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>
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 */
70 /* Private to the work thread, until prepare_done is true,
71 * then protected by kvm->resize_hpt_sem */
72 struct kvm_hpt_info hpt;
75 static void kvmppc_rmap_reset(struct kvm *kvm);
77 int kvmppc_allocate_hpt(struct kvm_hpt_info *info, u32 order)
79 unsigned long hpt = 0;
81 struct page *page = NULL;
82 struct revmap_entry *rev;
85 if ((order < PPC_MIN_HPT_ORDER) || (order > PPC_MAX_HPT_ORDER))
88 page = kvm_alloc_hpt_cma(1ul << (order - PAGE_SHIFT));
90 hpt = (unsigned long)pfn_to_kaddr(page_to_pfn(page));
91 memset((void *)hpt, 0, (1ul << order));
96 hpt = __get_free_pages(GFP_KERNEL|__GFP_ZERO|__GFP_RETRY_MAYFAIL
97 |__GFP_NOWARN, order - PAGE_SHIFT);
102 /* HPTEs are 2**4 bytes long */
103 npte = 1ul << (order - 4);
105 /* Allocate reverse map array */
106 rev = vmalloc(sizeof(struct revmap_entry) * npte);
108 pr_err("kvmppc_allocate_hpt: Couldn't alloc reverse map array\n");
110 kvm_free_hpt_cma(page, 1 << (order - PAGE_SHIFT));
112 free_pages(hpt, order - PAGE_SHIFT);
124 void kvmppc_set_hpt(struct kvm *kvm, struct kvm_hpt_info *info)
126 atomic64_set(&kvm->arch.mmio_update, 0);
127 kvm->arch.hpt = *info;
128 kvm->arch.sdr1 = __pa(info->virt) | (info->order - 18);
130 pr_debug("KVM guest htab at %lx (order %ld), LPID %x\n",
131 info->virt, (long)info->order, kvm->arch.lpid);
134 long kvmppc_alloc_reset_hpt(struct kvm *kvm, int order)
137 struct kvm_hpt_info info;
139 if (kvm_is_radix(kvm))
142 mutex_lock(&kvm->lock);
143 if (kvm->arch.hpte_setup_done) {
144 kvm->arch.hpte_setup_done = 0;
145 /* order hpte_setup_done vs. vcpus_running */
147 if (atomic_read(&kvm->arch.vcpus_running)) {
148 kvm->arch.hpte_setup_done = 1;
152 if (kvm->arch.hpt.order == order) {
153 /* We already have a suitable HPT */
155 /* Set the entire HPT to 0, i.e. invalid HPTEs */
156 memset((void *)kvm->arch.hpt.virt, 0, 1ul << order);
158 * Reset all the reverse-mapping chains for all memslots
160 kvmppc_rmap_reset(kvm);
161 /* Ensure that each vcpu will flush its TLB on next entry. */
162 cpumask_setall(&kvm->arch.need_tlb_flush);
167 if (kvm->arch.hpt.virt)
168 kvmppc_free_hpt(&kvm->arch.hpt);
170 err = kvmppc_allocate_hpt(&info, order);
173 kvmppc_set_hpt(kvm, &info);
176 mutex_unlock(&kvm->lock);
180 void kvmppc_free_hpt(struct kvm_hpt_info *info)
184 kvm_free_hpt_cma(virt_to_page(info->virt),
185 1 << (info->order - PAGE_SHIFT));
187 free_pages(info->virt, info->order - PAGE_SHIFT);
192 /* Bits in first HPTE dword for pagesize 4k, 64k or 16M */
193 static inline unsigned long hpte0_pgsize_encoding(unsigned long pgsize)
195 return (pgsize > 0x1000) ? HPTE_V_LARGE : 0;
198 /* Bits in second HPTE dword for pagesize 4k, 64k or 16M */
199 static inline unsigned long hpte1_pgsize_encoding(unsigned long pgsize)
201 return (pgsize == 0x10000) ? 0x1000 : 0;
204 void kvmppc_map_vrma(struct kvm_vcpu *vcpu, struct kvm_memory_slot *memslot,
205 unsigned long porder)
208 unsigned long npages;
209 unsigned long hp_v, hp_r;
210 unsigned long addr, hash;
212 unsigned long hp0, hp1;
213 unsigned long idx_ret;
215 struct kvm *kvm = vcpu->kvm;
217 psize = 1ul << porder;
218 npages = memslot->npages >> (porder - PAGE_SHIFT);
220 /* VRMA can't be > 1TB */
221 if (npages > 1ul << (40 - porder))
222 npages = 1ul << (40 - porder);
223 /* Can't use more than 1 HPTE per HPTEG */
224 if (npages > kvmppc_hpt_mask(&kvm->arch.hpt) + 1)
225 npages = kvmppc_hpt_mask(&kvm->arch.hpt) + 1;
227 hp0 = HPTE_V_1TB_SEG | (VRMA_VSID << (40 - 16)) |
228 HPTE_V_BOLTED | hpte0_pgsize_encoding(psize);
229 hp1 = hpte1_pgsize_encoding(psize) |
230 HPTE_R_R | HPTE_R_C | HPTE_R_M | PP_RWXX;
232 for (i = 0; i < npages; ++i) {
234 /* can't use hpt_hash since va > 64 bits */
235 hash = (i ^ (VRMA_VSID ^ (VRMA_VSID << 25)))
236 & kvmppc_hpt_mask(&kvm->arch.hpt);
238 * We assume that the hash table is empty and no
239 * vcpus are using it at this stage. Since we create
240 * at most one HPTE per HPTEG, we just assume entry 7
241 * is available and use it.
