep with kvm->arch.mmu_setup_lock held for a while.
1458 mutex_unlock(&kvm->arch.mmu_setup_lock);
1460 resize_hpt_debug(resize, "%s(): order = %d\n", __func__,
1463 err = resize_hpt_allocate(resize);
1465 /* We have strict assumption about -EBUSY
1466 * when preparing for HPT resize.
1468 if (WARN_ON(err == -EBUSY))
1471 mutex_lock(&kvm->arch.mmu_setup_lock);
1472 /* It is possible that kvm->arch.resize_hpt != resize
1473 * after we grab kvm->arch.mmu_setup_lock again.
1477 resize->error = err;
1479 if (kvm->arch.resize_hpt != resize)
1480 resize_hpt_release(kvm, resize);
1482 mutex_unlock(&kvm->arch.mmu_setup_lock);
1485 int kvm_vm_ioctl_resize_hpt_prepare(struct kvm *kvm,
1486 struct kvm_ppc_resize_hpt *rhpt)
1488 unsigned long flags = rhpt->flags;
1489 unsigned long shift = rhpt->shift;
1490 struct kvm_resize_hpt *resize;
1493 if (flags != 0 || kvm_is_radix(kvm))
1496 if (shift && ((shift < 18) || (shift > 46)))
1499 mutex_lock(&kvm->arch.mmu_setup_lock);
1501 resize = kvm->arch.resize_hpt;
1504 if (resize->order == shift) {
1505 /* Suitable resize in progress? */
1506 ret = resize->error;
1508 ret = 100; /* estimated time in ms */
1510 resize_hpt_release(kvm, resize);
1515 /* not suitable, cancel it */
1516 resize_hpt_release(kvm, resize);
1521 goto out; /* nothing to do */
1523 /* start new resize */
1525 resize = kzalloc(sizeof(*resize), GFP_KERNEL);
1531 resize->error = -EBUSY;
1532 resize->order = shift;
1534 INIT_WORK(&resize->work, resize_hpt_prepare_work);
1535 kvm->arch.resize_hpt = resize;
1537 schedule_work(&resize->work);
1539 ret = 100; /* estimated time in ms */
1542 mutex_unlock(&kvm->arch.mmu_setup_lock);
1546 static void resize_hpt_boot_vcpu(void *opaque)
1548 /* Nothing to do, just force a KVM exit */
1551 int kvm_vm_ioctl_resize_hpt_commit(struct kvm *kvm,
1552 struct kvm_ppc_resize_hpt *rhpt)
1554 unsigned long flags = rhpt->flags;
1555 unsigned long shift = rhpt->shift;
1556 struct kvm_resize_hpt *resize;
1559 if (flags != 0 || kvm_is_radix(kvm))
1562 if (shift && ((shift < 18) || (shift > 46)))
1565 mutex_lock(&kvm->arch.mmu_setup_lock);
1567 resize = kvm->arch.resize_hpt;
1569 /* This shouldn't be possible */
1571 if (WARN_ON(!kvm->arch.mmu_ready))
1574 /* Stop VCPUs from running while we mess with the HPT */
1575 kvm->arch.mmu_ready = 0;
1578 /* Boot all CPUs out of the guest so they re-read
1580 on_each_cpu(resize_hpt_boot_vcpu, NULL, 1);
1583 if (!resize || (resize->order != shift))
1586 ret = resize->error;
1590 ret = resize_hpt_rehash(resize);
1594 resize_hpt_pivot(resize);
1597 /* Let VCPUs run again */
1598 kvm->arch.mmu_ready = 1;
1601 resize_hpt_release(kvm, resize);
1602 mutex_unlock(&kvm->arch.mmu_setup_lock);
1607 * Functions for reading and writing the hash table via reads and
1608 * writes on a file descriptor.
1610 * Reads return the guest view of the hash table, which has to be
1611 * pieced together from the real hash table and the guest_rpte
1612 * values in the revmap array.
