2 * This program is free software; you can redistribute it and/or modify
3 * it under the terms of the GNU General Public License, version 2, as
4 * published by the Free Software Foundation.
6 * This program is distributed in the hope that it will be useful,
7 * but WITHOUT ANY WARRANTY; without even the implied warranty of
8 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
9 * GNU General Public License for more details.
11 * You should have received a copy of the GNU General Public License
12 * along with this program; if not, write to the Free Software
13 * Foundation, 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
15 * Copyright 2010 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
18 #include <linux/types.h>
19 #include <linux/string.h>
20 #include <linux/kvm.h>
21 #include <linux/kvm_host.h>
22 #include <linux/highmem.h>
23 #include <linux/gfp.h>
24 #include <linux/slab.h>
25 #include <linux/hugetlb.h>
26 #include <linux/vmalloc.h>
27 #include <linux/srcu.h>
28 #include <linux/anon_inodes.h>
29 #include <linux/file.h>
30 #include <linux/debugfs.h>
32 #include <asm/tlbflush.h>
33 #include <asm/kvm_ppc.h>
34 #include <asm/kvm_book3s.h>
35 #include <asm/book3s/64/mmu-hash.h>
36 #include <asm/hvcall.h>
37 #include <asm/synch.h>
38 #include <asm/ppc-opcode.h>
39 #include <asm/cputable.h>
40 #include <asm/pte-walk.h>
44 //#define DEBUG_RESIZE_HPT 1
46 #ifdef DEBUG_RESIZE_HPT
47 #define resize_hpt_debug(resize, ...) \
49 printk(KERN_DEBUG "RESIZE HPT %p: ", resize); \
50 printk(__VA_ARGS__); \
53 #define resize_hpt_debug(resize, ...) \
57 static long kvmppc_virtmode_do_h_enter(struct kvm *kvm, unsigned long flags,
58 long pte_index, unsigned long pteh,
59 unsigned long ptel, unsigned long *pte_idx_ret);
61 struct kvm_resize_hpt {
62 /* These fields read-only after init */
64 struct work_struct work;
67 /* These fields protected by kvm->lock */
71 /* Private to the work thread, until prepare_done is true,
72 * then protected by kvm->resize_hpt_sem */
73 struct kvm_hpt_info hpt;
76 int kvmppc_allocate_hpt(struct kvm_hpt_info *info, u32 order)
78 unsigned long hpt = 0;
80 struct page *page = NULL;
81 struct revmap_entry *rev;
84 if ((order < PPC_MIN_HPT_ORDER) || (order > PPC_MAX_HPT_ORDER))
87 page = kvm_alloc_hpt_cma(1ul << (order - PAGE_SHIFT));
89 hpt = (unsigned long)pfn_to_kaddr(page_to_pfn(page));
90 memset((void *)hpt, 0, (1ul << order));
95 hpt = __get_free_pages(GFP_KERNEL|__GFP_ZERO|__GFP_RETRY_MAYFAIL
96 |__GFP_NOWARN, order - PAGE_SHIFT);
101 /* HPTEs are 2**4 bytes long */
102 npte = 1ul << (order - 4);
104 /* Allocate reverse map array */
105 rev = vmalloc(sizeof(struct revmap_entry) * npte);
108 kvm_free_hpt_cma(page, 1 << (order - PAGE_SHIFT));
110 free_pages(hpt, order - PAGE_SHIFT);
122 void kvmppc_set_hpt(struct kvm *kvm, struct kvm_hpt_info *info)
124 atomic64_set(&kvm->arch.mmio_update, 0);
125 kvm->arch.hpt = *info;
126 kvm->arch.sdr1 = __pa(info->virt) | (info->order - 18);
128 pr_debug("KVM guest htab at %lx (order %ld), LPID %x\n",
129 info->virt, (long)info->order, kvm->arch.lpid);
132 long kvmppc_alloc_reset_hpt(struct kvm *kvm, int order)
135 struct kvm_hpt_info info;
137 mutex_lock(&kvm->lock);
138 if (kvm->arch.mmu_ready) {
139 kvm->arch.mmu_ready = 0;
140 /* order mmu_ready vs. vcpus_running */
142 if (atomic_read(&kvm->arch.vcpus_running)) {
143 kvm->arch.mmu_ready = 1;
147 if (kvm_is_radix(kvm)) {
148 err = kvmppc_switch_mmu_to_hpt(kvm);
153 if (kvm->arch.hpt.order == order) {
154 /* We already have a suitable HPT */
156 /* Set the entire HPT to 0, i.e. invalid HPTEs */
157 memset((void *)kvm->arch.hpt.virt, 0, 1ul << order);
159 * Reset all the reverse-mapping chains for all memslots
161 kvmppc_rmap_reset(kvm);
162 /* Ensure that each vcpu will flush its TLB on next entry. */
163 cpumask_setall(&kvm->arch.need_tlb_flush);
168 if (kvm->arch.hpt.virt) {
169 kvmppc_free_hpt(&kvm->arch.hpt);
170 kvmppc_rmap_reset(kvm);
173 err = kvmppc_allocate_hpt(&info, order);
176 kvmppc_set_hpt(kvm, &info);
179 mutex_unlock(&kvm->lock);
183 void kvmppc_free_hpt(struct kvm_hpt_info *info)
188 kvm_free_hpt_cma(virt_to_page(info->virt),
189 1 << (info->order - PAGE_SHIFT));
191 free_pages(info->virt, info->order - PAGE_SHIFT);
196 /* Bits in first HPTE dword for pagesize 4k, 64k or 16M */
197 static inline unsigned long hpte0_pgsize_encoding(unsigned long pgsize)
199 return (pgsize > 0x1000) ? HPTE_V_LARGE : 0;
202 /* Bits in second HPTE dword for pagesize 4k, 64k or 16M */
203 static inline unsigned long hpte1_pgsize_encoding(unsigned long pgsize)
205 return (pgsize == 0x10000) ? 0x1000 : 0;
208 void kvmppc_map_vrma(struct kvm_vcpu *vcpu, struct kvm_memory_slot *memslot,
209 unsigned long porder)
212 unsigned long npages;
213 unsigned long hp_v, hp_r;
214 unsigned long addr, hash;
216 unsigned long hp0, hp1;
217 unsigned long idx_ret;
219 struct kvm *kvm = vcpu->kvm;
221 psize = 1ul << porder;
222 npages = memslot->npages >> (porder - PAGE_SHIFT);
224 /* VRMA can't be > 1TB */
225 if (npages > 1ul << (40 - porder))
226 npages = 1ul << (40 - porder);
227 /* Can't use more than 1 HPTE per HPTEG */
228 if (npages > kvmppc_hpt_mask(&kvm->arch.hpt) + 1)
229 npages = kvmppc_hpt_mask(&kvm->arch.hpt) + 1;
231 hp0 = HPTE_V_1TB_SEG | (VRMA_VSID << (40 - 16)) |
232 HPTE_V_BOLTED | hpte0_pgsize_encoding(psize);
233 hp1 = hpte1_pgsize_encoding(psize) |
234 HPTE_R_R | HPTE_R_C | HPTE_R_M | PP_RWXX;
236 for (i = 0; i < npages; ++i) {
238 /* can't use hpt_hash since va > 64 bits */
239 hash = (i ^ (VRMA_VSID ^ (VRMA_VSID << 25)))
240 & kvmppc_hpt_mask(&kvm->arch.hpt);
242 * We assume that the hash table is empty and no
243 * vcpus are using it at this stage. Since we create
244 * at most one HPTE per HPTEG, we just assume entry 7
245 * is available and use it.