243 hash = (hash << 3) + 7;
244 hp_v = hp0 | ((addr >> 16) & ~0x7fUL);
246 ret = kvmppc_virtmode_do_h_enter(kvm, H_EXACT, hash, hp_v, hp_r,
248 if (ret != H_SUCCESS) {
249 pr_err("KVM: map_vrma at %lx failed, ret=%ld\n",
256 int kvmppc_mmu_hv_init(void)
258 unsigned long host_lpid, rsvd_lpid;
260 if (!cpu_has_feature(CPU_FTR_HVMODE))
263 /* POWER7 has 10-bit LPIDs (12-bit in POWER8) */
264 host_lpid = mfspr(SPRN_LPID);
265 rsvd_lpid = LPID_RSVD;
267 kvmppc_init_lpid(rsvd_lpid + 1);
269 kvmppc_claim_lpid(host_lpid);
270 /* rsvd_lpid is reserved for use in partition switching */
271 kvmppc_claim_lpid(rsvd_lpid);
276 static void kvmppc_mmu_book3s_64_hv_reset_msr(struct kvm_vcpu *vcpu)
278 unsigned long msr = vcpu->arch.intr_msr;
280 /* If transactional, change to suspend mode on IRQ delivery */
281 if (MSR_TM_TRANSACTIONAL(vcpu->arch.shregs.msr))
284 msr |= vcpu->arch.shregs.msr & MSR_TS_MASK;
285 kvmppc_set_msr(vcpu, msr);
288 static long kvmppc_virtmode_do_h_enter(struct kvm *kvm, unsigned long flags,
289 long pte_index, unsigned long pteh,
290 unsigned long ptel, unsigned long *pte_idx_ret)
294 /* Protect linux PTE lookup from page table destruction */
295 rcu_read_lock_sched(); /* this disables preemption too */
296 ret = kvmppc_do_h_enter(kvm, flags, pte_index, pteh, ptel,
297 current->mm->pgd, false, pte_idx_ret);
298 rcu_read_unlock_sched();
299 if (ret == H_TOO_HARD) {
300 /* this can't happen */
301 pr_err("KVM: Oops, kvmppc_h_enter returned too hard!\n");
302 ret = H_RESOURCE; /* or something */
308 static struct kvmppc_slb *kvmppc_mmu_book3s_hv_find_slbe(struct kvm_vcpu *vcpu,
314 for (i = 0; i < vcpu->arch.slb_nr; i++) {
315 if (!(vcpu->arch.slb[i].orige & SLB_ESID_V))
318 if (vcpu->arch.slb[i].origv & SLB_VSID_B_1T)
323 if (((vcpu->arch.slb[i].orige ^ eaddr) & mask) == 0)
324 return &vcpu->arch.slb[i];
329 static unsigned long kvmppc_mmu_get_real_addr(unsigned long v, unsigned long r,
332 unsigned long ra_mask;
334 ra_mask = hpte_page_size(v, r) - 1;
335 return (r & HPTE_R_RPN & ~ra_mask) | (ea & ra_mask);
338 static int kvmppc_mmu_book3s_64_hv_xlate(struct kvm_vcpu *vcpu, gva_t eaddr,
339 struct kvmppc_pte *gpte, bool data, bool iswrite)
341 struct kvm *kvm = vcpu->kvm;
342 struct kvmppc_slb *slbe;
344 unsigned long pp, key;
345 unsigned long v, orig_v, gr;
348 int virtmode = vcpu->arch.shregs.msr & (data ? MSR_DR : MSR_IR);
352 slbe = kvmppc_mmu_book3s_hv_find_slbe(vcpu, eaddr);
357 /* real mode access */
358 slb_v = vcpu->kvm->arch.vrma_slb_v;
362 /* Find the HPTE in the hash table */
363 index = kvmppc_hv_find_lock_hpte(kvm, eaddr, slb_v,
364 HPTE_V_VALID | HPTE_V_ABSENT);
369 hptep = (__be64 *)(kvm->arch.hpt.virt + (index << 4));
370 v = orig_v = be64_to_cpu(hptep[0]) & ~HPTE_V_HVLOCK;
371 if (cpu_has_feature(CPU_FTR_ARCH_300))
372 v = hpte_new_to_old_v(v, be64_to_cpu(hptep[1]));
373 gr = kvm->arch.hpt.rev[index].guest_rpte;
375 unlock_hpte(hptep, orig_v);
379 gpte->vpage = ((v & HPTE_V_AVPN) << 4) | ((eaddr >> 12) & 0xfff);
381 /* Get PP bits and key for permission check */
382 pp = gr & (HPTE_R_PP0 | HPTE_R_PP);
383 key = (vcpu->arch.shregs.msr & MSR_PR) ? SLB_VSID_KP : SLB_VSID_KS;
386 /* Calculate permissions */
387 gpte->may_read = hpte_read_permission(pp, key);
388 gpte->may_write = hpte_write_permission(pp, key);
389 gpte->may_execute = gpte->may_read && !(gr & (HPTE_R_N | HPTE_R_G));
391 /* Storage key permission check for POWER7 */
392 if (data && virtmode) {
393 int amrfield = hpte_get_skey_perm(gr, vcpu->arch.amr);
400 /* Get the guest physical address */
401 gpte->raddr = kvmppc_mmu_get_real_addr(v, gr, eaddr);
406 * Quick test for whether an instruction is a load or a store.
407 * If the instruction is a load or a store, then this will indicate
408 * which it is, at least on server processors. (Embedded processors
409 * have some external PID instructions that don't follow the rule
410 * embodied here.) If the instruction isn't a load or store, then
411 * this doesn't return anything useful.
413 static int instruction_is_store(unsigned int instr)
418 if ((instr & 0xfc000000) == 0x7c000000)
419 mask = 0x100; /* major opcode 31 */
420 return (instr & mask) != 0;
423 int kvmppc_hv_emulate_mmio(struct kvm_run *run, struct kvm_vcpu *vcpu,
424 unsigned long gpa, gva_t ea, int is_store)
429 * If we fail, we just return to the guest and try executing it again.
431 if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
436 * WARNING: We do not know for sure whether the instruction we just
437 * read from memory is the same that caused the fault in the first
438 * place. If the instruction we read is neither an load or a store,
439 * then it can't access memory, so we don't need to worry about
440 * enforcing access permissions. So, assuming it is a load or
441 * store, we just check that its direction (load or store) is
442 * consistent with the original fault, since that's what we
443 * checked the access permissions against. If there is a mismatch
444 * we just return and retry the instruction.
447 if (instruction_is_store(last_inst) != !!is_store)
451 * Emulated accesses are emulated by looking at the hash for
452 * translation once, then performing the access later. The
453 * translation could be invalidated in the meantime in which
454 * point performing the subsequent memory access on the old
455 * physical address could possibly be a security hole for the
456 * guest (but not the host).
458 * This is less of an issue for MMIO stores since they aren't
459 * globally visible. It could be an issue for MMIO loads to
460 * a certain extent but we'll ignore it for now.
463 vcpu->arch.paddr_accessed = gpa;
464 vcpu->arch.vaddr_accessed = ea;
465 return kvmppc_emulate_mmio(run, vcpu);
468 int kvmppc_book3s_hv_page_fault(struct kvm_run *run, struct kvm_vcpu *vcpu,
469 unsigned long ea, unsigned long dsisr)
471 struct kvm *kvm = vcpu->kvm;
472 unsigned long hpte[3], r;
473 unsigned long hnow_v, hnow_r;
475 unsigned long mmu_seq, psize, pte_size;
476 unsigned long gpa_base, gfn_base;
477 unsigned long gpa, gfn, hva, pfn;
478 struct kvm_memory_slot *memslot;
480 struct revmap_entry *rev;
481 struct page *page, *pages[1];
482 long index, ret, npages;
484 unsigned int writing, write_ok;
485 struct vm_area_struct *vma;
486 unsigned long rcbits;
489 if (kvm_is_radix(kvm))
490 return kvmppc_book3s_radix_page_fault(run, vcpu, ea, dsisr);
493 * Real-mode code has already searched the HPT and found the
494 * entry we're interested in. Lock the entry and check that
495 * it hasn't changed. If it has, just return and re-execute the
498 if (ea != vcpu->arch.pgfault_addr)
501 if (vcpu->arch.pgfault_cache) {
502 mmio_update = atomic64_read(&kvm->arch.mmio_update);
503 if (mmio_update == vcpu->arch.pgfault_cache->mmio_update) {
504 r = vcpu->arch.pgfault_cache->rpte;
505 psize = hpte_page_size(vcpu->arch.pgfault_hpte[0], r);
506 gpa_base = r & HPTE_R_RPN & ~(psize - 1);
507 gfn_base = gpa_base >> PAGE_SHIFT;
508 gpa = gpa_base | (ea & (psize - 1));
509 return kvmppc_hv_emulate_mmio(run, vcpu, gpa, ea,
510 dsisr & DSISR_ISSTORE);
513 index = vcpu->arch.pgfault_index;
514 hptep = (__be64 *)(kvm->arch.hpt.virt + (index << 4));
515 rev = &kvm->arch.hpt.rev[index];
517 while (!try_lock_hpte(hptep, HPTE_V_HVLOCK))
519 hpte[0] = be64_to_cpu(hptep[0]) & ~HPTE_V_HVLOCK;
520 hpte[1] = be64_to_cpu(hptep[1]);
521 hpte[2] = r = rev->guest_rpte;
522 unlock_hpte(hptep, hpte[0]);
525 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
526 hpte[0] = hpte_new_to_old_v(hpte[0], hpte[1]);
527 hpte[1] = hpte_new_to_old_r(hpte[1]);
529 if (hpte[0] != vcpu->arch.pgfault_hpte[0] ||
530 hpte[1] != vcpu->arch.pgfault_hpte[1])
533 /* Translate the logical address and get the page */
534 psize = hpte_page_size(hpte[0], r);
535 gpa_base = r & HPTE_R_RPN & ~(psize - 1);
536 gfn_base = gpa_base >> PAGE_SHIFT;
537 gpa = gpa_base | (ea & (psize - 1));
538 gfn = gpa >> PAGE_SHIFT;
539 memslot = gfn_to_memslot(kvm, gfn);
541 trace_kvm_page_fault_enter(vcpu, hpte, memslot, ea, dsisr);
543 /* No memslot means it's an emulated MMIO region */
544 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
545 return kvmppc_hv_emulate_mmio(run, vcpu, gpa, ea,
546 dsisr & DSISR_ISSTORE);
549 * This should never happen, because of the slot_is_aligned()
550 * check in kvmppc_do_h_enter().