1614 * On writes, each HPTE written is considered in turn, and if it
1615 * is valid, it is written to the HPT as if an H_ENTER with the
1616 * exact flag set was done. When the invalid count is non-zero
1617 * in the header written to the stream, the kernel will make
1618 * sure that that many HPTEs are invalid, and invalidate them
1622 struct kvm_htab_ctx {
1623 unsigned long index;
1624 unsigned long flags;
1629 #define HPTE_SIZE (2 * sizeof(unsigned long))
1632 * Returns 1 if this HPT entry has been modified or has pending
1635 static int hpte_dirty(struct revmap_entry *revp, __be64 *hptp)
1637 unsigned long rcbits_unset;
1639 if (revp->guest_rpte & HPTE_GR_MODIFIED)
1642 /* Also need to consider changes in reference and changed bits */
1643 rcbits_unset = ~revp->guest_rpte & (HPTE_R_R | HPTE_R_C);
1644 if ((be64_to_cpu(hptp[0]) & HPTE_V_VALID) &&
1645 (be64_to_cpu(hptp[1]) & rcbits_unset))
1651 static long record_hpte(unsigned long flags, __be64 *hptp,
1652 unsigned long *hpte, struct revmap_entry *revp,
1653 int want_valid, int first_pass)
1655 unsigned long v, r, hr;
1656 unsigned long rcbits_unset;
1660 /* Unmodified entries are uninteresting except on the first pass */
1661 dirty = hpte_dirty(revp, hptp);
1662 if (!first_pass && !dirty)
1666 if (be64_to_cpu(hptp[0]) & (HPTE_V_VALID | HPTE_V_ABSENT)) {
1668 if ((flags & KVM_GET_HTAB_BOLTED_ONLY) &&
1669 !(be64_to_cpu(hptp[0]) & HPTE_V_BOLTED))
1672 if (valid != want_valid)
1676 if (valid || dirty) {
1677 /* lock the HPTE so it's stable and read it */
1679 while (!try_lock_hpte(hptp, HPTE_V_HVLOCK))
1681 v = be64_to_cpu(hptp[0]);
1682 hr = be64_to_cpu(hptp[1]);
1683 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
1684 v = hpte_new_to_old_v(v, hr);
1685 hr = hpte_new_to_old_r(hr);
1688 /* re-evaluate valid and dirty from synchronized HPTE value */
1689 valid = !!(v & HPTE_V_VALID);
1690 dirty = !!(revp->guest_rpte & HPTE_GR_MODIFIED);
1692 /* Harvest R and C into guest view if necessary */
1693 rcbits_unset = ~revp->guest_rpte & (HPTE_R_R | HPTE_R_C);
1694 if (valid && (rcbits_unset & hr)) {
1695 revp->guest_rpte |= (hr &
1696 (HPTE_R_R | HPTE_R_C)) | HPTE_GR_MODIFIED;
1700 if (v & HPTE_V_ABSENT) {
1701 v &= ~HPTE_V_ABSENT;
1705 if ((flags & KVM_GET_HTAB_BOLTED_ONLY) && !(v & HPTE_V_BOLTED))
1708 r = revp->guest_rpte;
1709 /* only clear modified if this is the right sort of entry */
1710 if (valid == want_valid && dirty) {
1711 r &= ~HPTE_GR_MODIFIED;
1712 revp->guest_rpte = r;
1714 unlock_hpte(hptp, be64_to_cpu(hptp[0]));
1716 if (!(valid == want_valid && (first_pass || dirty)))
1719 hpte[0] = cpu_to_be64(v);
1720 hpte[1] = cpu_to_be64(r);
1724 static ssize_t kvm_htab_read(struct file *file, char __user *buf,
1725 size_t count, loff_t *ppos)
1727 struct kvm_htab_ctx *ctx = file->private_data;
1728 struct kvm *kvm = ctx->kvm;
1729 struct kvm_get_htab_header hdr;
1731 struct revmap_entry *revp;
1732 unsigned long i, nb, nw;
1733 unsigned long __user *lbuf;
1734 struct kvm_get_htab_header __user *hptr;
1735 unsigned long flags;
1737 unsigned long hpte[2];
1739 if (!access_ok(buf, count))
1741 if (kvm_is_radix(kvm))
1744 first_pass = ctx->first_pass;
1748 hptp = (__be64 *)(kvm->arch.hpt.virt + (i * HPTE_SIZE));
1749 revp = kvm->arch.hpt.rev + i;
1750 lbuf = (unsigned long __user *)buf;
1753 while (nb + sizeof(hdr) + HPTE_SIZE < count) {
1754 /* Initialize header */
1755 hptr = (struct kvm_get_htab_header __user *)buf;
