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/kvm_ppc.h>
33 #include <asm/kvm_book3s.h>
34 #include <asm/book3s/64/mmu-hash.h>
35 #include <asm/hvcall.h>
36 #include <asm/synch.h>
37 #include <asm/ppc-opcode.h>
38 #include <asm/cputable.h>
39 #include <asm/pte-walk.h>
43 //#define DEBUG_RESIZE_HPT 1
45 #ifdef DEBUG_RESIZE_HPT
46 #define resize_hpt_debug(resize, ...) \
48 printk(KERN_DEBUG "RESIZE HPT %p: ", resize); \
49 printk(__VA_ARGS__); \
52 #define resize_hpt_debug(resize, ...) \
56 static long kvmppc_virtmode_do_h_enter(struct kvm *kvm, unsigned long flags,
57 long pte_index, unsigned long pteh,
58 unsigned long ptel, unsigned long *pte_idx_ret);
60 struct kvm_resize_hpt {
61 /* These fields read-only after init */
63 struct work_struct work;
66 /* These fields protected by kvm->lock */
68 /* Possible values and their usage:
69 * <0 an error occurred during allocation,
70 * -EBUSY allocation is in the progress,
71 * 0 allocation made successfuly.
75 /* Private to the work thread, until error != -EBUSY,
76 * then protected by kvm->lock.
78 struct kvm_hpt_info hpt;
81 int kvmppc_allocate_hpt(struct kvm_hpt_info *info, u32 order)
83 unsigned long hpt = 0;
85 struct page *page = NULL;
86 struct revmap_entry *rev;
89 if ((order < PPC_MIN_HPT_ORDER) || (order > PPC_MAX_HPT_ORDER))
92 page = kvm_alloc_hpt_cma(1ul << (order - PAGE_SHIFT));
94 hpt = (unsigned long)pfn_to_kaddr(page_to_pfn(page));
95 memset((void *)hpt, 0, (1ul << order));
100 hpt = __get_free_pages(GFP_KERNEL|__GFP_ZERO|__GFP_RETRY_MAYFAIL
101 |__GFP_NOWARN, order - PAGE_SHIFT);
106 /* HPTEs are 2**4 bytes long */
107 npte = 1ul << (order - 4);
109 /* Allocate reverse map array */
110 rev = vmalloc(array_size(npte, sizeof(struct revmap_entry)));
113 kvm_free_hpt_cma(page, 1 << (order - PAGE_SHIFT));
115 free_pages(hpt, order - PAGE_SHIFT);
127 void kvmppc_set_hpt(struct kvm *kvm, struct kvm_hpt_info *info)
129 atomic64_set(&kvm->arch.mmio_update, 0);
130 kvm->arch.hpt = *info;
131 kvm->arch.sdr1 = __pa(info->virt) | (info->order - 18);
133 pr_debug("KVM guest htab at %lx (order %ld), LPID %x\n",
134 info->virt, (long)info->order, kvm->arch.lpid);
137 long kvmppc_alloc_reset_hpt(struct kvm *kvm, int order)
140 struct kvm_hpt_info info;
142 mutex_lock(&kvm->lock);
143 if (kvm->arch.mmu_ready) {
144 kvm->arch.mmu_ready = 0;
145 /* order mmu_ready vs. vcpus_running */
147 if (atomic_read(&kvm->arch.vcpus_running)) {
148 kvm->arch.mmu_ready = 1;
152 if (kvm_is_radix(kvm)) {
153 err = kvmppc_switch_mmu_to_hpt(kvm);
158 if (kvm->arch.hpt.order == order) {
159 /* We already have a suitable HPT */
161 /* Set the entire HPT to 0, i.e. invalid HPTEs */
162 memset((void *)kvm->arch.hpt.virt, 0, 1ul << order);
164 * Reset all the reverse-mapping chains for all memslots
166 kvmppc_rmap_reset(kvm);
171 if (kvm->arch.hpt.virt) {
172 kvmppc_free_hpt(&kvm->arch.hpt);
173 kvmppc_rmap_reset(kvm);
176 err = kvmppc_allocate_hpt(&info, order);
179 kvmppc_set_hpt(kvm, &info);
183 /* Ensure that each vcpu will flush its TLB on next entry. */
184 cpumask_setall(&kvm->arch.need_tlb_flush);
186 mutex_unlock(&kvm->lock);
190 void kvmppc_free_hpt(struct kvm_hpt_info *info)
195 kvm_free_hpt_cma(virt_to_page(info->virt),
196 1 << (info->order - PAGE_SHIFT));
198 free_pages(info->virt, info->order - PAGE_SHIFT);
203 /* Bits in first HPTE dword for pagesize 4k, 64k or 16M */
204 static inline unsigned long hpte0_pgsize_encoding(unsigned long pgsize)
206 return (pgsize > 0x1000) ? HPTE_V_LARGE : 0;
209 /* Bits in second HPTE dword for pagesize 4k, 64k or 16M */
210 static inline unsigned long hpte1_pgsize_encoding(unsigned long pgsize)
212 return (pgsize == 0x10000) ? 0x1000 : 0;
215 void kvmppc_map_vrma(struct kvm_vcpu *vcpu, struct kvm_memory_slot *memslot,
216 unsigned long porder)
219 unsigned long npages;
220 unsigned long hp_v, hp_r;
221 unsigned long addr, hash;
223 unsigned long hp0, hp1;
224 unsigned long idx_ret;
226 struct kvm *kvm = vcpu->kvm;
228 psize = 1ul << porder;
229 npages = memslot->npages >> (porder - PAGE_SHIFT);
231 /* VRMA can't be > 1TB */
232 if (npages > 1ul << (40 - porder))
233 npages = 1ul << (40 - porder);
234 /* Can't use more than 1 HPTE per HPTEG */
235 if (npages > kvmppc_hpt_mask(&kvm->arch.hpt) + 1)
236 npages = kvmppc_hpt_mask(&kvm->arch.hpt) + 1;
238 hp0 = HPTE_V_1TB_SEG | (VRMA_VSID << (40 - 16)) |
239 HPTE_V_BOLTED | hpte0_pgsize_encoding(psize);
240 hp1 = hpte1_pgsize_encoding(psize) |
241 HPTE_R_R | HPTE_R_C | HPTE_R_M | PP_RWXX;
243 for (i = 0; i < npages; ++i) {
245 /* can't use hpt_hash since va > 64 bits */
246 hash = (i ^ (VRMA_VSID ^ (VRMA_VSID << 25)))
247 & kvmppc_hpt_mask(&kvm->arch.hpt);
249 * We assume that the hash table is empty and no
250 * vcpus are using it at this stage. Since we create
251 * at most one HPTE per HPTEG, we just assume entry 7
252 * is available and use it.