552 if (gfn_base < memslot->base_gfn)
555 /* used to check for invalidations in progress */
556 mmu_seq = kvm->mmu_notifier_seq;
563 pte_size = PAGE_SIZE;
564 writing = (dsisr & DSISR_ISSTORE) != 0;
565 /* If writing != 0, then the HPTE must allow writing, if we get here */
567 hva = gfn_to_hva_memslot(memslot, gfn);
568 npages = get_user_pages_fast(hva, 1, writing, pages);
570 /* Check if it's an I/O mapping */
571 down_read(¤t->mm->mmap_sem);
572 vma = find_vma(current->mm, hva);
573 if (vma && vma->vm_start <= hva && hva + psize <= vma->vm_end &&
574 (vma->vm_flags & VM_PFNMAP)) {
575 pfn = vma->vm_pgoff +
576 ((hva - vma->vm_start) >> PAGE_SHIFT);
578 is_ci = pte_ci(__pte((pgprot_val(vma->vm_page_prot))));
579 write_ok = vma->vm_flags & VM_WRITE;
581 up_read(¤t->mm->mmap_sem);
586 pfn = page_to_pfn(page);
587 if (PageHuge(page)) {
588 page = compound_head(page);
589 pte_size <<= compound_order(page);
591 /* if the guest wants write access, see if that is OK */
592 if (!writing && hpte_is_writable(r)) {
596 * We need to protect against page table destruction
597 * hugepage split and collapse.
599 local_irq_save(flags);
600 ptep = find_linux_pte_or_hugepte(current->mm->pgd,
603 pte = kvmppc_read_update_linux_pte(ptep, 1);
604 if (__pte_write(pte))
607 local_irq_restore(flags);
611 if (psize > pte_size)
614 /* Check WIMG vs. the actual page we're accessing */
615 if (!hpte_cache_flags_ok(r, is_ci)) {
619 * Allow guest to map emulated device memory as
620 * uncacheable, but actually make it cacheable.
622 r = (r & ~(HPTE_R_W|HPTE_R_I|HPTE_R_G)) | HPTE_R_M;
626 * Set the HPTE to point to pfn.
627 * Since the pfn is at PAGE_SIZE granularity, make sure we
628 * don't mask out lower-order bits if psize < PAGE_SIZE.
630 if (psize < PAGE_SIZE)
632 r = (r & HPTE_R_KEY_HI) | (r & ~(HPTE_R_PP0 - psize)) |
633 ((pfn << PAGE_SHIFT) & ~(psize - 1));
634 if (hpte_is_writable(r) && !write_ok)
635 r = hpte_make_readonly(r);
638 while (!try_lock_hpte(hptep, HPTE_V_HVLOCK))
640 hnow_v = be64_to_cpu(hptep[0]);
641 hnow_r = be64_to_cpu(hptep[1]);
642 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
643 hnow_v = hpte_new_to_old_v(hnow_v, hnow_r);
644 hnow_r = hpte_new_to_old_r(hnow_r);
646 if ((hnow_v & ~HPTE_V_HVLOCK) != hpte[0] || hnow_r != hpte[1] ||
647 rev->guest_rpte != hpte[2])
648 /* HPTE has been changed under us; let the guest retry */
650 hpte[0] = (hpte[0] & ~HPTE_V_ABSENT) | HPTE_V_VALID;
652 /* Always put the HPTE in the rmap chain for the page base address */
653 rmap = &memslot->arch.rmap[gfn_base - memslot->base_gfn];
656 /* Check if we might have been invalidated; let the guest retry if so */
658 if (mmu_notifier_retry(vcpu->kvm, mmu_seq)) {
663 /* Only set R/C in real HPTE if set in both *rmap and guest_rpte */
664 rcbits = *rmap >> KVMPPC_RMAP_RC_SHIFT;
665 r &= rcbits | ~(HPTE_R_R | HPTE_R_C);
667 if (be64_to_cpu(hptep[0]) & HPTE_V_VALID) {
668 /* HPTE was previously valid, so we need to invalidate it */
670 hptep[0] |= cpu_to_be64(HPTE_V_ABSENT);
671 kvmppc_invalidate_hpte(kvm, hptep, index);
672 /* don't lose previous R and C bits */
673 r |= be64_to_cpu(hptep[1]) & (HPTE_R_R | HPTE_R_C);
675 kvmppc_add_revmap_chain(kvm, rev, rmap, index, 0);
678 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
679 r = hpte_old_to_new_r(hpte[0], r);
680 hpte[0] = hpte_old_to_new_v(hpte[0]);
682 hptep[1] = cpu_to_be64(r);
684 __unlock_hpte(hptep, hpte[0]);
685 asm volatile("ptesync" : : : "memory");
687 if (page && hpte_is_writable(r))
691 trace_kvm_page_fault_exit(vcpu, hpte, ret);
695 * We drop pages[0] here, not page because page might
696 * have been set to the head page of a compound, but
697 * we have to drop the reference on the correct tail
698 * page to match the get inside gup()
705 __unlock_hpte(hptep, be64_to_cpu(hptep[0]));
710 static void kvmppc_rmap_reset(struct kvm *kvm)
712 struct kvm_memslots *slots;
713 struct kvm_memory_slot *memslot;
716 srcu_idx = srcu_read_lock(&kvm->srcu);
717 slots = kvm_memslots(kvm);
718 kvm_for_each_memslot(memslot, slots) {
720 * This assumes it is acceptable to lose reference and
721 * change bits across a reset.