247 hash = (hash << 3) + 7;
248 hp_v = hp0 | ((addr >> 16) & ~0x7fUL);
250 ret = kvmppc_virtmode_do_h_enter(kvm, H_EXACT, hash, hp_v, hp_r,
252 if (ret != H_SUCCESS) {
253 pr_err("KVM: map_vrma at %lx failed, ret=%ld\n",
260 int kvmppc_mmu_hv_init(void)
262 unsigned long host_lpid, rsvd_lpid;
264 if (!cpu_has_feature(CPU_FTR_HVMODE))
267 /* POWER7 has 10-bit LPIDs (12-bit in POWER8) */
268 host_lpid = mfspr(SPRN_LPID);
269 rsvd_lpid = LPID_RSVD;
271 kvmppc_init_lpid(rsvd_lpid + 1);
273 kvmppc_claim_lpid(host_lpid);
274 /* rsvd_lpid is reserved for use in partition switching */
275 kvmppc_claim_lpid(rsvd_lpid);
280 static void kvmppc_mmu_book3s_64_hv_reset_msr(struct kvm_vcpu *vcpu)
282 unsigned long msr = vcpu->arch.intr_msr;
284 /* If transactional, change to suspend mode on IRQ delivery */
285 if (MSR_TM_TRANSACTIONAL(vcpu->arch.shregs.msr))
288 msr |= vcpu->arch.shregs.msr & MSR_TS_MASK;
289 kvmppc_set_msr(vcpu, msr);
292 static long kvmppc_virtmode_do_h_enter(struct kvm *kvm, unsigned long flags,
293 long pte_index, unsigned long pteh,
294 unsigned long ptel, unsigned long *pte_idx_ret)
298 /* Protect linux PTE lookup from page table destruction */
299 rcu_read_lock_sched(); /* this disables preemption too */
300 ret = kvmppc_do_h_enter(kvm, flags, pte_index, pteh, ptel,
301 current->mm->pgd, false, pte_idx_ret);
302 rcu_read_unlock_sched();
303 if (ret == H_TOO_HARD) {
304 /* this can't happen */
305 pr_err("KVM: Oops, kvmppc_h_enter returned too hard!\n");
306 ret = H_RESOURCE; /* or something */
312 static struct kvmppc_slb *kvmppc_mmu_book3s_hv_find_slbe(struct kvm_vcpu *vcpu,
318 for (i = 0; i < vcpu->arch.slb_nr; i++) {
319 if (!(vcpu->arch.slb[i].orige & SLB_ESID_V))
322 if (vcpu->arch.slb[i].origv & SLB_VSID_B_1T)
327 if (((vcpu->arch.slb[i].orige ^ eaddr) & mask) == 0)
328 return &vcpu->arch.slb[i];
333 static unsigned long kvmppc_mmu_get_real_addr(unsigned long v, unsigned long r,
336 unsigned long ra_mask;
338 ra_mask = kvmppc_actual_pgsz(v, r) - 1;
339 return (r & HPTE_R_RPN & ~ra_mask) | (ea & ra_mask);
342 static int kvmppc_mmu_book3s_64_hv_xlate(struct kvm_vcpu *vcpu, gva_t eaddr,
343 struct kvmppc_pte *gpte, bool data, bool iswrite)
345 struct kvm *kvm = vcpu->kvm;
346 struct kvmppc_slb *slbe;
348 unsigned long pp, key;
349 unsigned long v, orig_v, gr;
352 int virtmode = vcpu->arch.shregs.msr & (data ? MSR_DR : MSR_IR);
354 if (kvm_is_radix(vcpu->kvm))
355 return kvmppc_mmu_radix_xlate(vcpu, eaddr, gpte, data, iswrite);
359 slbe = kvmppc_mmu_book3s_hv_find_slbe(vcpu, eaddr);
364 /* real mode access */
365 slb_v = vcpu->kvm->arch.vrma_slb_v;
369 /* Find the HPTE in the hash table */
370 index = kvmppc_hv_find_lock_hpte(kvm, eaddr, slb_v,
371 HPTE_V_VALID | HPTE_V_ABSENT);
376 hptep = (__be64 *)(kvm->arch.hpt.virt + (index << 4));
377 v = orig_v = be64_to_cpu(hptep[0]) & ~HPTE_V_HVLOCK;
378 if (cpu_has_feature(CPU_FTR_ARCH_300))
379 v = hpte_new_to_old_v(v, be64_to_cpu(hptep[1]));
380 gr = kvm->arch.hpt.rev[index].guest_rpte;
382 unlock_hpte(hptep, orig_v);
386 gpte->vpage = ((v & HPTE_V_AVPN) << 4) | ((eaddr >> 12) & 0xfff);
388 /* Get PP bits and key for permission check */
389 pp = gr & (HPTE_R_PP0 | HPTE_R_PP);
390 key = (vcpu->arch.shregs.msr & MSR_PR) ? SLB_VSID_KP : SLB_VSID_KS;
393 /* Calculate permissions */
394 gpte->may_read = hpte_read_permission(pp, key);
395 gpte->may_write = hpte_write_permission(pp, key);
396 gpte->may_execute = gpte->may_read && !(gr & (HPTE_R_N | HPTE_R_G));
398 /* Storage key permission check for POWER7 */
399 if (data && virtmode) {
400 int amrfield = hpte_get_skey_perm(gr, vcpu->arch.amr);
407 /* Get the guest physical address */
408 gpte->raddr = kvmppc_mmu_get_real_addr(v, gr, eaddr);
413 * Quick test for whether an instruction is a load or a store.
414 * If the instruction is a load or a store, then this will indicate
415 * which it is, at least on server processors. (Embedded processors
416 * have some external PID instructions that don't follow the rule
417 * embodied here.) If the instruction isn't a load or store, then
418 * this doesn't return anything useful.
420 static int instruction_is_store(unsigned int instr)
425 if ((instr & 0xfc000000) == 0x7c000000)
426 mask = 0x100; /* major opcode 31 */
427 return (instr & mask) != 0;
430 int kvmppc_hv_emulate_mmio(struct kvm_run *run, struct kvm_vcpu *vcpu,
431 unsigned long gpa, gva_t ea, int is_store)
436 * If we fail, we just return to the guest and try executing it again.
438 if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
443 * WARNING: We do not know for sure whether the instruction we just
444 * read from memory is the same that caused the fault in the first
445 * place. If the instruction we read is neither an load or a store,
446 * then it can't access memory, so we don't need to worry about
447 * enforcing access permissions. So, assuming it is a load or
448 * store, we just check that its direction (load or store) is
449 * consistent with the original fault, since that's what we
450 * checked the access permissions against. If there is a mismatch
451 * we just return and retry the instruction.
454 if (instruction_is_store(last_inst) != !!is_store)
458 * Emulated accesses are emulated by looking at the hash for
459 * translation once, then performing the access later. The
460 * translation could be invalidated in the meantime in which
461 * point performing the subsequent memory access on the old
462 * physical address could possibly be a security hole for the
463 * guest (but not the host).
465 * This is less of an issue for MMIO stores since they aren't
466 * globally visible. It could be an issue for MMIO loads to
467 * a certain extent but we'll ignore it for now.