254 hash = (hash << 3) + 7;
255 hp_v = hp0 | ((addr >> 16) & ~0x7fUL);
257 ret = kvmppc_virtmode_do_h_enter(kvm, H_EXACT, hash, hp_v, hp_r,
259 if (ret != H_SUCCESS) {
260 pr_err("KVM: map_vrma at %lx failed, ret=%ld\n",
267 int kvmppc_mmu_hv_init(void)
269 unsigned long host_lpid, rsvd_lpid;
271 if (!mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE))
274 /* POWER7 has 10-bit LPIDs (12-bit in POWER8) */
276 if (cpu_has_feature(CPU_FTR_HVMODE))
277 host_lpid = mfspr(SPRN_LPID);
278 rsvd_lpid = LPID_RSVD;
280 kvmppc_init_lpid(rsvd_lpid + 1);
282 kvmppc_claim_lpid(host_lpid);
283 /* rsvd_lpid is reserved for use in partition switching */
284 kvmppc_claim_lpid(rsvd_lpid);
289 static void kvmppc_mmu_book3s_64_hv_reset_msr(struct kvm_vcpu *vcpu)
291 unsigned long msr = vcpu->arch.intr_msr;
293 /* If transactional, change to suspend mode on IRQ delivery */
294 if (MSR_TM_TRANSACTIONAL(vcpu->arch.shregs.msr))
297 msr |= vcpu->arch.shregs.msr & MSR_TS_MASK;
298 kvmppc_set_msr(vcpu, msr);
301 static long kvmppc_virtmode_do_h_enter(struct kvm *kvm, unsigned long flags,
302 long pte_index, unsigned long pteh,
303 unsigned long ptel, unsigned long *pte_idx_ret)
307 /* Protect linux PTE lookup from page table destruction */
308 rcu_read_lock_sched(); /* this disables preemption too */
309 ret = kvmppc_do_h_enter(kvm, flags, pte_index, pteh, ptel,
310 current->mm->pgd, false, pte_idx_ret);
311 rcu_read_unlock_sched();
312 if (ret == H_TOO_HARD) {
313 /* this can't happen */
314 pr_err("KVM: Oops, kvmppc_h_enter returned too hard!\n");
315 ret = H_RESOURCE; /* or something */
321 static struct kvmppc_slb *kvmppc_mmu_book3s_hv_find_slbe(struct kvm_vcpu *vcpu,
327 for (i = 0; i < vcpu->arch.slb_nr; i++) {
328 if (!(vcpu->arch.slb[i].orige & SLB_ESID_V))
331 if (vcpu->arch.slb[i].origv & SLB_VSID_B_1T)
336 if (((vcpu->arch.slb[i].orige ^ eaddr) & mask) == 0)
337 return &vcpu->arch.slb[i];
342 static unsigned long kvmppc_mmu_get_real_addr(unsigned long v, unsigned long r,
345 unsigned long ra_mask;
347 ra_mask = kvmppc_actual_pgsz(v, r) - 1;
348 return (r & HPTE_R_RPN & ~ra_mask) | (ea & ra_mask);
351 static int kvmppc_mmu_book3s_64_hv_xlate(struct kvm_vcpu *vcpu, gva_t eaddr,
352 struct kvmppc_pte *gpte, bool data, bool iswrite)
354 struct kvm *kvm = vcpu->kvm;
355 struct kvmppc_slb *slbe;
357 unsigned long pp, key;
358 unsigned long v, orig_v, gr;
361 int virtmode = vcpu->arch.shregs.msr & (data ? MSR_DR : MSR_IR);
363 if (kvm_is_radix(vcpu->kvm))
364 return kvmppc_mmu_radix_xlate(vcpu, eaddr, gpte, data, iswrite);
368 slbe = kvmppc_mmu_book3s_hv_find_slbe(vcpu, eaddr);
373 /* real mode access */
374 slb_v = vcpu->kvm->arch.vrma_slb_v;
378 /* Find the HPTE in the hash table */
379 index = kvmppc_hv_find_lock_hpte(kvm, eaddr, slb_v,
380 HPTE_V_VALID | HPTE_V_ABSENT);
385 hptep = (__be64 *)(kvm->arch.hpt.virt + (index << 4));
386 v = orig_v = be64_to_cpu(hptep[0]) & ~HPTE_V_HVLOCK;
387 if (cpu_has_feature(CPU_FTR_ARCH_300))
388 v = hpte_new_to_old_v(v, be64_to_cpu(hptep[1]));
389 gr = kvm->arch.hpt.rev[index].guest_rpte;
391 unlock_hpte(hptep, orig_v);
395 gpte->vpage = ((v & HPTE_V_AVPN) << 4) | ((eaddr >> 12) & 0xfff);
397 /* Get PP bits and key for permission check */
398 pp = gr & (HPTE_R_PP0 | HPTE_R_PP);
399 key = (vcpu->arch.shregs.msr & MSR_PR) ? SLB_VSID_KP : SLB_VSID_KS;
402 /* Calculate permissions */
403 gpte->may_read = hpte_read_permission(pp, key);
404 gpte->may_write = hpte_write_permission(pp, key);
405 gpte->may_execute = gpte->may_read && !(gr & (HPTE_R_N | HPTE_R_G));
407 /* Storage key permission check for POWER7 */
408 if (data && virtmode) {
409 int amrfield = hpte_get_skey_perm(gr, vcpu->arch.amr);
416 /* Get the guest physical address */
417 gpte->raddr = kvmppc_mmu_get_real_addr(v, gr, eaddr);
422 * Quick test for whether an instruction is a load or a store.
423 * If the instruction is a load or a store, then this will indicate
424 * which it is, at least on server processors. (Embedded processors
425 * have some external PID instructions that don't follow the rule
426 * embodied here.) If the instruction isn't a load or store, then
427 * this doesn't return anything useful.
429 static int instruction_is_store(unsigned int instr)
434 if ((instr & 0xfc000000) == 0x7c000000)
435 mask = 0x100; /* major opcode 31 */
436 return (instr & mask) != 0;
439 int kvmppc_hv_emulate_mmio(struct kvm_run *run, struct kvm_vcpu *vcpu,
440 unsigned long gpa, gva_t ea, int is_store)
445 * Fast path - check if the guest physical address corresponds to a
446 * device on the FAST_MMIO_BUS, if so we can avoid loading the
447 * instruction all together, then we can just handle it and return.
452 idx = srcu_read_lock(&vcpu->kvm->srcu);
453 ret = kvm_io_bus_write(vcpu, KVM_FAST_MMIO_BUS, (gpa_t) gpa, 0,
455 srcu_read_unlock(&vcpu->kvm->srcu, idx);
457 kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4);
463 * If we fail, we just return to the guest and try executing it again.
465 if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
470 * WARNING: We do not know for sure whether the instruction we just
471 * read from memory is the same that caused the fault in the first
472 * place. If the instruction we read is neither an load or a store,
473 * then it can't access memory, so we don't need to worry about
474 * enforcing access permissions. So, assuming it is a load or
475 * store, we just check that its direction (load or store) is
476 * consistent with the original fault, since that's what we
477 * checked the access permissions against. If there is a mismatch
478 * we just return and retry the instruction.
481 if (instruction_is_store(last_inst) != !!is_store)
485 * Emulated accesses are emulated by looking at the hash for
486 * translation once, then performing the access later. The
487 * translation could be invalidated in the meantime in which
488 * point performing the subsequent memory access on the old
489 * physical address could possibly be a security hole for the
490 * guest (but not the host).
492 * This is less of an issue for MMIO stores since they aren't
493 * globally visible. It could be an issue for MMIO loads to
494 * a certain extent but we'll ignore it for now.