723 memset(memslot->arch.rmap, 0,
724 memslot->npages * sizeof(*memslot->arch.rmap));
726 srcu_read_unlock(&kvm->srcu, srcu_idx);
729 typedef int (*hva_handler_fn)(struct kvm *kvm, struct kvm_memory_slot *memslot,
732 static int kvm_handle_hva_range(struct kvm *kvm,
735 hva_handler_fn handler)
739 struct kvm_memslots *slots;
740 struct kvm_memory_slot *memslot;
742 slots = kvm_memslots(kvm);
743 kvm_for_each_memslot(memslot, slots) {
744 unsigned long hva_start, hva_end;
747 hva_start = max(start, memslot->userspace_addr);
748 hva_end = min(end, memslot->userspace_addr +
749 (memslot->npages << PAGE_SHIFT));
750 if (hva_start >= hva_end)
753 * {gfn(page) | page intersects with [hva_start, hva_end)} =
754 * {gfn, gfn+1, ..., gfn_end-1}.
756 gfn = hva_to_gfn_memslot(hva_start, memslot);
757 gfn_end = hva_to_gfn_memslot(hva_end + PAGE_SIZE - 1, memslot);
759 for (; gfn < gfn_end; ++gfn) {
760 ret = handler(kvm, memslot, gfn);
768 static int kvm_handle_hva(struct kvm *kvm, unsigned long hva,
769 hva_handler_fn handler)
771 return kvm_handle_hva_range(kvm, hva, hva + 1, handler);
774 /* Must be called with both HPTE and rmap locked */
775 static void kvmppc_unmap_hpte(struct kvm *kvm, unsigned long i,
776 unsigned long *rmapp, unsigned long gfn)
778 __be64 *hptep = (__be64 *) (kvm->arch.hpt.virt + (i << 4));
779 struct revmap_entry *rev = kvm->arch.hpt.rev;
781 unsigned long ptel, psize, rcbits;
785 /* chain is now empty */
786 *rmapp &= ~(KVMPPC_RMAP_PRESENT | KVMPPC_RMAP_INDEX);
788 /* remove i from chain */
792 rev[i].forw = rev[i].back = i;
793 *rmapp = (*rmapp & ~KVMPPC_RMAP_INDEX) | j;
796 /* Now check and modify the HPTE */
797 ptel = rev[i].guest_rpte;
798 psize = hpte_page_size(be64_to_cpu(hptep[0]), ptel);
799 if ((be64_to_cpu(hptep[0]) & HPTE_V_VALID) &&
800 hpte_rpn(ptel, psize) == gfn) {
801 hptep[0] |= cpu_to_be64(HPTE_V_ABSENT);
802 kvmppc_invalidate_hpte(kvm, hptep, i);
803 hptep[1] &= ~cpu_to_be64(HPTE_R_KEY_HI | HPTE_R_KEY_LO);
804 /* Harvest R and C */
805 rcbits = be64_to_cpu(hptep[1]) & (HPTE_R_R | HPTE_R_C);
806 *rmapp |= rcbits << KVMPPC_RMAP_RC_SHIFT;
807 if (rcbits & HPTE_R_C)
808 kvmppc_update_rmap_change(rmapp, psize);
809 if (rcbits & ~rev[i].guest_rpte) {
810 rev[i].guest_rpte = ptel | rcbits;
811 note_hpte_modification(kvm, &rev[i]);
816 static int kvm_unmap_rmapp(struct kvm *kvm, struct kvm_memory_slot *memslot,
821 unsigned long *rmapp;
823 rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn];
826 if (!(*rmapp & KVMPPC_RMAP_PRESENT)) {
832 * To avoid an ABBA deadlock with the HPTE lock bit,
833 * we can't spin on the HPTE lock while holding the
836 i = *rmapp & KVMPPC_RMAP_INDEX;
837 hptep = (__be64 *) (kvm->arch.hpt.virt + (i << 4));
838 if (!try_lock_hpte(hptep, HPTE_V_HVLOCK)) {
839 /* unlock rmap before spinning on the HPTE lock */
841 while (be64_to_cpu(hptep[0]) & HPTE_V_HVLOCK)
846 kvmppc_unmap_hpte(kvm, i, rmapp, gfn);
848 __unlock_hpte(hptep, be64_to_cpu(hptep[0]));
853 int kvm_unmap_hva_hv(struct kvm *kvm, unsigned long hva)
855 hva_handler_fn handler;
857 handler = kvm_is_radix(kvm) ? kvm_unmap_radix : kvm_unmap_rmapp;
858 kvm_handle_hva(kvm, hva, handler);
862 int kvm_unmap_hva_range_hv(struct kvm *kvm, unsigned long start, unsigned long end)
864 hva_handler_fn handler;
866 handler = kvm_is_radix(kvm) ? kvm_unmap_radix : kvm_unmap_rmapp;
867 kvm_handle_hva_range(kvm, start, end, handler);
871 void kvmppc_core_flush_memslot_hv(struct kvm *kvm,
872 struct kvm_memory_slot *memslot)
876 unsigned long *rmapp;
878 gfn = memslot->base_gfn;
879 rmapp = memslot->arch.rmap;
880 for (n = memslot->npages; n; --n, ++gfn) {
881 if (kvm_is_radix(kvm)) {
882 kvm_unmap_radix(kvm, memslot, gfn);
886 * Testing the present bit without locking is OK because
887 * the memslot has been marked invalid already, and hence
888 * no new HPTEs referencing this page can be created,
889 * thus the present bit can't go from 0 to 1.