470 vcpu->arch.paddr_accessed = gpa;
471 vcpu->arch.vaddr_accessed = ea;
472 return kvmppc_emulate_mmio(run, vcpu);
475 int kvmppc_book3s_hv_page_fault(struct kvm_run *run, struct kvm_vcpu *vcpu,
476 unsigned long ea, unsigned long dsisr)
478 struct kvm *kvm = vcpu->kvm;
479 unsigned long hpte[3], r;
480 unsigned long hnow_v, hnow_r;
482 unsigned long mmu_seq, psize, pte_size;
483 unsigned long gpa_base, gfn_base;
484 unsigned long gpa, gfn, hva, pfn;
485 struct kvm_memory_slot *memslot;
487 struct revmap_entry *rev;
488 struct page *page, *pages[1];
489 long index, ret, npages;
491 unsigned int writing, write_ok;
492 struct vm_area_struct *vma;
493 unsigned long rcbits;
496 if (kvm_is_radix(kvm))
497 return kvmppc_book3s_radix_page_fault(run, vcpu, ea, dsisr);
500 * Real-mode code has already searched the HPT and found the
501 * entry we're interested in. Lock the entry and check that
502 * it hasn't changed. If it has, just return and re-execute the
505 if (ea != vcpu->arch.pgfault_addr)
508 if (vcpu->arch.pgfault_cache) {
509 mmio_update = atomic64_read(&kvm->arch.mmio_update);
510 if (mmio_update == vcpu->arch.pgfault_cache->mmio_update) {
511 r = vcpu->arch.pgfault_cache->rpte;
512 psize = kvmppc_actual_pgsz(vcpu->arch.pgfault_hpte[0],
514 gpa_base = r & HPTE_R_RPN & ~(psize - 1);
515 gfn_base = gpa_base >> PAGE_SHIFT;
516 gpa = gpa_base | (ea & (psize - 1));
517 return kvmppc_hv_emulate_mmio(run, vcpu, gpa, ea,
518 dsisr & DSISR_ISSTORE);
521 index = vcpu->arch.pgfault_index;
522 hptep = (__be64 *)(kvm->arch.hpt.virt + (index << 4));
523 rev = &kvm->arch.hpt.rev[index];
525 while (!try_lock_hpte(hptep, HPTE_V_HVLOCK))
527 hpte[0] = be64_to_cpu(hptep[0]) & ~HPTE_V_HVLOCK;
528 hpte[1] = be64_to_cpu(hptep[1]);
529 hpte[2] = r = rev->guest_rpte;
530 unlock_hpte(hptep, hpte[0]);
533 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
534 hpte[0] = hpte_new_to_old_v(hpte[0], hpte[1]);
535 hpte[1] = hpte_new_to_old_r(hpte[1]);
537 if (hpte[0] != vcpu->arch.pgfault_hpte[0] ||
538 hpte[1] != vcpu->arch.pgfault_hpte[1])
541 /* Translate the logical address and get the page */
542 psize = kvmppc_actual_pgsz(hpte[0], r);
543 gpa_base = r & HPTE_R_RPN & ~(psize - 1);
544 gfn_base = gpa_base >> PAGE_SHIFT;
545 gpa = gpa_base | (ea & (psize - 1));
546 gfn = gpa >> PAGE_SHIFT;
547 memslot = gfn_to_memslot(kvm, gfn);
549 trace_kvm_page_fault_enter(vcpu, hpte, memslot, ea, dsisr);
551 /* No memslot means it's an emulated MMIO region */
552 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
553 return kvmppc_hv_emulate_mmio(run, vcpu, gpa, ea,
554 dsisr & DSISR_ISSTORE);
557 * This should never happen, because of the slot_is_aligned()
558 * check in kvmppc_do_h_enter().
560 if (gfn_base < memslot->base_gfn)
563 /* used to check for invalidations in progress */
564 mmu_seq = kvm->mmu_notifier_seq;
571 pte_size = PAGE_SIZE;
572 writing = (dsisr & DSISR_ISSTORE) != 0;
573 /* If writing != 0, then the HPTE must allow writing, if we get here */
575 hva = gfn_to_hva_memslot(memslot, gfn);
576 npages = get_user_pages_fast(hva, 1, writing, pages);
578 /* Check if it's an I/O mapping */
579 down_read(¤t->mm->mmap_sem);
580 vma = find_vma(current->mm, hva);
581 if (vma && vma->vm_start <= hva && hva + psize <= vma->vm_end &&
582 (vma->vm_flags & VM_PFNMAP)) {
583 pfn = vma->vm_pgoff +
584 ((hva - vma->vm_start) >> PAGE_SHIFT);
586 is_ci = pte_ci(__pte((pgprot_val(vma->vm_page_prot))));
587 write_ok = vma->vm_flags & VM_WRITE;
589 up_read(¤t->mm->mmap_sem);
594 pfn = page_to_pfn(page);
595 if (PageHuge(page)) {
596 page = compound_head(page);
597 pte_size <<= compound_order(page);
599 /* if the guest wants write access, see if that is OK */
600 if (!writing && hpte_is_writable(r)) {
604 * We need to protect against page table destruction
605 * hugepage split and collapse.
607 local_irq_save(flags);
608 ptep = find_current_mm_pte(current->mm->pgd,
611 pte = kvmppc_read_update_linux_pte(ptep, 1);
612 if (__pte_write(pte))
615 local_irq_restore(flags);
619 if (psize > pte_size)
622 /* Check WIMG vs. the actual page we're accessing */
623 if (!hpte_cache_flags_ok(r, is_ci)) {
627 * Allow guest to map emulated device memory as
628 * uncacheable, but actually make it cacheable.
630 r = (r & ~(HPTE_R_W|HPTE_R_I|HPTE_R_G)) | HPTE_R_M;
634 * Set the HPTE to point to pfn.
635 * Since the pfn is at PAGE_SIZE granularity, make sure we
636 * don't mask out lower-order bits if psize < PAGE_SIZE.
638 if (psize < PAGE_SIZE)
640 r = (r & HPTE_R_KEY_HI) | (r & ~(HPTE_R_PP0 - psize)) |
641 ((pfn << PAGE_SHIFT) & ~(psize - 1));
642 if (hpte_is_writable(r) && !write_ok)
643 r = hpte_make_readonly(r);
646 while (!try_lock_hpte(hptep, HPTE_V_HVLOCK))
648 hnow_v = be64_to_cpu(hptep[0]);
649 hnow_r = be64_to_cpu(hptep[1]);
650 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
651 hnow_v = hpte_new_to_old_v(hnow_v, hnow_r);
652 hnow_r = hpte_new_to_old_r(hnow_r);
656 * If the HPT is being resized, don't update the HPTE,
657 * instead let the guest retry after the resize operation is complete.
658 * The synchronization for mmu_ready test vs. set is provided
661 if (!kvm->arch.mmu_ready)
664 if ((hnow_v & ~HPTE_V_HVLOCK) != hpte[0] || hnow_r != hpte[1] ||
665 rev->guest_rpte != hpte[2])
666 /* HPTE has been changed under us; let the guest retry */
668 hpte[0] = (hpte[0] & ~HPTE_V_ABSENT) | HPTE_V_VALID;
670 /* Always put the HPTE in the rmap chain for the page base address */
671 rmap = &memslot->arch.rmap[gfn_base - memslot->base_gfn];
674 /* Check if we might have been invalidated; let the guest retry if so */
676 if (mmu_notifier_retry(vcpu->kvm, mmu_seq)) {
681 /* Only set R/C in real HPTE if set in both *rmap and guest_rpte */
682 rcbits = *rmap >> KVMPPC_RMAP_RC_SHIFT;
683 r &= rcbits | ~(HPTE_R_R | HPTE_R_C);
685 if (be64_to_cpu(hptep[0]) & HPTE_V_VALID) {
686 /* HPTE was previously valid, so we need to invalidate it */
688 hptep[0] |= cpu_to_be64(HPTE_V_ABSENT);
689 kvmppc_invalidate_hpte(kvm, hptep, index);
690 /* don't lose previous R and C bits */
691 r |= be64_to_cpu(hptep[1]) & (HPTE_R_R | HPTE_R_C);
693 kvmppc_add_revmap_chain(kvm, rev, rmap, index, 0);
696 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
697 r = hpte_old_to_new_r(hpte[0], r);
698 hpte[0] = hpte_old_to_new_v(hpte[0]);
700 hptep[1] = cpu_to_be64(r);
702 __unlock_hpte(hptep, hpte[0]);
703 asm volatile("ptesync" : : : "memory");
705 if (page && hpte_is_writable(r))
709 trace_kvm_page_fault_exit(vcpu, hpte, ret);
713 * We drop pages[0] here, not page because page might
714 * have been set to the head page of a compound, but
715 * we have to drop the reference on the correct tail
716 * page to match the get inside gup()
723 __unlock_hpte(hptep, be64_to_cpu(hptep[0]));
728 void kvmppc_rmap_reset(struct kvm *kvm)
730 struct kvm_memslots *slots;
731 struct kvm_memory_slot *memslot;
734 srcu_idx = srcu_read_lock(&kvm->srcu);
735 slots = kvm_memslots(kvm);
736 kvm_for_each_memslot(memslot, slots) {
738 * This assumes it is acceptable to lose reference and
739 * change bits across a reset.