497 vcpu->arch.paddr_accessed = gpa;
498 vcpu->arch.vaddr_accessed = ea;
499 return kvmppc_emulate_mmio(run, vcpu);
502 int kvmppc_book3s_hv_page_fault(struct kvm_run *run, struct kvm_vcpu *vcpu,
503 unsigned long ea, unsigned long dsisr)
505 struct kvm *kvm = vcpu->kvm;
506 unsigned long hpte[3], r;
507 unsigned long hnow_v, hnow_r;
509 unsigned long mmu_seq, psize, pte_size;
510 unsigned long gpa_base, gfn_base;
511 unsigned long gpa, gfn, hva, pfn;
512 struct kvm_memory_slot *memslot;
514 struct revmap_entry *rev;
515 struct page *page, *pages[1];
516 long index, ret, npages;
518 unsigned int writing, write_ok;
519 struct vm_area_struct *vma;
520 unsigned long rcbits;
523 if (kvm_is_radix(kvm))
524 return kvmppc_book3s_radix_page_fault(run, vcpu, ea, dsisr);
527 * Real-mode code has already searched the HPT and found the
528 * entry we're interested in. Lock the entry and check that
529 * it hasn't changed. If it has, just return and re-execute the
532 if (ea != vcpu->arch.pgfault_addr)
535 if (vcpu->arch.pgfault_cache) {
536 mmio_update = atomic64_read(&kvm->arch.mmio_update);
537 if (mmio_update == vcpu->arch.pgfault_cache->mmio_update) {
538 r = vcpu->arch.pgfault_cache->rpte;
539 psize = kvmppc_actual_pgsz(vcpu->arch.pgfault_hpte[0],
541 gpa_base = r & HPTE_R_RPN & ~(psize - 1);
542 gfn_base = gpa_base >> PAGE_SHIFT;
543 gpa = gpa_base | (ea & (psize - 1));
544 return kvmppc_hv_emulate_mmio(run, vcpu, gpa, ea,
545 dsisr & DSISR_ISSTORE);
548 index = vcpu->arch.pgfault_index;
549 hptep = (__be64 *)(kvm->arch.hpt.virt + (index << 4));
550 rev = &kvm->arch.hpt.rev[index];
552 while (!try_lock_hpte(hptep, HPTE_V_HVLOCK))
554 hpte[0] = be64_to_cpu(hptep[0]) & ~HPTE_V_HVLOCK;
555 hpte[1] = be64_to_cpu(hptep[1]);
556 hpte[2] = r = rev->guest_rpte;
557 unlock_hpte(hptep, hpte[0]);
560 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
561 hpte[0] = hpte_new_to_old_v(hpte[0], hpte[1]);
562 hpte[1] = hpte_new_to_old_r(hpte[1]);
564 if (hpte[0] != vcpu->arch.pgfault_hpte[0] ||
565 hpte[1] != vcpu->arch.pgfault_hpte[1])
568 /* Translate the logical address and get the page */
569 psize = kvmppc_actual_pgsz(hpte[0], r);
570 gpa_base = r & HPTE_R_RPN & ~(psize - 1);
571 gfn_base = gpa_base >> PAGE_SHIFT;
572 gpa = gpa_base | (ea & (psize - 1));
573 gfn = gpa >> PAGE_SHIFT;
574 memslot = gfn_to_memslot(kvm, gfn);
576 trace_kvm_page_fault_enter(vcpu, hpte, memslot, ea, dsisr);
578 /* No memslot means it's an emulated MMIO region */
579 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
580 return kvmppc_hv_emulate_mmio(run, vcpu, gpa, ea,
581 dsisr & DSISR_ISSTORE);
584 * This should never happen, because of the slot_is_aligned()
585 * check in kvmppc_do_h_enter().
587 if (gfn_base < memslot->base_gfn)
590 /* used to check for invalidations in progress */
591 mmu_seq = kvm->mmu_notifier_seq;
598 pte_size = PAGE_SIZE;
599 writing = (dsisr & DSISR_ISSTORE) != 0;
600 /* If writing != 0, then the HPTE must allow writing, if we get here */
602 hva = gfn_to_hva_memslot(memslot, gfn);
603 npages = get_user_pages_fast(hva, 1, writing, pages);
605 /* Check if it's an I/O mapping */
606 down_read(¤t->mm->mmap_sem);
607 vma = find_vma(current->mm, hva);
608 if (vma && vma->vm_start <= hva && hva + psize <= vma->vm_end &&
609 (vma->vm_flags & VM_PFNMAP)) {
610 pfn = vma->vm_pgoff +
611 ((hva - vma->vm_start) >> PAGE_SHIFT);
613 is_ci = pte_ci(__pte((pgprot_val(vma->vm_page_prot))));
614 write_ok = vma->vm_flags & VM_WRITE;
616 up_read(¤t->mm->mmap_sem);
621 pfn = page_to_pfn(page);
622 if (PageHuge(page)) {
623 page = compound_head(page);
624 pte_size <<= compound_order(page);
626 /* if the guest wants write access, see if that is OK */
627 if (!writing && hpte_is_writable(r)) {
631 * We need to protect against page table destruction
632 * hugepage split and collapse.
634 local_irq_save(flags);
635 ptep = find_current_mm_pte(current->mm->pgd,
638 pte = kvmppc_read_update_linux_pte(ptep, 1);
639 if (__pte_write(pte))
642 local_irq_restore(flags);
646 if (psize > pte_size)
649 /* Check WIMG vs. the actual page we're accessing */
650 if (!hpte_cache_flags_ok(r, is_ci)) {
654 * Allow guest to map emulated device memory as
655 * uncacheable, but actually make it cacheable.
657 r = (r & ~(HPTE_R_W|HPTE_R_I|HPTE_R_G)) | HPTE_R_M;
661 * Set the HPTE to point to pfn.
662 * Since the pfn is at PAGE_SIZE granularity, make sure we
663 * don't mask out lower-order bits if psize < PAGE_SIZE.
665 if (psize < PAGE_SIZE)
667 r = (r & HPTE_R_KEY_HI) | (r & ~(HPTE_R_PP0 - psize)) |
668 ((pfn << PAGE_SHIFT) & ~(psize - 1));
669 if (hpte_is_writable(r) && !write_ok)
670 r = hpte_make_readonly(r);
673 while (!try_lock_hpte(hptep, HPTE_V_HVLOCK))
675 hnow_v = be64_to_cpu(hptep[0]);
676 hnow_r = be64_to_cpu(hptep[1]);
677 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
678 hnow_v = hpte_new_to_old_v(hnow_v, hnow_r);
679 hnow_r = hpte_new_to_old_r(hnow_r);
683 * If the HPT is being resized, don't update the HPTE,
684 * instead let the guest retry after the resize operation is complete.
685 * The synchronization for mmu_ready test vs. set is provided
688 if (!kvm->arch.mmu_ready)
691 if ((hnow_v & ~HPTE_V_HVLOCK) != hpte[0] || hnow_r != hpte[1] ||
692 rev->guest_rpte != hpte[2])
693 /* HPTE has been changed under us; let the guest retry */
695 hpte[0] = (hpte[0] & ~HPTE_V_ABSENT) | HPTE_V_VALID;
697 /* Always put the HPTE in the rmap chain for the page base address */
698 rmap = &memslot->arch.rmap[gfn_base - memslot->base_gfn];
701 /* Check if we might have been invalidated; let the guest retry if so */
703 if (mmu_notifier_retry(vcpu->kvm, mmu_seq)) {
708 /* Only set R/C in real HPTE if set in both *rmap and guest_rpte */
709 rcbits = *rmap >> KVMPPC_RMAP_RC_SHIFT;
710 r &= rcbits | ~(HPTE_R_R | HPTE_R_C);
712 if (be64_to_cpu(hptep[0]) & HPTE_V_VALID) {
713 /* HPTE was previously valid, so we need to invalidate it */
715 hptep[0] |= cpu_to_be64(HPTE_V_ABSENT);
716 kvmppc_invalidate_hpte(kvm, hptep, index);
717 /* don't lose previous R and C bits */
718 r |= be64_to_cpu(hptep[1]) & (HPTE_R_R | HPTE_R_C);
720 kvmppc_add_revmap_chain(kvm, rev, rmap, index, 0);
723 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
724 r = hpte_old_to_new_r(hpte[0], r);
725 hpte[0] = hpte_old_to_new_v(hpte[0]);
727 hptep[1] = cpu_to_be64(r);
729 __unlock_hpte(hptep, hpte[0]);
730 asm volatile("ptesync" : : : "memory");
732 if (page && hpte_is_writable(r))
736 trace_kvm_page_fault_exit(vcpu, hpte, ret);
740 * We drop pages[0] here, not page because page might
741 * have been set to the head page of a compound, but
742 * we have to drop the reference on the correct tail
743 * page to match the get inside gup()
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, 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 typedef int (*hva_handler_fn)(struct kvm *kvm, struct kvm_memory_slot *memslot,
780 static int kvm_handle_hva_range(struct kvm *kvm,
783 hva_handler_fn handler)
787 struct kvm_memslots *slots;
788 struct kvm_memory_slot *memslot;
790 slots = kvm_memslots(kvm);
791 kvm_for_each_memslot(memslot, slots) {
792 unsigned long hva_start, hva_end;
795 hva_start = max(start, memslot->userspace_addr);
796 hva_end = min(end, memslot->userspace_addr +
797 (memslot->npages << PAGE_SHIFT));
798 if (hva_start >= hva_end)
801 * {gfn(page) | page intersects with [hva_start, hva_end)} =
802 * {gfn, gfn+1, ..., gfn_end-1}.