891 if (*rmapp & KVMPPC_RMAP_PRESENT)
892 kvm_unmap_rmapp(kvm, memslot, gfn);
897 static int kvm_age_rmapp(struct kvm *kvm, struct kvm_memory_slot *memslot,
900 struct revmap_entry *rev = kvm->arch.hpt.rev;
901 unsigned long head, i, j;
904 unsigned long *rmapp;
906 rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn];
909 if (*rmapp & KVMPPC_RMAP_REFERENCED) {
910 *rmapp &= ~KVMPPC_RMAP_REFERENCED;
913 if (!(*rmapp & KVMPPC_RMAP_PRESENT)) {
918 i = head = *rmapp & KVMPPC_RMAP_INDEX;
920 hptep = (__be64 *) (kvm->arch.hpt.virt + (i << 4));
923 /* If this HPTE isn't referenced, ignore it */
924 if (!(be64_to_cpu(hptep[1]) & HPTE_R_R))
927 if (!try_lock_hpte(hptep, HPTE_V_HVLOCK)) {
928 /* unlock rmap before spinning on the HPTE lock */
930 while (be64_to_cpu(hptep[0]) & HPTE_V_HVLOCK)
935 /* Now check and modify the HPTE */
936 if ((be64_to_cpu(hptep[0]) & HPTE_V_VALID) &&
937 (be64_to_cpu(hptep[1]) & HPTE_R_R)) {
938 kvmppc_clear_ref_hpte(kvm, hptep, i);
939 if (!(rev[i].guest_rpte & HPTE_R_R)) {
940 rev[i].guest_rpte |= HPTE_R_R;
941 note_hpte_modification(kvm, &rev[i]);
945 __unlock_hpte(hptep, be64_to_cpu(hptep[0]));
946 } while ((i = j) != head);
952 int kvm_age_hva_hv(struct kvm *kvm, unsigned long start, unsigned long end)
954 hva_handler_fn handler;
956 handler = kvm_is_radix(kvm) ? kvm_age_radix : kvm_age_rmapp;
957 return kvm_handle_hva_range(kvm, start, end, handler);
960 static int kvm_test_age_rmapp(struct kvm *kvm, struct kvm_memory_slot *memslot,
963 struct revmap_entry *rev = kvm->arch.hpt.rev;
964 unsigned long head, i, j;
967 unsigned long *rmapp;
969 rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn];
970 if (*rmapp & KVMPPC_RMAP_REFERENCED)
974 if (*rmapp & KVMPPC_RMAP_REFERENCED)
977 if (*rmapp & KVMPPC_RMAP_PRESENT) {
978 i = head = *rmapp & KVMPPC_RMAP_INDEX;
980 hp = (unsigned long *)(kvm->arch.hpt.virt + (i << 4));
982 if (be64_to_cpu(hp[1]) & HPTE_R_R)
984 } while ((i = j) != head);
993 int kvm_test_age_hva_hv(struct kvm *kvm, unsigned long hva)
995 hva_handler_fn handler;
997 handler = kvm_is_radix(kvm) ? kvm_test_age_radix : kvm_test_age_rmapp;
998 return kvm_handle_hva(kvm, hva, handler);
1001 void kvm_set_spte_hva_hv(struct kvm *kvm, unsigned long hva, pte_t pte)
1003 hva_handler_fn handler;
1005 handler = kvm_is_radix(kvm) ? kvm_unmap_radix : kvm_unmap_rmapp;
1006 kvm_handle_hva(kvm, hva, handler);
1009 static int vcpus_running(struct kvm *kvm)
1011 return atomic_read(&kvm->arch.vcpus_running) != 0;
1015 * Returns the number of system pages that are dirty.
1016 * This can be more than 1 if we find a huge-page HPTE.
1018 static int kvm_test_clear_dirty_npages(struct kvm *kvm, unsigned long *rmapp)
1020 struct revmap_entry *rev = kvm->arch.hpt.rev;
1021 unsigned long head, i, j;
1025 int npages_dirty = 0;
1029 if (*rmapp & KVMPPC_RMAP_CHANGED) {
1030 long change_order = (*rmapp & KVMPPC_RMAP_CHG_ORDER)
1031 >> KVMPPC_RMAP_CHG_SHIFT;
1032 *rmapp &= ~(KVMPPC_RMAP_CHANGED | KVMPPC_RMAP_CHG_ORDER);
1034 if (change_order > PAGE_SHIFT)
1035 npages_dirty = 1ul << (change_order - PAGE_SHIFT);
1037 if (!(*rmapp & KVMPPC_RMAP_PRESENT)) {
1039 return npages_dirty;
1042 i = head = *rmapp & KVMPPC_RMAP_INDEX;
1044 unsigned long hptep1;
1045 hptep = (__be64 *) (kvm->arch.hpt.virt + (i << 4));
1049 * Checking the C (changed) bit here is racy since there
1050 * is no guarantee about when the hardware writes it back.
1051 * If the HPTE is not writable then it is stable since the
1052 * page can't be written to, and we would have done a tlbie
1053 * (which forces the hardware to complete any writeback)
1054 * when making the HPTE read-only.
1055 * If vcpus are running then this call is racy anyway
1056 * since the page could get dirtied subsequently, so we
1057 * expect there to be a further call which would pick up
1058 * any delayed C bit writeback.
1059 * Otherwise we need to do the tlbie even if C==0 in
1060 * order to pick up any delayed writeback of C.
1062 hptep1 = be64_to_cpu(hptep[1]);
1063 if (!(hptep1 & HPTE_R_C) &&
1064 (!hpte_is_writable(hptep1) || vcpus_running(kvm)))
1067 if (!try_lock_hpte(hptep, HPTE_V_HVLOCK)) {
1068 /* unlock rmap before spinning on the HPTE lock */
1070 while (hptep[0] & cpu_to_be64(HPTE_V_HVLOCK))
1075 /* Now check and modify the HPTE */
1076 if (!(hptep[0] & cpu_to_be64(HPTE_V_VALID))) {
1077 __unlock_hpte(hptep, be64_to_cpu(hptep[0]));
1081 /* need to make it temporarily absent so C is stable */
1082 hptep[0] |= cpu_to_be64(HPTE_V_ABSENT);
1083 kvmppc_invalidate_hpte(kvm, hptep, i);
1084 v = be64_to_cpu(hptep[0]);
1085 r = be64_to_cpu(hptep[1]);
1087 hptep[1] = cpu_to_be64(r & ~HPTE_R_C);
1088 if (!(rev[i].guest_rpte & HPTE_R_C)) {
1089 rev[i].guest_rpte |= HPTE_R_C;
1090 note_hpte_modification(kvm, &rev[i]);
1092 n = hpte_page_size(v, r);
1093 n = (n + PAGE_SIZE - 1) >> PAGE_SHIFT;
1094 if (n > npages_dirty)
1098 v &= ~HPTE_V_ABSENT;
1100 __unlock_hpte(hptep, v);
1101 } while ((i = j) != head);
1104 return npages_dirty;
1107 void kvmppc_harvest_vpa_dirty(struct kvmppc_vpa *vpa,
1108 struct kvm_memory_slot *memslot,
1113 if (!vpa->dirty || !vpa->pinned_addr)
1115 gfn = vpa->gpa >> PAGE_SHIFT;
1116 if (gfn < memslot->base_gfn ||
1117 gfn >= memslot->base_gfn + memslot->npages)
1122 __set_bit_le(gfn - memslot->base_gfn, map);
1125 long kvmppc_hv_get_dirty_log_hpt(struct kvm *kvm,
1126 struct kvm_memory_slot *memslot, unsigned long *map)
1129 unsigned long *rmapp;
1132 rmapp = memslot->arch.rmap;
1133 for (i = 0; i < memslot->npages; ++i) {
1134 int npages = kvm_test_clear_dirty_npages(kvm, rmapp);
1136 * Note that if npages > 0 then i must be a multiple of npages,
1137 * since we always put huge-page HPTEs in the rmap chain
1138 * corresponding to their page base address.