741 memset(memslot->arch.rmap, 0,
742 memslot->npages * sizeof(*memslot->arch.rmap));
744 srcu_read_unlock(&kvm->srcu, srcu_idx);
747 typedef int (*hva_handler_fn)(struct kvm *kvm, struct kvm_memory_slot *memslot,
750 static int kvm_handle_hva_range(struct kvm *kvm,
753 hva_handler_fn handler)
757 struct kvm_memslots *slots;
758 struct kvm_memory_slot *memslot;
760 slots = kvm_memslots(kvm);
761 kvm_for_each_memslot(memslot, slots) {
762 unsigned long hva_start, hva_end;
765 hva_start = max(start, memslot->userspace_addr);
766 hva_end = min(end, memslot->userspace_addr +
767 (memslot->npages << PAGE_SHIFT));
768 if (hva_start >= hva_end)
771 * {gfn(page) | page intersects with [hva_start, hva_end)} =
772 * {gfn, gfn+1, ..., gfn_end-1}.
774 gfn = hva_to_gfn_memslot(hva_start, memslot);
775 gfn_end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, memslot);
777 for (; gfn < gfn_end; ++gfn) {
778 ret = handler(kvm, memslot, gfn);
786 static int kvm_handle_hva(struct kvm *kvm, unsigned long hva,
787 hva_handler_fn handler)
789 return kvm_handle_hva_range(kvm, hva, hva + 1, handler);
792 /* Must be called with both HPTE and rmap locked */
793 static void kvmppc_unmap_hpte(struct kvm *kvm, unsigned long i,
794 struct kvm_memory_slot *memslot,
795 unsigned long *rmapp, unsigned long gfn)
797 __be64 *hptep = (__be64 *) (kvm->arch.hpt.virt + (i << 4));
798 struct revmap_entry *rev = kvm->arch.hpt.rev;
800 unsigned long ptel, psize, rcbits;
804 /* chain is now empty */
805 *rmapp &= ~(KVMPPC_RMAP_PRESENT | KVMPPC_RMAP_INDEX);
807 /* remove i from chain */
811 rev[i].forw = rev[i].back = i;
812 *rmapp = (*rmapp & ~KVMPPC_RMAP_INDEX) | j;
815 /* Now check and modify the HPTE */
816 ptel = rev[i].guest_rpte;
817 psize = kvmppc_actual_pgsz(be64_to_cpu(hptep[0]), ptel);
818 if ((be64_to_cpu(hptep[0]) & HPTE_V_VALID) &&
819 hpte_rpn(ptel, psize) == gfn) {
820 hptep[0] |= cpu_to_be64(HPTE_V_ABSENT);
821 kvmppc_invalidate_hpte(kvm, hptep, i);
822 hptep[1] &= ~cpu_to_be64(HPTE_R_KEY_HI | HPTE_R_KEY_LO);
823 /* Harvest R and C */
824 rcbits = be64_to_cpu(hptep[1]) & (HPTE_R_R | HPTE_R_C);
825 *rmapp |= rcbits << KVMPPC_RMAP_RC_SHIFT;
826 if ((rcbits & HPTE_R_C) && memslot->dirty_bitmap)
827 kvmppc_update_dirty_map(memslot, gfn, psize);
828 if (rcbits & ~rev[i].guest_rpte) {
829 rev[i].guest_rpte = ptel | rcbits;
830 note_hpte_modification(kvm, &rev[i]);
835 static int kvm_unmap_rmapp(struct kvm *kvm, struct kvm_memory_slot *memslot,
840 unsigned long *rmapp;
842 rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn];
845 if (!(*rmapp & KVMPPC_RMAP_PRESENT)) {
851 * To avoid an ABBA deadlock with the HPTE lock bit,
852 * we can't spin on the HPTE lock while holding the
855 i = *rmapp & KVMPPC_RMAP_INDEX;
856 hptep = (__be64 *) (kvm->arch.hpt.virt + (i << 4));
857 if (!try_lock_hpte(hptep, HPTE_V_HVLOCK)) {
858 /* unlock rmap before spinning on the HPTE lock */
860 while (be64_to_cpu(hptep[0]) & HPTE_V_HVLOCK)
865 kvmppc_unmap_hpte(kvm, i, memslot, rmapp, gfn);
867 __unlock_hpte(hptep, be64_to_cpu(hptep[0]));
872 int kvm_unmap_hva_hv(struct kvm *kvm, unsigned long hva)
874 hva_handler_fn handler;
876 handler = kvm_is_radix(kvm) ? kvm_unmap_radix : kvm_unmap_rmapp;
877 kvm_handle_hva(kvm, hva, handler);
881 int kvm_unmap_hva_range_hv(struct kvm *kvm, unsigned long start, unsigned long end)
883 hva_handler_fn handler;
885 handler = kvm_is_radix(kvm) ? kvm_unmap_radix : kvm_unmap_rmapp;
886 kvm_handle_hva_range(kvm, start, end, handler);
890 void kvmppc_core_flush_memslot_hv(struct kvm *kvm,
891 struct kvm_memory_slot *memslot)
895 unsigned long *rmapp;
897 gfn = memslot->base_gfn;
898 rmapp = memslot->arch.rmap;
899 for (n = memslot->npages; n; --n, ++gfn) {
900 if (kvm_is_radix(kvm)) {
901 kvm_unmap_radix(kvm, memslot, gfn);
905 * Testing the present bit without locking is OK because
906 * the memslot has been marked invalid already, and hence
907 * no new HPTEs referencing this page can be created,
908 * thus the present bit can't go from 0 to 1.
910 if (*rmapp & KVMPPC_RMAP_PRESENT)
911 kvm_unmap_rmapp(kvm, memslot, gfn);
916 static int kvm_age_rmapp(struct kvm *kvm, struct kvm_memory_slot *memslot,
919 struct revmap_entry *rev = kvm->arch.hpt.rev;
920 unsigned long head, i, j;
923 unsigned long *rmapp;
925 rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn];
928 if (*rmapp & KVMPPC_RMAP_REFERENCED) {
929 *rmapp &= ~KVMPPC_RMAP_REFERENCED;
932 if (!(*rmapp & KVMPPC_RMAP_PRESENT)) {
937 i = head = *rmapp & KVMPPC_RMAP_INDEX;
939 hptep = (__be64 *) (kvm->arch.hpt.virt + (i << 4));
942 /* If this HPTE isn't referenced, ignore it */
943 if (!(be64_to_cpu(hptep[1]) & HPTE_R_R))
946 if (!try_lock_hpte(hptep, HPTE_V_HVLOCK)) {
947 /* unlock rmap before spinning on the HPTE lock */
949 while (be64_to_cpu(hptep[0]) & HPTE_V_HVLOCK)
954 /* Now check and modify the HPTE */
955 if ((be64_to_cpu(hptep[0]) & HPTE_V_VALID) &&
956 (be64_to_cpu(hptep[1]) & HPTE_R_R)) {
957 kvmppc_clear_ref_hpte(kvm, hptep, i);
958 if (!(rev[i].guest_rpte & HPTE_R_R)) {
959 rev[i].guest_rpte |= HPTE_R_R;
960 note_hpte_modification(kvm, &rev[i]);
964 __unlock_hpte(hptep, be64_to_cpu(hptep[0]));
965 } while ((i = j) != head);
971 int kvm_age_hva_hv(struct kvm *kvm, unsigned long start, unsigned long end)
973 hva_handler_fn handler;
975 handler = kvm_is_radix(kvm) ? kvm_age_radix : kvm_age_rmapp;
976 return kvm_handle_hva_range(kvm, start, end, handler);
979 static int kvm_test_age_rmapp(struct kvm *kvm, struct kvm_memory_slot *memslot,
982 struct revmap_entry *rev = kvm->arch.hpt.rev;
983 unsigned long head, i, j;
986 unsigned long *rmapp;
988 rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn];
989 if (*rmapp & KVMPPC_RMAP_REFERENCED)
993 if (*rmapp & KVMPPC_RMAP_REFERENCED)
996 if (*rmapp & KVMPPC_RMAP_PRESENT) {
997 i = head = *rmapp & KVMPPC_RMAP_INDEX;
999 hp = (unsigned long *)(kvm->arch.hpt.virt + (i << 4));
1001 if (be64_to_cpu(hp[1]) & HPTE_R_R)
1003 } while ((i = j) != head);
1012 int kvm_test_age_hva_hv(struct kvm *kvm, unsigned long hva)
1014 hva_handler_fn handler;
1016 handler = kvm_is_radix(kvm) ? kvm_test_age_radix : kvm_test_age_rmapp;
1017 return kvm_handle_hva(kvm, hva, handler);
1020 void kvm_set_spte_hva_hv(struct kvm *kvm, unsigned long hva, pte_t pte)
1022 hva_handler_fn handler;
1024 handler = kvm_is_radix(kvm) ? kvm_unmap_radix : kvm_unmap_rmapp;
1025 kvm_handle_hva(kvm, hva, handler);
1028 static int vcpus_running(struct kvm *kvm)
1030 return atomic_read(&kvm->arch.vcpus_running) != 0;
1034 * Returns the number of system pages that are dirty.