804 gfn = hva_to_gfn_memslot(hva_start, memslot);
805 gfn_end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, memslot);
807 for (; gfn < gfn_end; ++gfn) {
808 ret = handler(kvm, memslot, gfn);
816 static int kvm_handle_hva(struct kvm *kvm, unsigned long hva,
817 hva_handler_fn handler)
819 return kvm_handle_hva_range(kvm, hva, hva + 1, handler);
822 /* Must be called with both HPTE and rmap locked */
823 static void kvmppc_unmap_hpte(struct kvm *kvm, unsigned long i,
824 struct kvm_memory_slot *memslot,
825 unsigned long *rmapp, unsigned long gfn)
827 __be64 *hptep = (__be64 *) (kvm->arch.hpt.virt + (i << 4));
828 struct revmap_entry *rev = kvm->arch.hpt.rev;
830 unsigned long ptel, psize, rcbits;
834 /* chain is now empty */
835 *rmapp &= ~(KVMPPC_RMAP_PRESENT | KVMPPC_RMAP_INDEX);
837 /* remove i from chain */
841 rev[i].forw = rev[i].back = i;
842 *rmapp = (*rmapp & ~KVMPPC_RMAP_INDEX) | j;
845 /* Now check and modify the HPTE */
846 ptel = rev[i].guest_rpte;
847 psize = kvmppc_actual_pgsz(be64_to_cpu(hptep[0]), ptel);
848 if ((be64_to_cpu(hptep[0]) & HPTE_V_VALID) &&
849 hpte_rpn(ptel, psize) == gfn) {
850 hptep[0] |= cpu_to_be64(HPTE_V_ABSENT);
851 kvmppc_invalidate_hpte(kvm, hptep, i);
852 hptep[1] &= ~cpu_to_be64(HPTE_R_KEY_HI | HPTE_R_KEY_LO);
853 /* Harvest R and C */
854 rcbits = be64_to_cpu(hptep[1]) & (HPTE_R_R | HPTE_R_C);
855 *rmapp |= rcbits << KVMPPC_RMAP_RC_SHIFT;
856 if ((rcbits & HPTE_R_C) && memslot->dirty_bitmap)
857 kvmppc_update_dirty_map(memslot, gfn, psize);
858 if (rcbits & ~rev[i].guest_rpte) {
859 rev[i].guest_rpte = ptel | rcbits;
860 note_hpte_modification(kvm, &rev[i]);
865 static int kvm_unmap_rmapp(struct kvm *kvm, struct kvm_memory_slot *memslot,
870 unsigned long *rmapp;
872 rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn];
875 if (!(*rmapp & KVMPPC_RMAP_PRESENT)) {
881 * To avoid an ABBA deadlock with the HPTE lock bit,
882 * we can't spin on the HPTE lock while holding the
885 i = *rmapp & KVMPPC_RMAP_INDEX;
886 hptep = (__be64 *) (kvm->arch.hpt.virt + (i << 4));
887 if (!try_lock_hpte(hptep, HPTE_V_HVLOCK)) {
888 /* unlock rmap before spinning on the HPTE lock */
890 while (be64_to_cpu(hptep[0]) & HPTE_V_HVLOCK)
895 kvmppc_unmap_hpte(kvm, i, memslot, rmapp, gfn);
897 __unlock_hpte(hptep, be64_to_cpu(hptep[0]));
902 int kvm_unmap_hva_range_hv(struct kvm *kvm, unsigned long start, unsigned long end)
904 hva_handler_fn handler;
906 handler = kvm_is_radix(kvm) ? kvm_unmap_radix : kvm_unmap_rmapp;
907 kvm_handle_hva_range(kvm, start, end, handler);
911 void kvmppc_core_flush_memslot_hv(struct kvm *kvm,
912 struct kvm_memory_slot *memslot)
916 unsigned long *rmapp;
918 gfn = memslot->base_gfn;
919 rmapp = memslot->arch.rmap;
920 if (kvm_is_radix(kvm)) {
921 kvmppc_radix_flush_memslot(kvm, memslot);
925 for (n = memslot->npages; n; --n, ++gfn) {
927 * Testing the present bit without locking is OK because
928 * the memslot has been marked invalid already, and hence
929 * no new HPTEs referencing this page can be created,
930 * thus the present bit can't go from 0 to 1.
932 if (*rmapp & KVMPPC_RMAP_PRESENT)
933 kvm_unmap_rmapp(kvm, memslot, gfn);
938 static int kvm_age_rmapp(struct kvm *kvm, struct kvm_memory_slot *memslot,
941 struct revmap_entry *rev = kvm->arch.hpt.rev;
942 unsigned long head, i, j;
945 unsigned long *rmapp;
947 rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn];
950 if (*rmapp & KVMPPC_RMAP_REFERENCED) {
951 *rmapp &= ~KVMPPC_RMAP_REFERENCED;
954 if (!(*rmapp & KVMPPC_RMAP_PRESENT)) {
959 i = head = *rmapp & KVMPPC_RMAP_INDEX;
961 hptep = (__be64 *) (kvm->arch.hpt.virt + (i << 4));
964 /* If this HPTE isn't referenced, ignore it */
965 if (!(be64_to_cpu(hptep[1]) & HPTE_R_R))
968 if (!try_lock_hpte(hptep, HPTE_V_HVLOCK)) {
969 /* unlock rmap before spinning on the HPTE lock */
971 while (be64_to_cpu(hptep[0]) & HPTE_V_HVLOCK)
976 /* Now check and modify the HPTE */
977 if ((be64_to_cpu(hptep[0]) & HPTE_V_VALID) &&
978 (be64_to_cpu(hptep[1]) & HPTE_R_R)) {
979 kvmppc_clear_ref_hpte(kvm, hptep, i);
980 if (!(rev[i].guest_rpte & HPTE_R_R)) {
981 rev[i].guest_rpte |= HPTE_R_R;
982 note_hpte_modification(kvm, &rev[i]);
986 __unlock_hpte(hptep, be64_to_cpu(hptep[0]));
987 } while ((i = j) != head);
993 int kvm_age_hva_hv(struct kvm *kvm, unsigned long start, unsigned long end)
995 hva_handler_fn handler;
997 handler = kvm_is_radix(kvm) ? kvm_age_radix : kvm_age_rmapp;
998 return kvm_handle_hva_range(kvm, start, end, handler);
1001 static int kvm_test_age_rmapp(struct kvm *kvm, struct kvm_memory_slot *memslot,
1004 struct revmap_entry *rev = kvm->arch.hpt.rev;
1005 unsigned long head, i, j;
1008 unsigned long *rmapp;
1010 rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn];
1011 if (*rmapp & KVMPPC_RMAP_REFERENCED)
1015 if (*rmapp & KVMPPC_RMAP_REFERENCED)
1018 if (*rmapp & KVMPPC_RMAP_PRESENT) {
1019 i = head = *rmapp & KVMPPC_RMAP_INDEX;
1021 hp = (unsigned long *)(kvm->arch.hpt.virt + (i << 4));
1023 if (be64_to_cpu(hp[1]) & HPTE_R_R)
1025 } while ((i = j) != head);
1034 int kvm_test_age_hva_hv(struct kvm *kvm, unsigned long hva)
1036 hva_handler_fn handler;
1038 handler = kvm_is_radix(kvm) ? kvm_test_age_radix : kvm_test_age_rmapp;
1039 return kvm_handle_hva(kvm, hva, handler);
1042 void kvm_set_spte_hva_hv(struct kvm *kvm, unsigned long hva, pte_t pte)
1044 hva_handler_fn handler;
1046 handler = kvm_is_radix(kvm) ? kvm_unmap_radix : kvm_unmap_rmapp;
1047 kvm_handle_hva(kvm, hva, handler);
1050 static int vcpus_running(struct kvm *kvm)
1052 return atomic_read(&kvm->arch.vcpus_running) != 0;
1056 * Returns the number of system pages that are dirty.