1760 lbuf = (unsigned long __user *)(buf + sizeof(hdr));
1762 /* Skip uninteresting entries, i.e. clean on not-first pass */
1764 while (i < kvmppc_hpt_npte(&kvm->arch.hpt) &&
1765 !hpte_dirty(revp, hptp)) {
1773 /* Grab a series of valid entries */
1774 while (i < kvmppc_hpt_npte(&kvm->arch.hpt) &&
1775 hdr.n_valid < 0xffff &&
1776 nb + HPTE_SIZE < count &&
1777 record_hpte(flags, hptp, hpte, revp, 1, first_pass)) {
1778 /* valid entry, write it out */
1780 if (__put_user(hpte[0], lbuf) ||
1781 __put_user(hpte[1], lbuf + 1))
1789 /* Now skip invalid entries while we can */
1790 while (i < kvmppc_hpt_npte(&kvm->arch.hpt) &&
1791 hdr.n_invalid < 0xffff &&
1792 record_hpte(flags, hptp, hpte, revp, 0, first_pass)) {
1793 /* found an invalid entry */
1800 if (hdr.n_valid || hdr.n_invalid) {
1801 /* write back the header */
1802 if (__copy_to_user(hptr, &hdr, sizeof(hdr)))
1805 buf = (char __user *)lbuf;
1810 /* Check if we've wrapped around the hash table */
1811 if (i >= kvmppc_hpt_npte(&kvm->arch.hpt)) {
1813 ctx->first_pass = 0;
1823 static ssize_t kvm_htab_write(struct file *file, const char __user *buf,
1824 size_t count, loff_t *ppos)
1826 struct kvm_htab_ctx *ctx = file->private_data;
1827 struct kvm *kvm = ctx->kvm;
1828 struct kvm_get_htab_header hdr;
1831 unsigned long __user *lbuf;
1833 unsigned long tmp[2];
1839 if (!access_ok(buf, count))
1841 if (kvm_is_radix(kvm))
1844 /* lock out vcpus from running while we're doing this */
1845 mutex_lock(&kvm->arch.mmu_setup_lock);
1846 mmu_ready = kvm->arch.mmu_ready;
1848 kvm->arch.mmu_ready = 0; /* temporarily */
1849 /* order mmu_ready vs. vcpus_running */
1851 if (atomic_read(&kvm->arch.vcpus_running)) {
1852 kvm->arch.mmu_ready = 1;
1853 mutex_unlock(&kvm->arch.mmu_setup_lock);
1859 for (nb = 0; nb + sizeof(hdr) <= count; ) {
1861 if (__copy_from_user(&hdr, buf, sizeof(hdr)))
1865 if (nb + hdr.n_valid * HPTE_SIZE > count)
1873 if (i >= kvmppc_hpt_npte(&kvm->arch.hpt) ||
1874 i + hdr.n_valid + hdr.n_invalid > kvmppc_hpt_npte(&kvm->arch.hpt))
1877 hptp = (__be64 *)(kvm->arch.hpt.virt + (i * HPTE_SIZE));
1878 lbuf = (unsigned long __user *)buf;
1879 for (j = 0; j < hdr.n_valid; ++j) {
1884 if (__get_user(hpte_v, lbuf) ||
1885 __get_user(hpte_r, lbuf + 1))
1887 v = be64_to_cpu(hpte_v);
1888 r = be64_to_cpu(hpte_r);
1890 if (!(v & HPTE_V_VALID))
1892 pshift = kvmppc_hpte_base_page_shift(v, r);
1898 if (be64_to_cpu(hptp[0]) & (HPTE_V_VALID | HPTE_V_ABSENT))
1899 kvmppc_do_h_remove(kvm, 0, i, 0, tmp);
1901 ret = kvmppc_virtmode_do_h_enter(kvm, H_EXACT, i, v, r,
1903 if (ret != H_SUCCESS) {
1904 pr_err("%s ret %ld i=%ld v=%lx r=%lx\n", __func__, ret, i, v, r);
1907 if (!mmu_ready && is_vrma_hpte(v)) {
1908 unsigned long senc, lpcr;
1910 senc = slb_pgsize_encoding(1ul << pshift);
1911 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
1912 (VRMA_VSID << SLB_VSID_SHIFT_1T);
1913 if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
1914 lpcr = senc << (LPCR_VRMASD_SH - 4);
1915 kvmppc_update_lpcr(kvm, lpcr,
1918 kvmppc_setup_partition_table(kvm);
1926 for (j = 0; j < hdr.n_invalid; ++j) {
1927 if (be64_to_cpu(hptp[0]) & (HPTE_V_VALID | HPTE_V_ABSENT))
1928 kvmppc_do_h_remove(kvm, 0, i, 0, tmp);
1936 /* Order HPTE updates vs. mmu_ready */
1938 kvm->arch.mmu_ready = mmu_ready;
1939 mutex_unlock(&kvm->arch.mmu_setup_lock);
1946 static int kvm_htab_release(struct inode *inode, struct file *filp)
1948 struct kvm_htab_ctx *ctx = filp->private_data;