1035 * This can be more than 1 if we find a huge-page HPTE.
1037 static int kvm_test_clear_dirty_npages(struct kvm *kvm, unsigned long *rmapp)
1039 struct revmap_entry *rev = kvm->arch.hpt.rev;
1040 unsigned long head, i, j;
1044 int npages_dirty = 0;
1048 if (!(*rmapp & KVMPPC_RMAP_PRESENT)) {
1050 return npages_dirty;
1053 i = head = *rmapp & KVMPPC_RMAP_INDEX;
1055 unsigned long hptep1;
1056 hptep = (__be64 *) (kvm->arch.hpt.virt + (i << 4));
1060 * Checking the C (changed) bit here is racy since there
1061 * is no guarantee about when the hardware writes it back.
1062 * If the HPTE is not writable then it is stable since the
1063 * page can't be written to, and we would have done a tlbie
1064 * (which forces the hardware to complete any writeback)
1065 * when making the HPTE read-only.
1066 * If vcpus are running then this call is racy anyway
1067 * since the page could get dirtied subsequently, so we
1068 * expect there to be a further call which would pick up
1069 * any delayed C bit writeback.
1070 * Otherwise we need to do the tlbie even if C==0 in
1071 * order to pick up any delayed writeback of C.
1073 hptep1 = be64_to_cpu(hptep[1]);
1074 if (!(hptep1 & HPTE_R_C) &&
1075 (!hpte_is_writable(hptep1) || vcpus_running(kvm)))
1078 if (!try_lock_hpte(hptep, HPTE_V_HVLOCK)) {
1079 /* unlock rmap before spinning on the HPTE lock */
1081 while (hptep[0] & cpu_to_be64(HPTE_V_HVLOCK))
1086 /* Now check and modify the HPTE */
1087 if (!(hptep[0] & cpu_to_be64(HPTE_V_VALID))) {
1088 __unlock_hpte(hptep, be64_to_cpu(hptep[0]));
1092 /* need to make it temporarily absent so C is stable */
1093 hptep[0] |= cpu_to_be64(HPTE_V_ABSENT);
1094 kvmppc_invalidate_hpte(kvm, hptep, i);
1095 v = be64_to_cpu(hptep[0]);
1096 r = be64_to_cpu(hptep[1]);
1098 hptep[1] = cpu_to_be64(r & ~HPTE_R_C);
1099 if (!(rev[i].guest_rpte & HPTE_R_C)) {
1100 rev[i].guest_rpte |= HPTE_R_C;
1101 note_hpte_modification(kvm, &rev[i]);
1103 n = kvmppc_actual_pgsz(v, r);
1104 n = (n + PAGE_SIZE - 1) >> PAGE_SHIFT;
1105 if (n > npages_dirty)
1109 v &= ~HPTE_V_ABSENT;
1111 __unlock_hpte(hptep, v);
1112 } while ((i = j) != head);
1115 return npages_dirty;
1118 void kvmppc_harvest_vpa_dirty(struct kvmppc_vpa *vpa,
1119 struct kvm_memory_slot *memslot,
1124 if (!vpa->dirty || !vpa->pinned_addr)
1126 gfn = vpa->gpa >> PAGE_SHIFT;
1127 if (gfn < memslot->base_gfn ||
1128 gfn >= memslot->base_gfn + memslot->npages)
1133 __set_bit_le(gfn - memslot->base_gfn, map);
1136 long kvmppc_hv_get_dirty_log_hpt(struct kvm *kvm,
1137 struct kvm_memory_slot *memslot, unsigned long *map)
1140 unsigned long *rmapp;
1143 rmapp = memslot->arch.rmap;
1144 for (i = 0; i < memslot->npages; ++i) {
1145 int npages = kvm_test_clear_dirty_npages(kvm, rmapp);
1147 * Note that if npages > 0 then i must be a multiple of npages,
1148 * since we always put huge-page HPTEs in the rmap chain
1149 * corresponding to their page base address.
1152 set_dirty_bits(map, i, npages);
1159 void *kvmppc_pin_guest_page(struct kvm *kvm, unsigned long gpa,
1160 unsigned long *nb_ret)
1162 struct kvm_memory_slot *memslot;
1163 unsigned long gfn = gpa >> PAGE_SHIFT;
1164 struct page *page, *pages[1];
1166 unsigned long hva, offset;
1169 srcu_idx = srcu_read_lock(&kvm->srcu);
1170 memslot = gfn_to_memslot(kvm, gfn);
1171 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
1173 hva = gfn_to_hva_memslot(memslot, gfn);
1174 npages = get_user_pages_fast(hva, 1, 1, pages);
1178 srcu_read_unlock(&kvm->srcu, srcu_idx);
1180 offset = gpa & (PAGE_SIZE - 1);
1182 *nb_ret = PAGE_SIZE - offset;
1183 return page_address(page) + offset;
1186 srcu_read_unlock(&kvm->srcu, srcu_idx);
1190 void kvmppc_unpin_guest_page(struct kvm *kvm, void *va, unsigned long gpa,
1193 struct page *page = virt_to_page(va);
1194 struct kvm_memory_slot *memslot;
1203 /* We need to mark this page dirty in the memslot dirty_bitmap, if any */
1204 gfn = gpa >> PAGE_SHIFT;
1205 srcu_idx = srcu_read_lock(&kvm->srcu);
1206 memslot = gfn_to_memslot(kvm, gfn);
1207 if (memslot && memslot->dirty_bitmap)
1208 set_bit_le(gfn - memslot->base_gfn, memslot->dirty_bitmap);
1209 srcu_read_unlock(&kvm->srcu, srcu_idx);
1215 static int resize_hpt_allocate(struct kvm_resize_hpt *resize)
1219 rc = kvmppc_allocate_hpt(&resize->hpt, resize->order);
1223 resize_hpt_debug(resize, "resize_hpt_allocate(): HPT @ 0x%lx\n",
1229 static unsigned long resize_hpt_rehash_hpte(struct kvm_resize_hpt *resize,
1232 struct kvm *kvm = resize->kvm;
1233 struct kvm_hpt_info *old = &kvm->arch.hpt;
1234 struct kvm_hpt_info *new = &resize->hpt;
1235 unsigned long old_hash_mask = (1ULL << (old->order - 7)) - 1;
1236 unsigned long new_hash_mask = (1ULL << (new->order - 7)) - 1;
1237 __be64 *hptep, *new_hptep;
1238 unsigned long vpte, rpte, guest_rpte;
1240 struct revmap_entry *rev;
1241 unsigned long apsize, avpn, pteg, hash;
1242 unsigned long new_idx, new_pteg, replace_vpte;
1245 hptep = (__be64 *)(old->virt + (idx << 4));
1247 /* Guest is stopped, so new HPTEs can't be added or faulted
1248 * in, only unmapped or altered by host actions. So, it's
1249 * safe to check this before we take the HPTE lock */
1250 vpte = be64_to_cpu(hptep[0]);
1251 if (!(vpte & HPTE_V_VALID) && !(vpte & HPTE_V_ABSENT))
1252 return 0; /* nothing to do */
1254 while (!try_lock_hpte(hptep, HPTE_V_HVLOCK))
1257 vpte = be64_to_cpu(hptep[0]);
1260 if (!(vpte & HPTE_V_VALID) && !(vpte & HPTE_V_ABSENT))
1265 rev = &old->rev[idx];
1266 guest_rpte = rev->guest_rpte;
1269 apsize = kvmppc_actual_pgsz(vpte, guest_rpte);
1273 if (vpte & HPTE_V_VALID) {
1274 unsigned long gfn = hpte_rpn(guest_rpte, apsize);
1275 int srcu_idx = srcu_read_lock(&kvm->srcu);
1276 struct kvm_memory_slot *memslot =
1277 __gfn_to_memslot(kvm_memslots(kvm), gfn);