1057 * This can be more than 1 if we find a huge-page HPTE.
1059 static int kvm_test_clear_dirty_npages(struct kvm *kvm, unsigned long *rmapp)
1061 struct revmap_entry *rev = kvm->arch.hpt.rev;
1062 unsigned long head, i, j;
1066 int npages_dirty = 0;
1070 if (!(*rmapp & KVMPPC_RMAP_PRESENT)) {
1072 return npages_dirty;
1075 i = head = *rmapp & KVMPPC_RMAP_INDEX;
1077 unsigned long hptep1;
1078 hptep = (__be64 *) (kvm->arch.hpt.virt + (i << 4));
1082 * Checking the C (changed) bit here is racy since there
1083 * is no guarantee about when the hardware writes it back.
1084 * If the HPTE is not writable then it is stable since the
1085 * page can't be written to, and we would have done a tlbie
1086 * (which forces the hardware to complete any writeback)
1087 * when making the HPTE read-only.
1088 * If vcpus are running then this call is racy anyway
1089 * since the page could get dirtied subsequently, so we
1090 * expect there to be a further call which would pick up
1091 * any delayed C bit writeback.
1092 * Otherwise we need to do the tlbie even if C==0 in
1093 * order to pick up any delayed writeback of C.
1095 hptep1 = be64_to_cpu(hptep[1]);
1096 if (!(hptep1 & HPTE_R_C) &&
1097 (!hpte_is_writable(hptep1) || vcpus_running(kvm)))
1100 if (!try_lock_hpte(hptep, HPTE_V_HVLOCK)) {
1101 /* unlock rmap before spinning on the HPTE lock */
1103 while (hptep[0] & cpu_to_be64(HPTE_V_HVLOCK))
1108 /* Now check and modify the HPTE */
1109 if (!(hptep[0] & cpu_to_be64(HPTE_V_VALID))) {
1110 __unlock_hpte(hptep, be64_to_cpu(hptep[0]));
1114 /* need to make it temporarily absent so C is stable */
1115 hptep[0] |= cpu_to_be64(HPTE_V_ABSENT);
1116 kvmppc_invalidate_hpte(kvm, hptep, i);
1117 v = be64_to_cpu(hptep[0]);
1118 r = be64_to_cpu(hptep[1]);
1120 hptep[1] = cpu_to_be64(r & ~HPTE_R_C);
1121 if (!(rev[i].guest_rpte & HPTE_R_C)) {
1122 rev[i].guest_rpte |= HPTE_R_C;
1123 note_hpte_modification(kvm, &rev[i]);
1125 n = kvmppc_actual_pgsz(v, r);
1126 n = (n + PAGE_SIZE - 1) >> PAGE_SHIFT;
1127 if (n > npages_dirty)
1131 v &= ~HPTE_V_ABSENT;
1133 __unlock_hpte(hptep, v);
1134 } while ((i = j) != head);
1137 return npages_dirty;
1140 void kvmppc_harvest_vpa_dirty(struct kvmppc_vpa *vpa,
1141 struct kvm_memory_slot *memslot,
1146 if (!vpa->dirty || !vpa->pinned_addr)
1148 gfn = vpa->gpa >> PAGE_SHIFT;
1149 if (gfn < memslot->base_gfn ||
1150 gfn >= memslot->base_gfn + memslot->npages)
1155 __set_bit_le(gfn - memslot->base_gfn, map);
1158 long kvmppc_hv_get_dirty_log_hpt(struct kvm *kvm,
1159 struct kvm_memory_slot *memslot, unsigned long *map)
1162 unsigned long *rmapp;
1165 rmapp = memslot->arch.rmap;
1166 for (i = 0; i < memslot->npages; ++i) {
1167 int npages = kvm_test_clear_dirty_npages(kvm, rmapp);
1169 * Note that if npages > 0 then i must be a multiple of npages,
1170 * since we always put huge-page HPTEs in the rmap chain
1171 * corresponding to their page base address.
1174 set_dirty_bits(map, i, npages);
1181 void *kvmppc_pin_guest_page(struct kvm *kvm, unsigned long gpa,
1182 unsigned long *nb_ret)
1184 struct kvm_memory_slot *memslot;
1185 unsigned long gfn = gpa >> PAGE_SHIFT;
1186 struct page *page, *pages[1];
1188 unsigned long hva, offset;
1191 srcu_idx = srcu_read_lock(&kvm->srcu);
1192 memslot = gfn_to_memslot(kvm, gfn);
1193 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
1195 hva = gfn_to_hva_memslot(memslot, gfn);
1196 npages = get_user_pages_fast(hva, 1, 1, pages);
1200 srcu_read_unlock(&kvm->srcu, srcu_idx);
1202 offset = gpa & (PAGE_SIZE - 1);
1204 *nb_ret = PAGE_SIZE - offset;
1205 return page_address(page) + offset;
1208 srcu_read_unlock(&kvm->srcu, srcu_idx);
1212 void kvmppc_unpin_guest_page(struct kvm *kvm, void *va, unsigned long gpa,
1215 struct page *page = virt_to_page(va);
1216 struct kvm_memory_slot *memslot;
1225 /* We need to mark this page dirty in the memslot dirty_bitmap, if any */
1226 gfn = gpa >> PAGE_SHIFT;
1227 srcu_idx = srcu_read_lock(&kvm->srcu);
1228 memslot = gfn_to_memslot(kvm, gfn);
1229 if (memslot && memslot->dirty_bitmap)
1230 set_bit_le(gfn - memslot->base_gfn, memslot->dirty_bitmap);
1231 srcu_read_unlock(&kvm->srcu, srcu_idx);
1237 static int resize_hpt_allocate(struct kvm_resize_hpt *resize)
1241 rc = kvmppc_allocate_hpt(&resize->hpt, resize->order);
1245 resize_hpt_debug(resize, "resize_hpt_allocate(): HPT @ 0x%lx\n",
1251 static unsigned long resize_hpt_rehash_hpte(struct kvm_resize_hpt *resize,
1254 struct kvm *kvm = resize->kvm;
1255 struct kvm_hpt_info *old = &kvm->arch.hpt;
1256 struct kvm_hpt_info *new = &resize->hpt;
1257 unsigned long old_hash_mask = (1ULL << (old->order - 7)) - 1;
1258 unsigned long new_hash_mask = (1ULL << (new->order - 7)) - 1;
1259 __be64 *hptep, *new_hptep;
1260 unsigned long vpte, rpte, guest_rpte;
1262 struct revmap_entry *rev;
1263 unsigned long apsize, avpn, pteg, hash;
1264 unsigned long new_idx, new_pteg, replace_vpte;
1267 hptep = (__be64 *)(old->virt + (idx << 4));
1269 /* Guest is stopped, so new HPTEs can't be added or faulted
1270 * in, only unmapped or altered by host actions. So, it's
1271 * safe to check this before we take the HPTE lock */
1272 vpte = be64_to_cpu(hptep[0]);
1273 if (!(vpte & HPTE_V_VALID) && !(vpte & HPTE_V_ABSENT))
1274 return 0; /* nothing to do */
1276 while (!try_lock_hpte(hptep, HPTE_V_HVLOCK))
1279 vpte = be64_to_cpu(hptep[0]);
1282 if (!(vpte & HPTE_V_VALID) && !(vpte & HPTE_V_ABSENT))
1286 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
1287 rpte = be64_to_cpu(hptep[1]);
1288 vpte = hpte_new_to_old_v(vpte, rpte);
1292 rev = &old->rev[idx];
1293 guest_rpte = rev->guest_rpte;
1296 apsize = kvmppc_actual_pgsz(vpte, guest_rpte);
1300 if (vpte & HPTE_V_VALID) {
1301 unsigned long gfn = hpte_rpn(guest_rpte, apsize);
1302 int srcu_idx = srcu_read_lock(&kvm->srcu);
1303 struct kvm_memory_slot *memslot =
1304 __gfn_to_memslot(kvm_memslots(kvm), gfn);
1307 unsigned long *rmapp;
1308 rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn];
1311 kvmppc_unmap_hpte(kvm, idx, memslot, rmapp, gfn);
1315 srcu_read_unlock(&kvm->srcu, srcu_idx);
1318 /* Reload PTE after unmap */
1319 vpte = be64_to_cpu(hptep[0]);
1320 BUG_ON(vpte & HPTE_V_VALID);
1321 BUG_ON(!(vpte & HPTE_V_ABSENT));
1324 if (!(vpte & HPTE_V_BOLTED))
1327 rpte = be64_to_cpu(hptep[1]);