1141 for (j = i; npages; ++j, --npages)
1142 __set_bit_le(j, map);
1149 void *kvmppc_pin_guest_page(struct kvm *kvm, unsigned long gpa,
1150 unsigned long *nb_ret)
1152 struct kvm_memory_slot *memslot;
1153 unsigned long gfn = gpa >> PAGE_SHIFT;
1154 struct page *page, *pages[1];
1156 unsigned long hva, offset;
1159 srcu_idx = srcu_read_lock(&kvm->srcu);
1160 memslot = gfn_to_memslot(kvm, gfn);
1161 if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
1163 hva = gfn_to_hva_memslot(memslot, gfn);
1164 npages = get_user_pages_fast(hva, 1, 1, pages);
1168 srcu_read_unlock(&kvm->srcu, srcu_idx);
1170 offset = gpa & (PAGE_SIZE - 1);
1172 *nb_ret = PAGE_SIZE - offset;
1173 return page_address(page) + offset;
1176 srcu_read_unlock(&kvm->srcu, srcu_idx);
1180 void kvmppc_unpin_guest_page(struct kvm *kvm, void *va, unsigned long gpa,
1183 struct page *page = virt_to_page(va);
1184 struct kvm_memory_slot *memslot;
1186 unsigned long *rmap;
1194 /* We need to mark this page dirty in the rmap chain */
1195 gfn = gpa >> PAGE_SHIFT;
1196 srcu_idx = srcu_read_lock(&kvm->srcu);
1197 memslot = gfn_to_memslot(kvm, gfn);
1199 if (!kvm_is_radix(kvm)) {
1200 rmap = &memslot->arch.rmap[gfn - memslot->base_gfn];
1202 *rmap |= KVMPPC_RMAP_CHANGED;
1204 } else if (memslot->dirty_bitmap) {
1205 mark_page_dirty(kvm, gfn);
1208 srcu_read_unlock(&kvm->srcu, srcu_idx);
1214 static int resize_hpt_allocate(struct kvm_resize_hpt *resize)
1218 rc = kvmppc_allocate_hpt(&resize->hpt, resize->order);
1222 resize_hpt_debug(resize, "resize_hpt_allocate(): HPT @ 0x%lx\n",
1228 static unsigned long resize_hpt_rehash_hpte(struct kvm_resize_hpt *resize,
1231 struct kvm *kvm = resize->kvm;
1232 struct kvm_hpt_info *old = &kvm->arch.hpt;
1233 struct kvm_hpt_info *new = &resize->hpt;
1234 unsigned long old_hash_mask = (1ULL << (old->order - 7)) - 1;
1235 unsigned long new_hash_mask = (1ULL << (new->order - 7)) - 1;
1236 __be64 *hptep, *new_hptep;
1237 unsigned long vpte, rpte, guest_rpte;
1239 struct revmap_entry *rev;
1240 unsigned long apsize, psize, avpn, pteg, hash;
1241 unsigned long new_idx, new_pteg, replace_vpte;
1243 hptep = (__be64 *)(old->virt + (idx << 4));
1245 /* Guest is stopped, so new HPTEs can't be added or faulted
1246 * in, only unmapped or altered by host actions. So, it's
1247 * safe to check this before we take the HPTE lock */
1248 vpte = be64_to_cpu(hptep[0]);
1249 if (!(vpte & HPTE_V_VALID) && !(vpte & HPTE_V_ABSENT))
1250 return 0; /* nothing to do */
1252 while (!try_lock_hpte(hptep, HPTE_V_HVLOCK))
1255 vpte = be64_to_cpu(hptep[0]);
1258 if (!(vpte & HPTE_V_VALID) && !(vpte & HPTE_V_ABSENT))
1263 rev = &old->rev[idx];
1264 guest_rpte = rev->guest_rpte;
1267 apsize = hpte_page_size(vpte, guest_rpte);
1271 if (vpte & HPTE_V_VALID) {
1272 unsigned long gfn = hpte_rpn(guest_rpte, apsize);
1273 int srcu_idx = srcu_read_lock(&kvm->srcu);
1274 struct kvm_memory_slot *memslot =
1275 __gfn_to_memslot(kvm_memslots(kvm), gfn);
1278 unsigned long *rmapp;
1279 rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn];
1282 kvmppc_unmap_hpte(kvm, idx, rmapp, gfn);
1286 srcu_read_unlock(&kvm->srcu, srcu_idx);
1289 /* Reload PTE after unmap */
1290 vpte = be64_to_cpu(hptep[0]);
1292 BUG_ON(vpte & HPTE_V_VALID);
1293 BUG_ON(!(vpte & HPTE_V_ABSENT));
1296 if (!(vpte & HPTE_V_BOLTED))
1299 rpte = be64_to_cpu(hptep[1]);
1300 psize = hpte_base_page_size(vpte, rpte);
1301 avpn = HPTE_V_AVPN_VAL(vpte) & ~((psize - 1) >> 23);
1302 pteg = idx / HPTES_PER_GROUP;
1303 if (vpte & HPTE_V_SECONDARY)
1306 if (!(vpte & HPTE_V_1TB_SEG)) {
1307 unsigned long offset, vsid;
1309 /* We only have 28 - 23 bits of offset in avpn */
1310 offset = (avpn & 0x1f) << 23;
1312 /* We can find more bits from the pteg value */
1313 if (psize < (1ULL << 23))
1314 offset |= ((vsid ^ pteg) & old_hash_mask) * psize;
1316 hash = vsid ^ (offset / psize);
1318 unsigned long offset, vsid;
1320 /* We only have 40 - 23 bits of seg_off in avpn */
1321 offset = (avpn & 0x1ffff) << 23;
1323 if (psize < (1ULL << 23))
1324 offset |= ((vsid ^ (vsid << 25) ^ pteg) & old_hash_mask) * psize;
1326 hash = vsid ^ (vsid << 25) ^ (offset / psize);
1329 new_pteg = hash & new_hash_mask;
1330 if (vpte & HPTE_V_SECONDARY) {
1331 BUG_ON(~pteg != (hash & old_hash_mask));
1332 new_pteg = ~new_pteg;
1334 BUG_ON(pteg != (hash & old_hash_mask));
1337 new_idx = new_pteg * HPTES_PER_GROUP + (idx % HPTES_PER_GROUP);
1338 new_hptep = (__be64 *)(new->virt + (new_idx << 4));
1340 replace_vpte = be64_to_cpu(new_hptep[0]);
1342 if (replace_vpte & (HPTE_V_VALID | HPTE_V_ABSENT)) {
1343 BUG_ON(new->order >= old->order);
1345 if (replace_vpte & HPTE_V_BOLTED) {
1346 if (vpte & HPTE_V_BOLTED)
1347 /* Bolted collision, nothing we can do */
1349 /* Discard the new HPTE */
1353 /* Discard the previous HPTE */
1356 new_hptep[1] = cpu_to_be64(rpte);
1357 new->rev[new_idx].guest_rpte = guest_rpte;
1358 /* No need for a barrier, since new HPT isn't active */
1359 new_hptep[0] = cpu_to_be64(vpte);
1360 unlock_hpte(new_hptep, vpte);
1363 unlock_hpte(hptep, vpte);
1367 static int resize_hpt_rehash(struct kvm_resize_hpt *resize)
1369 struct kvm *kvm = resize->kvm;
1374 * resize_hpt_rehash_hpte() doesn't handle the new-format HPTEs
1375 * that POWER9 uses, and could well hit a BUG_ON on POWER9.