1280 unsigned long *rmapp;
1281 rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn];
1284 kvmppc_unmap_hpte(kvm, idx, memslot, rmapp, gfn);
1288 srcu_read_unlock(&kvm->srcu, srcu_idx);
1291 /* Reload PTE after unmap */
1292 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]);
1302 pshift = kvmppc_hpte_base_page_shift(vpte, rpte);
1303 avpn = HPTE_V_AVPN_VAL(vpte) & ~(((1ul << pshift) - 1) >> 23);
1304 pteg = idx / HPTES_PER_GROUP;
1305 if (vpte & HPTE_V_SECONDARY)
1308 if (!(vpte & HPTE_V_1TB_SEG)) {
1309 unsigned long offset, vsid;
1311 /* We only have 28 - 23 bits of offset in avpn */
1312 offset = (avpn & 0x1f) << 23;
1314 /* We can find more bits from the pteg value */
1316 offset |= ((vsid ^ pteg) & old_hash_mask) << pshift;
1318 hash = vsid ^ (offset >> pshift);
1320 unsigned long offset, vsid;
1322 /* We only have 40 - 23 bits of seg_off in avpn */
1323 offset = (avpn & 0x1ffff) << 23;
1326 offset |= ((vsid ^ (vsid << 25) ^ pteg) & old_hash_mask) << pshift;
1328 hash = vsid ^ (vsid << 25) ^ (offset >> pshift);
1331 new_pteg = hash & new_hash_mask;
1332 if (vpte & HPTE_V_SECONDARY) {
1333 BUG_ON(~pteg != (hash & old_hash_mask));
1334 new_pteg = ~new_pteg;
1336 BUG_ON(pteg != (hash & old_hash_mask));
1339 new_idx = new_pteg * HPTES_PER_GROUP + (idx % HPTES_PER_GROUP);
1340 new_hptep = (__be64 *)(new->virt + (new_idx << 4));
1342 replace_vpte = be64_to_cpu(new_hptep[0]);
1344 if (replace_vpte & (HPTE_V_VALID | HPTE_V_ABSENT)) {
1345 BUG_ON(new->order >= old->order);
1347 if (replace_vpte & HPTE_V_BOLTED) {
1348 if (vpte & HPTE_V_BOLTED)
1349 /* Bolted collision, nothing we can do */
1351 /* Discard the new HPTE */
1355 /* Discard the previous HPTE */
1358 new_hptep[1] = cpu_to_be64(rpte);
1359 new->rev[new_idx].guest_rpte = guest_rpte;
1360 /* No need for a barrier, since new HPT isn't active */
1361 new_hptep[0] = cpu_to_be64(vpte);
1362 unlock_hpte(new_hptep, vpte);
1365 unlock_hpte(hptep, vpte);
1369 static int resize_hpt_rehash(struct kvm_resize_hpt *resize)
1371 struct kvm *kvm = resize->kvm;
1376 * resize_hpt_rehash_hpte() doesn't handle the new-format HPTEs
1377 * that POWER9 uses, and could well hit a BUG_ON on POWER9.
1379 if (cpu_has_feature(CPU_FTR_ARCH_300))
1381 for (i = 0; i < kvmppc_hpt_npte(&kvm->arch.hpt); i++) {
1382 rc = resize_hpt_rehash_hpte(resize, i);
1390 static void resize_hpt_pivot(struct kvm_resize_hpt *resize)
1392 struct kvm *kvm = resize->kvm;
1393 struct kvm_hpt_info hpt_tmp;
1395 /* Exchange the pending tables in the resize structure with
1396 * the active tables */
1398 resize_hpt_debug(resize, "resize_hpt_pivot()\n");
1400 spin_lock(&kvm->mmu_lock);
1401 asm volatile("ptesync" : : : "memory");
1403 hpt_tmp = kvm->arch.hpt;
1404 kvmppc_set_hpt(kvm, &resize->hpt);
1405 resize->hpt = hpt_tmp;
1407 spin_unlock(&kvm->mmu_lock);
1409 synchronize_srcu_expedited(&kvm->srcu);
1411 resize_hpt_debug(resize, "resize_hpt_pivot() done\n");
1414 static void resize_hpt_release(struct kvm *kvm, struct kvm_resize_hpt *resize)
1416 BUG_ON(kvm->arch.resize_hpt != resize);
1421 if (resize->hpt.virt)
1422 kvmppc_free_hpt(&resize->hpt);
1424 kvm->arch.resize_hpt = NULL;
1428 static void resize_hpt_prepare_work(struct work_struct *work)
1430 struct kvm_resize_hpt *resize = container_of(work,
1431 struct kvm_resize_hpt,
1433 struct kvm *kvm = resize->kvm;
1436 resize_hpt_debug(resize, "resize_hpt_prepare_work(): order = %d\n",
1439 err = resize_hpt_allocate(resize);
1441 mutex_lock(&kvm->lock);
1443 resize->error = err;
1444 resize->prepare_done = true;
1446 mutex_unlock(&kvm->lock);
1449 long kvm_vm_ioctl_resize_hpt_prepare(struct kvm *kvm,
1450 struct kvm_ppc_resize_hpt *rhpt)
1452 unsigned long flags = rhpt->flags;
1453 unsigned long shift = rhpt->shift;
1454 struct kvm_resize_hpt *resize;
1457 if (flags != 0 || kvm_is_radix(kvm))
1460 if (shift && ((shift < 18) || (shift > 46)))
1463 mutex_lock(&kvm->lock);
1465 resize = kvm->arch.resize_hpt;
1468 if (resize->order == shift) {
1469 /* Suitable resize in progress */
1470 if (resize->prepare_done) {
1471 ret = resize->error;
1473 resize_hpt_release(kvm, resize);
1475 ret = 100; /* estimated time in ms */
1481 /* not suitable, cancel it */
1482 resize_hpt_release(kvm, resize);
1487 goto out; /* nothing to do */
1489 /* start new resize */
1491 resize = kzalloc(sizeof(*resize), GFP_KERNEL);
1496 resize->order = shift;
1498 INIT_WORK(&resize->work, resize_hpt_prepare_work);
1499 kvm->arch.resize_hpt = resize;
1501 schedule_work(&resize->work);
1503 ret = 100; /* estimated time in ms */
1506 mutex_unlock(&kvm->lock);
1510 static void resize_hpt_boot_vcpu(void *opaque)
1512 /* Nothing to do, just force a KVM exit */
1515 long kvm_vm_ioctl_resize_hpt_commit(struct kvm *kvm,
1516 struct kvm_ppc_resize_hpt *rhpt)
1518 unsigned long flags = rhpt->flags;
1519 unsigned long shift = rhpt->shift;
1520 struct kvm_resize_hpt *resize;
1523 if (flags != 0 || kvm_is_radix(kvm))
1526 if (shift && ((shift < 18) || (shift > 46)))
1529 mutex_lock(&kvm->lock);
1531 resize = kvm->arch.resize_hpt;
1533 /* This shouldn't be possible */
1535 if (WARN_ON(!kvm->arch.mmu_ready))
1538 /* Stop VCPUs from running while we mess with the HPT */
1539 kvm->arch.mmu_ready = 0;
1542 /* Boot all CPUs out of the guest so they re-read
1544 on_each_cpu(resize_hpt_boot_vcpu, NULL, 1);
1547 if (!resize || (resize->order != shift))
1551 if (!resize->prepare_done)
1554 ret = resize->error;
1558 ret = resize_hpt_rehash(resize);
1562 resize_hpt_pivot(resize);
1565 /* Let VCPUs run again */
1566 kvm->arch.mmu_ready = 1;
1569 resize_hpt_release(kvm, resize);
1570 mutex_unlock(&kvm->lock);
1575 * Functions for reading and writing the hash table via reads and
1576 * writes on a file descriptor.