1329 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
1330 vpte = hpte_new_to_old_v(vpte, rpte);
1331 rpte = hpte_new_to_old_r(rpte);
1334 pshift = kvmppc_hpte_base_page_shift(vpte, rpte);
1335 avpn = HPTE_V_AVPN_VAL(vpte) & ~(((1ul << pshift) - 1) >> 23);
1336 pteg = idx / HPTES_PER_GROUP;
1337 if (vpte & HPTE_V_SECONDARY)
1340 if (!(vpte & HPTE_V_1TB_SEG)) {
1341 unsigned long offset, vsid;
1343 /* We only have 28 - 23 bits of offset in avpn */
1344 offset = (avpn & 0x1f) << 23;
1346 /* We can find more bits from the pteg value */
1348 offset |= ((vsid ^ pteg) & old_hash_mask) << pshift;
1350 hash = vsid ^ (offset >> pshift);
1352 unsigned long offset, vsid;
1354 /* We only have 40 - 23 bits of seg_off in avpn */
1355 offset = (avpn & 0x1ffff) << 23;
1358 offset |= ((vsid ^ (vsid << 25) ^ pteg) & old_hash_mask) << pshift;
1360 hash = vsid ^ (vsid << 25) ^ (offset >> pshift);
1363 new_pteg = hash & new_hash_mask;
1364 if (vpte & HPTE_V_SECONDARY)
1365 new_pteg = ~hash & new_hash_mask;
1367 new_idx = new_pteg * HPTES_PER_GROUP + (idx % HPTES_PER_GROUP);
1368 new_hptep = (__be64 *)(new->virt + (new_idx << 4));
1370 replace_vpte = be64_to_cpu(new_hptep[0]);
1371 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
1372 unsigned long replace_rpte = be64_to_cpu(new_hptep[1]);
1373 replace_vpte = hpte_new_to_old_v(replace_vpte, replace_rpte);
1376 if (replace_vpte & (HPTE_V_VALID | HPTE_V_ABSENT)) {
1377 BUG_ON(new->order >= old->order);
1379 if (replace_vpte & HPTE_V_BOLTED) {
1380 if (vpte & HPTE_V_BOLTED)
1381 /* Bolted collision, nothing we can do */
1383 /* Discard the new HPTE */
1387 /* Discard the previous HPTE */
1390 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
1391 rpte = hpte_old_to_new_r(vpte, rpte);
1392 vpte = hpte_old_to_new_v(vpte);
1395 new_hptep[1] = cpu_to_be64(rpte);
1396 new->rev[new_idx].guest_rpte = guest_rpte;
1397 /* No need for a barrier, since new HPT isn't active */
1398 new_hptep[0] = cpu_to_be64(vpte);
1399 unlock_hpte(new_hptep, vpte);
1402 unlock_hpte(hptep, vpte);
1406 static int resize_hpt_rehash(struct kvm_resize_hpt *resize)
1408 struct kvm *kvm = resize->kvm;
1412 for (i = 0; i < kvmppc_hpt_npte(&kvm->arch.hpt); i++) {
1413 rc = resize_hpt_rehash_hpte(resize, i);
1421 static void resize_hpt_pivot(struct kvm_resize_hpt *resize)
1423 struct kvm *kvm = resize->kvm;
1424 struct kvm_hpt_info hpt_tmp;
1426 /* Exchange the pending tables in the resize structure with
1427 * the active tables */
1429 resize_hpt_debug(resize, "resize_hpt_pivot()\n");
1431 spin_lock(&kvm->mmu_lock);
1432 asm volatile("ptesync" : : : "memory");
1434 hpt_tmp = kvm->arch.hpt;
1435 kvmppc_set_hpt(kvm, &resize->hpt);
1436 resize->hpt = hpt_tmp;
1438 spin_unlock(&kvm->mmu_lock);
1440 synchronize_srcu_expedited(&kvm->srcu);
1442 if (cpu_has_feature(CPU_FTR_ARCH_300))
1443 kvmppc_setup_partition_table(kvm);
1445 resize_hpt_debug(resize, "resize_hpt_pivot() done\n");
1448 static void resize_hpt_release(struct kvm *kvm, struct kvm_resize_hpt *resize)
1450 if (WARN_ON(!mutex_is_locked(&kvm->lock)))
1456 if (resize->error != -EBUSY) {
1457 if (resize->hpt.virt)
1458 kvmppc_free_hpt(&resize->hpt);
1462 if (kvm->arch.resize_hpt == resize)
1463 kvm->arch.resize_hpt = NULL;
1466 static void resize_hpt_prepare_work(struct work_struct *work)
1468 struct kvm_resize_hpt *resize = container_of(work,
1469 struct kvm_resize_hpt,
1471 struct kvm *kvm = resize->kvm;
1474 if (WARN_ON(resize->error != -EBUSY))
1477 mutex_lock(&kvm->lock);
1479 /* Request is still current? */
1480 if (kvm->arch.resize_hpt == resize) {
1481 /* We may request large allocations here:
1482 * do not sleep with kvm->lock held for a while.
1484 mutex_unlock(&kvm->lock);
1486 resize_hpt_debug(resize, "resize_hpt_prepare_work(): order = %d\n",
1489 err = resize_hpt_allocate(resize);
1491 /* We have strict assumption about -EBUSY
1492 * when preparing for HPT resize.
1494 if (WARN_ON(err == -EBUSY))
1497 mutex_lock(&kvm->lock);
1498 /* It is possible that kvm->arch.resize_hpt != resize
1499 * after we grab kvm->lock again.
1503 resize->error = err;
1505 if (kvm->arch.resize_hpt != resize)
1506 resize_hpt_release(kvm, resize);
1508 mutex_unlock(&kvm->lock);
1511 long kvm_vm_ioctl_resize_hpt_prepare(struct kvm *kvm,
1512 struct kvm_ppc_resize_hpt *rhpt)
1514 unsigned long flags = rhpt->flags;
1515 unsigned long shift = rhpt->shift;
1516 struct kvm_resize_hpt *resize;
1519 if (flags != 0 || kvm_is_radix(kvm))
1522 if (shift && ((shift < 18) || (shift > 46)))
1525 mutex_lock(&kvm->lock);
1527 resize = kvm->arch.resize_hpt;
1530 if (resize->order == shift) {
1531 /* Suitable resize in progress? */
1532 ret = resize->error;
1534 ret = 100; /* estimated time in ms */
1536 resize_hpt_release(kvm, resize);
1541 /* not suitable, cancel it */
1542 resize_hpt_release(kvm, resize);
1547 goto out; /* nothing to do */
1549 /* start new resize */
1551 resize = kzalloc(sizeof(*resize), GFP_KERNEL);
1557 resize->error = -EBUSY;
1558 resize->order = shift;
1560 INIT_WORK(&resize->work, resize_hpt_prepare_work);
1561 kvm->arch.resize_hpt = resize;
1563 schedule_work(&resize->work);
1565 ret = 100; /* estimated time in ms */
1568 mutex_unlock(&kvm->lock);
1572 static void resize_hpt_boot_vcpu(void *opaque)
1574 /* Nothing to do, just force a KVM exit */
1577 long kvm_vm_ioctl_resize_hpt_commit(struct kvm *kvm,
1578 struct kvm_ppc_resize_hpt *rhpt)
1580 unsigned long flags = rhpt->flags;
1581 unsigned long shift = rhpt->shift;
1582 struct kvm_resize_hpt *resize;
1585 if (flags != 0 || kvm_is_radix(kvm))
1588 if (shift && ((shift < 18) || (shift > 46)))
1591 mutex_lock(&kvm->lock);
1593 resize = kvm->arch.resize_hpt;
1595 /* This shouldn't be possible */
1597 if (WARN_ON(!kvm->arch.mmu_ready))
1600 /* Stop VCPUs from running while we mess with the HPT */
1601 kvm->arch.mmu_ready = 0;
1604 /* Boot all CPUs out of the guest so they re-read
1606 on_each_cpu(resize_hpt_boot_vcpu, NULL, 1);
1609 if (!resize || (resize->order != shift))
1612 ret = resize->error;
1616 ret = resize_hpt_rehash(resize);
1620 resize_hpt_pivot(resize);
1623 /* Let VCPUs run again */
1624 kvm->arch.mmu_ready = 1;
1627 resize_hpt_release(kvm, resize);
1628 mutex_unlock(&kvm->lock);
1633 * Functions for reading and writing the hash table via reads and
1634 * writes on a file descriptor.