1950 filp->private_data = NULL;
1951 if (!(ctx->flags & KVM_GET_HTAB_WRITE))
1952 atomic_dec(&ctx->kvm->arch.hpte_mod_interest);
1953 kvm_put_kvm(ctx->kvm);
1958 static const struct file_operations kvm_htab_fops = {
1959 .read = kvm_htab_read,
1960 .write = kvm_htab_write,
1961 .llseek = default_llseek,
1962 .release = kvm_htab_release,
1965 int kvm_vm_ioctl_get_htab_fd(struct kvm *kvm, struct kvm_get_htab_fd *ghf)
1968 struct kvm_htab_ctx *ctx;
1971 /* reject flags we don't recognize */
1972 if (ghf->flags & ~(KVM_GET_HTAB_BOLTED_ONLY | KVM_GET_HTAB_WRITE))
1974 ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
1979 ctx->index = ghf->start_index;
1980 ctx->flags = ghf->flags;
1981 ctx->first_pass = 1;
1983 rwflag = (ghf->flags & KVM_GET_HTAB_WRITE) ? O_WRONLY : O_RDONLY;
1984 ret = anon_inode_getfd("kvm-htab", &kvm_htab_fops, ctx, rwflag | O_CLOEXEC);
1987 kvm_put_kvm_no_destroy(kvm);
1991 if (rwflag == O_RDONLY) {
1992 mutex_lock(&kvm->slots_lock);
1993 atomic_inc(&kvm->arch.hpte_mod_interest);
1994 /* make sure kvmppc_do_h_enter etc. see the increment */
1995 synchronize_srcu_expedited(&kvm->srcu);
1996 mutex_unlock(&kvm->slots_lock);
2002 struct debugfs_htab_state {
2005 unsigned long hpt_index;
2011 static int debugfs_htab_open(struct inode *inode, struct file *file)
2013 struct kvm *kvm = inode->i_private;
2014 struct debugfs_htab_state *p;
2016 p = kzalloc(sizeof(*p), GFP_KERNEL);
2022 mutex_init(&p->mutex);
2023 file->private_data = p;
2025 return nonseekable_open(inode, file);
2028 static int debugfs_htab_release(struct inode *inode, struct file *file)
2030 struct debugfs_htab_state *p = file->private_data;
2032 kvm_put_kvm(p->kvm);
2037 static ssize_t debugfs_htab_read(struct file *file, char __user *buf,
2038 size_t len, loff_t *ppos)
2040 struct debugfs_htab_state *p = file->private_data;
2043 unsigned long v, hr, gr;
2048 if (kvm_is_radix(kvm))
2051 ret = mutex_lock_interruptible(&p->mutex);
2055 if (p->chars_left) {
2059 r = copy_to_user(buf, p->buf + p->buf_index, n);
2074 hptp = (__be64 *)(kvm->arch.hpt.virt + (i * HPTE_SIZE));
2075 for (; len != 0 && i < kvmppc_hpt_npte(&kvm->arch.hpt);
2077 if (!(be64_to_cpu(hptp[0]) & (HPTE_V_VALID | HPTE_V_ABSENT)))
2080 /* lock the HPTE so it's stable and read it */
2082 while (!try_lock_hpte(hptp, HPTE_V_HVLOCK))
2084 v = be64_to_cpu(hptp[0]) & ~HPTE_V_HVLOCK;
2085 hr = be64_to_cpu(hptp[1]);
2086 gr = kvm->arch.hpt.rev[i].guest_rpte;
2087 unlock_hpte(hptp, v);
2090 if (!(v & (HPTE_V_VALID | HPTE_V_ABSENT)))
2093 n = scnprintf(p->buf, sizeof(p->buf),
2094 "%6lx %.16lx %.16lx %.16lx\n",
2099 r = copy_to_user(buf, p->buf, n);
2115 mutex_unlock(&p->mutex);
2119 static ssize_t debugfs_htab_write(struct file *file, const char __user *buf,
2120 size_t len, loff_t *ppos)
2125 static const struct file_operations debugfs_htab_fops = {
2126 .owner = THIS_MODULE,
2127 .open = debugfs_htab_open,
2128 .release = debugfs_htab_release,
2129 .read = debugfs_htab_read,
2130 .write = debugfs_htab_write,
2131 .llseek = generic_file_llseek,
2134 void kvmppc_mmu_debugfs_init(struct kvm *kvm)
2136 debugfs_create_file("htab", 0400, kvm->debugfs_dentry, kvm,
2137 &debugfs_htab_fops);
2140 void kvmppc_mmu_book3s_hv_init(struct kvm_vcpu *vcpu)
2142 struct kvmppc_mmu *mmu = &vcpu->arch.mmu;
2144 vcpu->arch.slb_nr = 32; /* POWER7/POWER8 */
2146 mmu->xlate = kvmppc_mmu_book3s_64_hv_xlate;
2148 vcpu->arch.hflags |= BOOK3S_HFLAG_SLB;