1377 if (cpu_has_feature(CPU_FTR_ARCH_300))
1379 for (i = 0; i < kvmppc_hpt_npte(&kvm->arch.hpt); i++) {
1380 rc = resize_hpt_rehash_hpte(resize, i);
1388 static void resize_hpt_pivot(struct kvm_resize_hpt *resize)
1390 struct kvm *kvm = resize->kvm;
1391 struct kvm_hpt_info hpt_tmp;
1393 /* Exchange the pending tables in the resize structure with
1394 * the active tables */
1396 resize_hpt_debug(resize, "resize_hpt_pivot()\n");
1398 spin_lock(&kvm->mmu_lock);
1399 asm volatile("ptesync" : : : "memory");
1401 hpt_tmp = kvm->arch.hpt;
1402 kvmppc_set_hpt(kvm, &resize->hpt);
1403 resize->hpt = hpt_tmp;
1405 spin_unlock(&kvm->mmu_lock);
1407 synchronize_srcu_expedited(&kvm->srcu);
1409 resize_hpt_debug(resize, "resize_hpt_pivot() done\n");
1412 static void resize_hpt_release(struct kvm *kvm, struct kvm_resize_hpt *resize)
1414 BUG_ON(kvm->arch.resize_hpt != resize);
1419 if (resize->hpt.virt)
1420 kvmppc_free_hpt(&resize->hpt);
1422 kvm->arch.resize_hpt = NULL;
1426 static void resize_hpt_prepare_work(struct work_struct *work)
1428 struct kvm_resize_hpt *resize = container_of(work,
1429 struct kvm_resize_hpt,
1431 struct kvm *kvm = resize->kvm;
1434 resize_hpt_debug(resize, "resize_hpt_prepare_work(): order = %d\n",
1437 err = resize_hpt_allocate(resize);
1439 mutex_lock(&kvm->lock);
1441 resize->error = err;
1442 resize->prepare_done = true;
1444 mutex_unlock(&kvm->lock);
1447 long kvm_vm_ioctl_resize_hpt_prepare(struct kvm *kvm,
1448 struct kvm_ppc_resize_hpt *rhpt)
1450 unsigned long flags = rhpt->flags;
1451 unsigned long shift = rhpt->shift;
1452 struct kvm_resize_hpt *resize;
1458 if (shift && ((shift < 18) || (shift > 46)))
1461 mutex_lock(&kvm->lock);
1463 resize = kvm->arch.resize_hpt;
1466 if (resize->order == shift) {
1467 /* Suitable resize in progress */
1468 if (resize->prepare_done) {
1469 ret = resize->error;
1471 resize_hpt_release(kvm, resize);
1473 ret = 100; /* estimated time in ms */
1479 /* not suitable, cancel it */
1480 resize_hpt_release(kvm, resize);
1485 goto out; /* nothing to do */
1487 /* start new resize */
1489 resize = kzalloc(sizeof(*resize), GFP_KERNEL);
1494 resize->order = shift;
1496 INIT_WORK(&resize->work, resize_hpt_prepare_work);
1497 kvm->arch.resize_hpt = resize;
1499 schedule_work(&resize->work);
1501 ret = 100; /* estimated time in ms */
1504 mutex_unlock(&kvm->lock);
1508 static void resize_hpt_boot_vcpu(void *opaque)
1510 /* Nothing to do, just force a KVM exit */
1513 long kvm_vm_ioctl_resize_hpt_commit(struct kvm *kvm,
1514 struct kvm_ppc_resize_hpt *rhpt)
1516 unsigned long flags = rhpt->flags;
1517 unsigned long shift = rhpt->shift;
1518 struct kvm_resize_hpt *resize;
1524 if (shift && ((shift < 18) || (shift > 46)))
1527 mutex_lock(&kvm->lock);
1529 resize = kvm->arch.resize_hpt;
1531 /* This shouldn't be possible */
1533 if (WARN_ON(!kvm->arch.hpte_setup_done))
1536 /* Stop VCPUs from running while we mess with the HPT */
1537 kvm->arch.hpte_setup_done = 0;
1540 /* Boot all CPUs out of the guest so they re-read
1541 * hpte_setup_done */
1542 on_each_cpu(resize_hpt_boot_vcpu, NULL, 1);
1545 if (!resize || (resize->order != shift))
1549 if (!resize->prepare_done)
1552 ret = resize->error;
1556 ret = resize_hpt_rehash(resize);
1560 resize_hpt_pivot(resize);
1563 /* Let VCPUs run again */
1564 kvm->arch.hpte_setup_done = 1;
1567 resize_hpt_release(kvm, resize);
1568 mutex_unlock(&kvm->lock);
1573 * Functions for reading and writing the hash table via reads and
1574 * writes on a file descriptor.
1576 * Reads return the guest view of the hash table, which has to be
1577 * pieced together from the real hash table and the guest_rpte
1578 * values in the revmap array.
1580 * On writes, each HPTE written is considered in turn, and if it
1581 * is valid, it is written to the HPT as if an H_ENTER with the
1582 * exact flag set was done. When the invalid count is non-zero
1583 * in the header written to the stream, the kernel will make
1584 * sure that that many HPTEs are invalid, and invalidate them
1588 struct kvm_htab_ctx {
1589 unsigned long index;
1590 unsigned long flags;
1595 #define HPTE_SIZE (2 * sizeof(unsigned long))
1598 * Returns 1 if this HPT entry has been modified or has pending
1601 static int hpte_dirty(struct revmap_entry *revp, __be64 *hptp)
1603 unsigned long rcbits_unset;
1605 if (revp->guest_rpte & HPTE_GR_MODIFIED)
1608 /* Also need to consider changes in reference and changed bits */
1609 rcbits_unset = ~revp->guest_rpte & (HPTE_R_R | HPTE_R_C);
1610 if ((be64_to_cpu(hptp[0]) & HPTE_V_VALID) &&
1611 (be64_to_cpu(hptp[1]) & rcbits_unset))
1617 static long record_hpte(unsigned long flags, __be64 *hptp,
1618 unsigned long *hpte, struct revmap_entry *revp,
1619 int want_valid, int first_pass)
1621 unsigned long v, r, hr;
1622 unsigned long rcbits_unset;
1626 /* Unmodified entries are uninteresting except on the first pass */
1627 dirty = hpte_dirty(revp, hptp);
1628 if (!first_pass && !dirty)
1632 if (be64_to_cpu(hptp[0]) & (HPTE_V_VALID | HPTE_V_ABSENT)) {
1634 if ((flags & KVM_GET_HTAB_BOLTED_ONLY) &&
1635 !(be64_to_cpu(hptp[0]) & HPTE_V_BOLTED))
1638 if (valid != want_valid)
1642 if (valid || dirty) {
1643 /* lock the HPTE so it's stable and read it */
1645 while (!try_lock_hpte(hptp, HPTE_V_HVLOCK))
1647 v = be64_to_cpu(hptp[0]);
1648 hr = be64_to_cpu(hptp[1]);
1649 if (cpu_has_feature(CPU_FTR_ARCH_300)) {
1650 v = hpte_new_to_old_v(v, hr);
1651 hr = hpte_new_to_old_r(hr);