1636 * Reads return the guest view of the hash table, which has to be
1637 * pieced together from the real hash table and the guest_rpte
1638 * values in the revmap array.
1640 * On writes, each HPTE written is considered in turn, and if it
1641 * is valid, it is written to the HPT as if an H_ENTER with the
1642 * exact flag set was done. When the invalid count is non-zero
1643 * in the header written to the stream, the kernel will make
1644 * sure that that many HPTEs are invalid, and invalidate them
1648 struct kvm_htab_ctx {
1649 unsigned long index;
1650 unsigned long flags;
1655 #define HPTE_SIZE (2 * sizeof(unsigned long))
1658 * Returns 1 if this HPT entry has been modified or has pending
1661 static int hpte_dirty(struct revmap_entry *revp, __be64 *hptp)
1663 unsigned long rcbits_unset;
1665 if (revp->guest_rpte & HPTE_GR_MODIFIED)
1668 /* Also need to consider changes in reference and changed bits */
1669 rcbits_unset = ~revp->guest_rpte & (HPTE_R_R | HPTE_R_C);
1670 if ((be64_to_cpu(hptp[0]) & HPTE_V_VALID) &&
1671 (be64_to_cpu(hptp[1]) & rcbits_unset))
1677 static long record_hpte(unsigned long flags, __be64 *hptp,
1678 unsigned long *hpte, struct revmap_entry *revp,
1679 int want_valid, int first_pass)
1681 unsigned long v, r, hr;
1682 unsigned long rcbits_unset;
1686 /* Unmodified entries are uninteresting except on the first pass */
1687 dirty = hpte_dirty(revp, hptp);
1688 if (!first_pass && !dirty)
1692 if (be64_to_cpu(hptp[0]) & (HPTE_V_VALID | HPTE_V_ABSENT)) {
1694 if ((flags & KVM_GET_HTAB_BOLTED_ONLY) &&
1695 !(be64_to_cpu(hptp[0]) & HPTE_V_BOLTED))
1698 if (valid != want_valid)
1702 if (valid || dirty) {
1703 /* lock the HPTE so it's stable and read it */
1705 while (!try_lock_hpte(hptp, HPTE_V_HVLOCK))
1707 v = be64_to_cpu(hptp[0]);
1708 hr = be64_to_cpu(hptp[1]);
1709 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
1710 v = hpte_new_to_old_v(v, hr);
1711 hr = hpte_new_to_old_r(hr);
1714 /* re-evaluate valid and dirty from synchronized HPTE value */
1715 valid = !!(v & HPTE_V_VALID);
1716 dirty = !!(revp->guest_rpte & HPTE_GR_MODIFIED);
1718 /* Harvest R and C into guest view if necessary */
1719 rcbits_unset = ~revp->guest_rpte & (HPTE_R_R | HPTE_R_C);
1720 if (valid && (rcbits_unset & hr)) {
1721 revp->guest_rpte |= (hr &
1722 (HPTE_R_R | HPTE_R_C)) | HPTE_GR_MODIFIED;
1726 if (v & HPTE_V_ABSENT) {
1727 v &= ~HPTE_V_ABSENT;
1731 if ((flags & KVM_GET_HTAB_BOLTED_ONLY) && !(v & HPTE_V_BOLTED))
1734 r = revp->guest_rpte;
1735 /* only clear modified if this is the right sort of entry */
1736 if (valid == want_valid && dirty) {
1737 r &= ~HPTE_GR_MODIFIED;
1738 revp->guest_rpte = r;
1740 unlock_hpte(hptp, be64_to_cpu(hptp[0]));
1742 if (!(valid == want_valid && (first_pass || dirty)))
1745 hpte[0] = cpu_to_be64(v);
1746 hpte[1] = cpu_to_be64(r);
1750 static ssize_t kvm_htab_read(struct file *file, char __user *buf,
1751 size_t count, loff_t *ppos)
1753 struct kvm_htab_ctx *ctx = file->private_data;
1754 struct kvm *kvm = ctx->kvm;
1755 struct kvm_get_htab_header hdr;
1757 struct revmap_entry *revp;
1758 unsigned long i, nb, nw;
1759 unsigned long __user *lbuf;
1760 struct kvm_get_htab_header __user *hptr;
1761 unsigned long flags;
1763 unsigned long hpte[2];
1765 if (!access_ok(buf, count))
1767 if (kvm_is_radix(kvm))
1770 first_pass = ctx->first_pass;
1774 hptp = (__be64 *)(kvm->arch.hpt.virt + (i * HPTE_SIZE));
1775 revp = kvm->arch.hpt.rev + i;
1776 lbuf = (unsigned long __user *)buf;
1779 while (nb + sizeof(hdr) + HPTE_SIZE < count) {
1780 /* Initialize header */
1781 hptr = (struct kvm_get_htab_header __user *)buf;
1786 lbuf = (unsigned long __user *)(buf + sizeof(hdr));
1788 /* Skip uninteresting entries, i.e. clean on not-first pass */
1790 while (i < kvmppc_hpt_npte(&kvm->arch.hpt) &&
1791 !hpte_dirty(revp, hptp)) {
1799 /* Grab a series of valid entries */
1800 while (i < kvmppc_hpt_npte(&kvm->arch.hpt) &&
1801 hdr.n_valid < 0xffff &&
1802 nb + HPTE_SIZE < count &&
1803 record_hpte(flags, hptp, hpte, revp, 1, first_pass)) {
1804 /* valid entry, write it out */
1806 if (__put_user(hpte[0], lbuf) ||
1807 __put_user(hpte[1], lbuf + 1))
1815 /* Now skip invalid entries while we can */
1816 while (i < kvmppc_hpt_npte(&kvm->arch.hpt) &&
1817 hdr.n_invalid < 0xffff &&
1818 record_hpte(flags, hptp, hpte, revp, 0, first_pass)) {
1819 /* found an invalid entry */
1826 if (hdr.n_valid || hdr.n_invalid) {
1827 /* write back the header */
1828 if (__copy_to_user(hptr, &hdr, sizeof(hdr)))
1831 buf = (char __user *)lbuf;
1836 /* Check if we've wrapped around the hash table */
1837 if (i >= kvmppc_hpt_npte(&kvm->arch.hpt)) {
1839 ctx->first_pass = 0;
1849 static ssize_t kvm_htab_write(struct file *file, const char __user *buf,
1850 size_t count, loff_t *ppos)
1852 struct kvm_htab_ctx *ctx = file->private_data;
1853 struct kvm *kvm = ctx->kvm;
1854 struct kvm_get_htab_header hdr;
1857 unsigned long __user *lbuf;
1859 unsigned long tmp[2];
1865 if (!access_ok(buf, count))
1867 if (kvm_is_radix(kvm))
1870 /* lock out vcpus from running while we're doing this */
1871 mutex_lock(&kvm->lock);
1872 mmu_ready = kvm->arch.mmu_ready;
1874 kvm->arch.mmu_ready = 0; /* temporarily */
1875 /* order mmu_ready vs. vcpus_running */