1578 * Reads return the guest view of the hash table, which has to be
1579 * pieced together from the real hash table and the guest_rpte
1580 * values in the revmap array.
1582 * On writes, each HPTE written is considered in turn, and if it
1583 * is valid, it is written to the HPT as if an H_ENTER with the
1584 * exact flag set was done. When the invalid count is non-zero
1585 * in the header written to the stream, the kernel will make
1586 * sure that that many HPTEs are invalid, and invalidate them
1590 struct kvm_htab_ctx {
1591 unsigned long index;
1592 unsigned long flags;
1597 #define HPTE_SIZE (2 * sizeof(unsigned long))
1600 * Returns 1 if this HPT entry has been modified or has pending
1603 static int hpte_dirty(struct revmap_entry *revp, __be64 *hptp)
1605 unsigned long rcbits_unset;
1607 if (revp->guest_rpte & HPTE_GR_MODIFIED)
1610 /* Also need to consider changes in reference and changed bits */
1611 rcbits_unset = ~revp->guest_rpte & (HPTE_R_R | HPTE_R_C);
1612 if ((be64_to_cpu(hptp[0]) & HPTE_V_VALID) &&
1613 (be64_to_cpu(hptp[1]) & rcbits_unset))
1619 static long record_hpte(unsigned long flags, __be64 *hptp,
1620 unsigned long *hpte, struct revmap_entry *revp,
1621 int want_valid, int first_pass)
1623 unsigned long v, r, hr;
1624 unsigned long rcbits_unset;
1628 /* Unmodified entries are uninteresting except on the first pass */
1629 dirty = hpte_dirty(revp, hptp);
1630 if (!first_pass && !dirty)
1634 if (be64_to_cpu(hptp[0]) & (HPTE_V_VALID | HPTE_V_ABSENT)) {
1636 if ((flags & KVM_GET_HTAB_BOLTED_ONLY) &&
1637 !(be64_to_cpu(hptp[0]) & HPTE_V_BOLTED))
1640 if (valid != want_valid)
1644 if (valid || dirty) {
1645 /* lock the HPTE so it's stable and read it */
1647 while (!try_lock_hpte(hptp, HPTE_V_HVLOCK))
1649 v = be64_to_cpu(hptp[0]);
1650 hr = be64_to_cpu(hptp[1]);
1651 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
1652 v = hpte_new_to_old_v(v, hr);
1653 hr = hpte_new_to_old_r(hr);
1656 /* re-evaluate valid and dirty from synchronized HPTE value */
1657 valid = !!(v & HPTE_V_VALID);
1658 dirty = !!(revp->guest_rpte & HPTE_GR_MODIFIED);
1660 /* Harvest R and C into guest view if necessary */
1661 rcbits_unset = ~revp->guest_rpte & (HPTE_R_R | HPTE_R_C);
1662 if (valid && (rcbits_unset & hr)) {
1663 revp->guest_rpte |= (hr &
1664 (HPTE_R_R | HPTE_R_C)) | HPTE_GR_MODIFIED;
1668 if (v & HPTE_V_ABSENT) {
1669 v &= ~HPTE_V_ABSENT;
1673 if ((flags & KVM_GET_HTAB_BOLTED_ONLY) && !(v & HPTE_V_BOLTED))
1676 r = revp->guest_rpte;
1677 /* only clear modified if this is the right sort of entry */
1678 if (valid == want_valid && dirty) {
1679 r &= ~HPTE_GR_MODIFIED;
1680 revp->guest_rpte = r;
1682 unlock_hpte(hptp, be64_to_cpu(hptp[0]));
1684 if (!(valid == want_valid && (first_pass || dirty)))
1687 hpte[0] = cpu_to_be64(v);
1688 hpte[1] = cpu_to_be64(r);
1692 static ssize_t kvm_htab_read(struct file *file, char __user *buf,
1693 size_t count, loff_t *ppos)
1695 struct kvm_htab_ctx *ctx = file->private_data;
1696 struct kvm *kvm = ctx->kvm;
1697 struct kvm_get_htab_header hdr;
1699 struct revmap_entry *revp;
1700 unsigned long i, nb, nw;
1701 unsigned long __user *lbuf;
1702 struct kvm_get_htab_header __user *hptr;
1703 unsigned long flags;
1705 unsigned long hpte[2];
1707 if (!access_ok(VERIFY_WRITE, buf, count))
1709 if (kvm_is_radix(kvm))
1712 first_pass = ctx->first_pass;
1716 hptp = (__be64 *)(kvm->arch.hpt.virt + (i * HPTE_SIZE));
1717 revp = kvm->arch.hpt.rev + i;
1718 lbuf = (unsigned long __user *)buf;
1721 while (nb + sizeof(hdr) + HPTE_SIZE < count) {
1722 /* Initialize header */
1723 hptr = (struct kvm_get_htab_header __user *)buf;
1728 lbuf = (unsigned long __user *)(buf + sizeof(hdr));
1730 /* Skip uninteresting entries, i.e. clean on not-first pass */
1732 while (i < kvmppc_hpt_npte(&kvm->arch.hpt) &&
1733 !hpte_dirty(revp, hptp)) {
1741 /* Grab a series of valid entries */
1742 while (i < kvmppc_hpt_npte(&kvm->arch.hpt) &&
1743 hdr.n_valid < 0xffff &&
1744 nb + HPTE_SIZE < count &&
1745 record_hpte(flags, hptp, hpte, revp, 1, first_pass)) {
1746 /* valid entry, write it out */
1748 if (__put_user(hpte[0], lbuf) ||
1749 __put_user(hpte[1], lbuf + 1))
1757 /* Now skip invalid entries while we can */
1758 while (i < kvmppc_hpt_npte(&kvm->arch.hpt) &&
1759 hdr.n_invalid < 0xffff &&
1760 record_hpte(flags, hptp, hpte, revp, 0, first_pass)) {
1761 /* found an invalid entry */
1768 if (hdr.n_valid || hdr.n_invalid) {
1769 /* write back the header */
1770 if (__copy_to_user(hptr, &hdr, sizeof(hdr)))
1773 buf = (char __user *)lbuf;
1778 /* Check if we've wrapped around the hash table */
1779 if (i >= kvmppc_hpt_npte(&kvm->arch.hpt)) {
1781 ctx->first_pass = 0;
1791 static ssize_t kvm_htab_write(struct file *file, const char __user *buf,
1792 size_t count, loff_t *ppos)
1794 struct kvm_htab_ctx *ctx = file->private_data;
1795 struct kvm *kvm = ctx->kvm;
1796 struct kvm_get_htab_header hdr;
1799 unsigned long __user *lbuf;
1801 unsigned long tmp[2];
1807 if (!access_ok(VERIFY_READ, buf, count))
1809 if (kvm_is_radix(kvm))
1812 /* lock out vcpus from running while we're doing this */
1813 mutex_lock(&kvm->lock);
1814 mmu_ready = kvm->arch.mmu_ready;
1816 kvm->arch.mmu_ready = 0; /* temporarily */
1817 /* order mmu_ready vs. vcpus_running */
1819 if (atomic_read(&kvm->arch.vcpus_running)) {
1820 kvm->arch.mmu_ready = 1;
1821 mutex_unlock(&kvm->lock);
1827 for (nb = 0; nb + sizeof(hdr) <= count; ) {
1829 if (__copy_from_user(&hdr, buf, sizeof(hdr)))
1833 if (nb + hdr.n_valid * HPTE_SIZE > count)
1841 if (i >= kvmppc_hpt_npte(&kvm->arch.hpt) ||
1842 i + hdr.n_valid + hdr.n_invalid > kvmppc_hpt_npte(&kvm->arch.hpt))
1845 hptp = (__be64 *)(kvm->arch.hpt.virt + (i * HPTE_SIZE));
1846 lbuf = (unsigned long __user *)buf;
1847 for (j = 0; j < hdr.n_valid; ++j) {
1852 if (__get_user(hpte_v, lbuf) ||
1853 __get_user(hpte_r, lbuf + 1))
1855 v = be64_to_cpu(hpte_v);
1856 r = be64_to_cpu(hpte_r);
1858 if (!(v & HPTE_V_VALID))
1860 pshift = kvmppc_hpte_base_page_shift(v, r);
1866 if (be64_to_cpu(hptp[0]) & (HPTE_V_VALID | HPTE_V_ABSENT))
1867 kvmppc_do_h_remove(kvm, 0, i, 0, tmp);
1869 ret = kvmppc_virtmode_do_h_enter(kvm, H_EXACT, i, v, r,
1871 if (ret != H_SUCCESS) {
1872 pr_err("kvm_htab_write ret %ld i=%ld v=%lx "
1873 "r=%lx\n", ret, i, v, r);
1876 if (!mmu_ready && is_vrma_hpte(v)) {
1877 unsigned long senc, lpcr;
1879 senc = slb_pgsize_encoding(1ul << pshift);
1880 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
1881 (VRMA_VSID << SLB_VSID_SHIFT_1T);
1882 if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
1883 lpcr = senc << (LPCR_VRMASD_SH - 4);
1884 kvmppc_update_lpcr(kvm, lpcr,
1887 kvmppc_setup_partition_table(kvm);
1895 for (j = 0; j < hdr.n_invalid; ++j) {
1896 if (be64_to_cpu(hptp[0]) & (HPTE_V_VALID | HPTE_V_ABSENT))
1897 kvmppc_do_h_remove(kvm, 0, i, 0, tmp);
1905 /* Order HPTE updates vs. mmu_ready */
1907 kvm->arch.mmu_ready = mmu_ready;
1908 mutex_unlock(&kvm->lock);
1915 static int kvm_htab_release(struct inode *inode, struct file *filp)
1917 struct kvm_htab_ctx *ctx = filp->private_data;
1919 filp->private_data = NULL;
1920 if (!(ctx->flags & KVM_GET_HTAB_WRITE))
1921 atomic_dec(&ctx->kvm->arch.hpte_mod_interest);
1922 kvm_put_kvm(ctx->kvm);
1927 static const struct file_operations kvm_htab_fops = {
1928 .read = kvm_htab_read,
1929 .write = kvm_htab_write,
1930 .llseek = default_llseek,
1931 .release = kvm_htab_release,
1934 int kvm_vm_ioctl_get_htab_fd(struct kvm *kvm, struct kvm_get_htab_fd *ghf)
1937 struct kvm_htab_ctx *ctx;
1940 /* reject flags we don't recognize */
1941 if (ghf->flags & ~(KVM_GET_HTAB_BOLTED_ONLY | KVM_GET_HTAB_WRITE))
1943 ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
1948 ctx->index = ghf->start_index;
1949 ctx->flags = ghf->flags;
1950 ctx->first_pass = 1;
1952 rwflag = (ghf->flags & KVM_GET_HTAB_WRITE) ? O_WRONLY : O_RDONLY;
1953 ret = anon_inode_getfd("kvm-htab", &kvm_htab_fops, ctx, rwflag | O_CLOEXEC);
1960 if (rwflag == O_RDONLY) {
1961 mutex_lock(&kvm->slots_lock);
1962 atomic_inc(&kvm->arch.hpte_mod_interest);
1963 /* make sure kvmppc_do_h_enter etc. see the increment */
1964 synchronize_srcu_expedited(&kvm->srcu);
1965 mutex_unlock(&kvm->slots_lock);
1971 struct debugfs_htab_state {
1974 unsigned long hpt_index;
1980 static int debugfs_htab_open(struct inode *inode, struct file *file)
1982 struct kvm *kvm = inode->i_private;
1983 struct debugfs_htab_state *p;
1985 p = kzalloc(sizeof(*p), GFP_KERNEL);
1991 mutex_init(&p->mutex);
1992 file->private_data = p;
1994 return nonseekable_open(inode, file);
1997 static int debugfs_htab_release(struct inode *inode, struct file *file)
1999 struct debugfs_htab_state *p = file->private_data;
2001 kvm_put_kvm(p->kvm);
2006 static ssize_t debugfs_htab_read(struct file *file, char __user *buf,
2007 size_t len, loff_t *ppos)
2009 struct debugfs_htab_state *p = file->private_data;
2012 unsigned long v, hr, gr;
2017 if (kvm_is_radix(kvm))
2020 ret = mutex_lock_interruptible(&p->mutex);
2024 if (p->chars_left) {
2028 r = copy_to_user(buf, p->buf + p->buf_index, n);
2043 hptp = (__be64 *)(kvm->arch.hpt.virt + (i * HPTE_SIZE));
2044 for (; len != 0 && i < kvmppc_hpt_npte(&kvm->arch.hpt);
2046 if (!(be64_to_cpu(hptp[0]) & (HPTE_V_VALID | HPTE_V_ABSENT)))
2049 /* lock the HPTE so it's stable and read it */
2051 while (!try_lock_hpte(hptp, HPTE_V_HVLOCK))
2053 v = be64_to_cpu(hptp[0]) & ~HPTE_V_HVLOCK;
2054 hr = be64_to_cpu(hptp[1]);
2055 gr = kvm->arch.hpt.rev[i].guest_rpte;
2056 unlock_hpte(hptp, v);
2059 if (!(v & (HPTE_V_VALID | HPTE_V_ABSENT)))
2062 n = scnprintf(p->buf, sizeof(p->buf),
2063 "%6lx %.16lx %.16lx %.16lx\n",
2068 r = copy_to_user(buf, p->buf, n);
2084 mutex_unlock(&p->mutex);
2088 static ssize_t debugfs_htab_write(struct file *file, const char __user *buf,
2089 size_t len, loff_t *ppos)
2094 static const struct file_operations debugfs_htab_fops = {
2095 .owner = THIS_MODULE,
2096 .open = debugfs_htab_open,
2097 .release = debugfs_htab_release,
2098 .read = debugfs_htab_read,
2099 .write = debugfs_htab_write,
2100 .llseek = generic_file_llseek,
2103 void kvmppc_mmu_debugfs_init(struct kvm *kvm)
2105 kvm->arch.htab_dentry = debugfs_create_file("htab", 0400,
2106 kvm->arch.debugfs_dir, kvm,
2107 &debugfs_htab_fops);
2110 void kvmppc_mmu_book3s_hv_init(struct kvm_vcpu *vcpu)
2112 struct kvmppc_mmu *mmu = &vcpu->arch.mmu;
2114 vcpu->arch.slb_nr = 32; /* POWER7/POWER8 */
2116 mmu->xlate = kvmppc_mmu_book3s_64_hv_xlate;
2117 mmu->reset_msr = kvmppc_mmu_book3s_64_hv_reset_msr;
2119 vcpu->arch.hflags |= BOOK3S_HFLAG_SLB;