1877 if (atomic_read(&kvm->arch.vcpus_running)) {
1878 kvm->arch.mmu_ready = 1;
1879 mutex_unlock(&kvm->lock);
1885 for (nb = 0; nb + sizeof(hdr) <= count; ) {
1887 if (__copy_from_user(&hdr, buf, sizeof(hdr)))
1891 if (nb + hdr.n_valid * HPTE_SIZE > count)
1899 if (i >= kvmppc_hpt_npte(&kvm->arch.hpt) ||
1900 i + hdr.n_valid + hdr.n_invalid > kvmppc_hpt_npte(&kvm->arch.hpt))
1903 hptp = (__be64 *)(kvm->arch.hpt.virt + (i * HPTE_SIZE));
1904 lbuf = (unsigned long __user *)buf;
1905 for (j = 0; j < hdr.n_valid; ++j) {
1910 if (__get_user(hpte_v, lbuf) ||
1911 __get_user(hpte_r, lbuf + 1))
1913 v = be64_to_cpu(hpte_v);
1914 r = be64_to_cpu(hpte_r);
1916 if (!(v & HPTE_V_VALID))
1918 pshift = kvmppc_hpte_base_page_shift(v, r);
1924 if (be64_to_cpu(hptp[0]) & (HPTE_V_VALID | HPTE_V_ABSENT))
1925 kvmppc_do_h_remove(kvm, 0, i, 0, tmp);
1927 ret = kvmppc_virtmode_do_h_enter(kvm, H_EXACT, i, v, r,
1929 if (ret != H_SUCCESS) {
1930 pr_err("kvm_htab_write ret %ld i=%ld v=%lx "
1931 "r=%lx\n", ret, i, v, r);
1934 if (!mmu_ready && is_vrma_hpte(v)) {
1935 unsigned long senc, lpcr;
1937 senc = slb_pgsize_encoding(1ul << pshift);
1938 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
1939 (VRMA_VSID << SLB_VSID_SHIFT_1T);
1940 if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
1941 lpcr = senc << (LPCR_VRMASD_SH - 4);
1942 kvmppc_update_lpcr(kvm, lpcr,
1945 kvmppc_setup_partition_table(kvm);
1953 for (j = 0; j < hdr.n_invalid; ++j) {
1954 if (be64_to_cpu(hptp[0]) & (HPTE_V_VALID | HPTE_V_ABSENT))
1955 kvmppc_do_h_remove(kvm, 0, i, 0, tmp);
1963 /* Order HPTE updates vs. mmu_ready */
1965 kvm->arch.mmu_ready = mmu_ready;
1966 mutex_unlock(&kvm->lock);
1973 static int kvm_htab_release(struct inode *inode, struct file *filp)
1975 struct kvm_htab_ctx *ctx = filp->private_data;
1977 filp->private_data = NULL;
1978 if (!(ctx->flags & KVM_GET_HTAB_WRITE))
1979 atomic_dec(&ctx->kvm->arch.hpte_mod_interest);
1980 kvm_put_kvm(ctx->kvm);
1985 static const struct file_operations kvm_htab_fops = {
1986 .read = kvm_htab_read,
1987 .write = kvm_htab_write,
1988 .llseek = default_llseek,
1989 .release = kvm_htab_release,
1992 int kvm_vm_ioctl_get_htab_fd(struct kvm *kvm, struct kvm_get_htab_fd *ghf)
1995 struct kvm_htab_ctx *ctx;
1998 /* reject flags we don't recognize */
1999 if (ghf->flags & ~(KVM_GET_HTAB_BOLTED_ONLY | KVM_GET_HTAB_WRITE))
2001 ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
2006 ctx->index = ghf->start_index;
2007 ctx->flags = ghf->flags;
2008 ctx->first_pass = 1;
2010 rwflag = (ghf->flags & KVM_GET_HTAB_WRITE) ? O_WRONLY : O_RDONLY;
2011 ret = anon_inode_getfd("kvm-htab", &kvm_htab_fops, ctx, rwflag | O_CLOEXEC);
2018 if (rwflag == O_RDONLY) {
2019 mutex_lock(&kvm->slots_lock);
2020 atomic_inc(&kvm->arch.hpte_mod_interest);
2021 /* make sure kvmppc_do_h_enter etc. see the increment */
2022 synchronize_srcu_expedited(&kvm->srcu);
2023 mutex_unlock(&kvm->slots_lock);
2029 struct debugfs_htab_state {
2032 unsigned long hpt_index;
2038 static int debugfs_htab_open(struct inode *inode, struct file *file)
2040 struct kvm *kvm = inode->i_private;
2041 struct debugfs_htab_state *p;
2043 p = kzalloc(sizeof(*p), GFP_KERNEL);
2049 mutex_init(&p->mutex);
2050 file->private_data = p;
2052 return nonseekable_open(inode, file);
2055 static int debugfs_htab_release(struct inode *inode, struct file *file)
2057 struct debugfs_htab_state *p = file->private_data;
2059 kvm_put_kvm(p->kvm);
2064 static ssize_t debugfs_htab_read(struct file *file, char __user *buf,
2065 size_t len, loff_t *ppos)
2067 struct debugfs_htab_state *p = file->private_data;
2070 unsigned long v, hr, gr;
2075 if (kvm_is_radix(kvm))
2078 ret = mutex_lock_interruptible(&p->mutex);
2082 if (p->chars_left) {
2086 r = copy_to_user(buf, p->buf + p->buf_index, n);
2101 hptp = (__be64 *)(kvm->arch.hpt.virt + (i * HPTE_SIZE));
2102 for (; len != 0 && i < kvmppc_hpt_npte(&kvm->arch.hpt);
2104 if (!(be64_to_cpu(hptp[0]) & (HPTE_V_VALID | HPTE_V_ABSENT)))
2107 /* lock the HPTE so it's stable and read it */
2109 while (!try_lock_hpte(hptp, HPTE_V_HVLOCK))
2111 v = be64_to_cpu(hptp[0]) & ~HPTE_V_HVLOCK;
2112 hr = be64_to_cpu(hptp[1]);
2113 gr = kvm->arch.hpt.rev[i].guest_rpte;
2114 unlock_hpte(hptp, v);
2117 if (!(v & (HPTE_V_VALID | HPTE_V_ABSENT)))
2120 n = scnprintf(p->buf, sizeof(p->buf),
2121 "%6lx %.16lx %.16lx %.16lx\n",
2126 r = copy_to_user(buf, p->buf, n);
2142 mutex_unlock(&p->mutex);
2146 static ssize_t debugfs_htab_write(struct file *file, const char __user *buf,
2147 size_t len, loff_t *ppos)
2152 static const struct file_operations debugfs_htab_fops = {
2153 .owner = THIS_MODULE,
2154 .open = debugfs_htab_open,
2155 .release = debugfs_htab_release,
2156 .read = debugfs_htab_read,
2157 .write = debugfs_htab_write,
2158 .llseek = generic_file_llseek,
2161 void kvmppc_mmu_debugfs_init(struct kvm *kvm)
2163 kvm->arch.htab_dentry = debugfs_create_file("htab", 0400,
2164 kvm->arch.debugfs_dir, kvm,
2165 &debugfs_htab_fops);
2168 void kvmppc_mmu_book3s_hv_init(struct kvm_vcpu *vcpu)
2170 struct kvmppc_mmu *mmu = &vcpu->arch.mmu;
2172 vcpu->arch.slb_nr = 32; /* POWER7/POWER8 */
2174 mmu->xlate = kvmppc_mmu_book3s_64_hv_xlate;
2175 mmu->reset_msr = kvmppc_mmu_book3s_64_hv_reset_msr;
2177 vcpu->arch.hflags |= BOOK3S_HFLAG_SLB;