1654 /* re-evaluate valid and dirty from synchronized HPTE value */
1655 valid = !!(v & HPTE_V_VALID);
1656 dirty = !!(revp->guest_rpte & HPTE_GR_MODIFIED);
1658 /* Harvest R and C into guest view if necessary */
1659 rcbits_unset = ~revp->guest_rpte & (HPTE_R_R | HPTE_R_C);
1660 if (valid && (rcbits_unset & hr)) {
1661 revp->guest_rpte |= (hr &
1662 (HPTE_R_R | HPTE_R_C)) | HPTE_GR_MODIFIED;
1666 if (v & HPTE_V_ABSENT) {
1667 v &= ~HPTE_V_ABSENT;
1671 if ((flags & KVM_GET_HTAB_BOLTED_ONLY) && !(v & HPTE_V_BOLTED))
1674 r = revp->guest_rpte;
1675 /* only clear modified if this is the right sort of entry */
1676 if (valid == want_valid && dirty) {
1677 r &= ~HPTE_GR_MODIFIED;
1678 revp->guest_rpte = r;
1680 unlock_hpte(hptp, be64_to_cpu(hptp[0]));
1682 if (!(valid == want_valid && (first_pass || dirty)))
1685 hpte[0] = cpu_to_be64(v);
1686 hpte[1] = cpu_to_be64(r);
1690 static ssize_t kvm_htab_read(struct file *file, char __user *buf,
1691 size_t count, loff_t *ppos)
1693 struct kvm_htab_ctx *ctx = file->private_data;
1694 struct kvm *kvm = ctx->kvm;
1695 struct kvm_get_htab_header hdr;
1697 struct revmap_entry *revp;
1698 unsigned long i, nb, nw;
1699 unsigned long __user *lbuf;
1700 struct kvm_get_htab_header __user *hptr;
1701 unsigned long flags;
1703 unsigned long hpte[2];
1705 if (!access_ok(VERIFY_WRITE, buf, count))
1708 first_pass = ctx->first_pass;
1712 hptp = (__be64 *)(kvm->arch.hpt.virt + (i * HPTE_SIZE));
1713 revp = kvm->arch.hpt.rev + i;
1714 lbuf = (unsigned long __user *)buf;
1717 while (nb + sizeof(hdr) + HPTE_SIZE < count) {
1718 /* Initialize header */
1719 hptr = (struct kvm_get_htab_header __user *)buf;
1724 lbuf = (unsigned long __user *)(buf + sizeof(hdr));
1726 /* Skip uninteresting entries, i.e. clean on not-first pass */
1728 while (i < kvmppc_hpt_npte(&kvm->arch.hpt) &&
1729 !hpte_dirty(revp, hptp)) {
1737 /* Grab a series of valid entries */
1738 while (i < kvmppc_hpt_npte(&kvm->arch.hpt) &&
1739 hdr.n_valid < 0xffff &&
1740 nb + HPTE_SIZE < count &&
1741 record_hpte(flags, hptp, hpte, revp, 1, first_pass)) {
1742 /* valid entry, write it out */
1744 if (__put_user(hpte[0], lbuf) ||
1745 __put_user(hpte[1], lbuf + 1))
1753 /* Now skip invalid entries while we can */
1754 while (i < kvmppc_hpt_npte(&kvm->arch.hpt) &&
1755 hdr.n_invalid < 0xffff &&
1756 record_hpte(flags, hptp, hpte, revp, 0, first_pass)) {
1757 /* found an invalid entry */
1764 if (hdr.n_valid || hdr.n_invalid) {
1765 /* write back the header */
1766 if (__copy_to_user(hptr, &hdr, sizeof(hdr)))
1769 buf = (char __user *)lbuf;
1774 /* Check if we've wrapped around the hash table */
1775 if (i >= kvmppc_hpt_npte(&kvm->arch.hpt)) {
1777 ctx->first_pass = 0;
1787 static ssize_t kvm_htab_write(struct file *file, const char __user *buf,
1788 size_t count, loff_t *ppos)
1790 struct kvm_htab_ctx *ctx = file->private_data;
1791 struct kvm *kvm = ctx->kvm;
1792 struct kvm_get_htab_header hdr;
1795 unsigned long __user *lbuf;
1797 unsigned long tmp[2];
1802 if (!access_ok(VERIFY_READ, buf, count))
1805 /* lock out vcpus from running while we're doing this */
1806 mutex_lock(&kvm->lock);
1807 hpte_setup = kvm->arch.hpte_setup_done;
1809 kvm->arch.hpte_setup_done = 0; /* temporarily */
1810 /* order hpte_setup_done vs. vcpus_running */
1812 if (atomic_read(&kvm->arch.vcpus_running)) {
1813 kvm->arch.hpte_setup_done = 1;
1814 mutex_unlock(&kvm->lock);
1820 for (nb = 0; nb + sizeof(hdr) <= count; ) {
1822 if (__copy_from_user(&hdr, buf, sizeof(hdr)))
1826 if (nb + hdr.n_valid * HPTE_SIZE > count)
1834 if (i >= kvmppc_hpt_npte(&kvm->arch.hpt) ||
1835 i + hdr.n_valid + hdr.n_invalid > kvmppc_hpt_npte(&kvm->arch.hpt))
1838 hptp = (__be64 *)(kvm->arch.hpt.virt + (i * HPTE_SIZE));
1839 lbuf = (unsigned long __user *)buf;
1840 for (j = 0; j < hdr.n_valid; ++j) {
1845 if (__get_user(hpte_v, lbuf) ||
1846 __get_user(hpte_r, lbuf + 1))
1848 v = be64_to_cpu(hpte_v);
1849 r = be64_to_cpu(hpte_r);
1851 if (!(v & HPTE_V_VALID))
1856 if (be64_to_cpu(hptp[0]) & (HPTE_V_VALID | HPTE_V_ABSENT))
1857 kvmppc_do_h_remove(kvm, 0, i, 0, tmp);
1859 ret = kvmppc_virtmode_do_h_enter(kvm, H_EXACT, i, v, r,
1861 if (ret != H_SUCCESS) {
1862 pr_err("kvm_htab_write ret %ld i=%ld v=%lx "
1863 "r=%lx\n", ret, i, v, r);
1866 if (!hpte_setup && is_vrma_hpte(v)) {
1867 unsigned long psize = hpte_base_page_size(v, r);
1868 unsigned long senc = slb_pgsize_encoding(psize);
1871 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
1872 (VRMA_VSID << SLB_VSID_SHIFT_1T);
1873 lpcr = senc << (LPCR_VRMASD_SH - 4);
1874 kvmppc_update_lpcr(kvm, lpcr, LPCR_VRMASD);
1881 for (j = 0; j < hdr.n_invalid; ++j) {
1882 if (be64_to_cpu(hptp[0]) & (HPTE_V_VALID | HPTE_V_ABSENT))
1883 kvmppc_do_h_remove(kvm, 0, i, 0, tmp);
1891 /* Order HPTE updates vs. hpte_setup_done */
1893 kvm->arch.hpte_setup_done = hpte_setup;
1894 mutex_unlock(&kvm->lock);
1901 static int kvm_htab_release(struct inode *inode, struct file *filp)
1903 struct kvm_htab_ctx *ctx = filp->private_data;
1905 filp->private_data = NULL;
1906 if (!(ctx->flags & KVM_GET_HTAB_WRITE))
1907 atomic_dec(&ctx->kvm->arch.hpte_mod_interest);
1908 kvm_put_kvm(ctx->kvm);
1913 static const struct file_operations kvm_htab_fops = {
1914 .read = kvm_htab_read,
1915 .write = kvm_htab_write,
1916 .llseek = default_llseek,
1917 .release = kvm_htab_release,
1920 int kvm_vm_ioctl_get_htab_fd(struct kvm *kvm, struct kvm_get_htab_fd *ghf)
1923 struct kvm_htab_ctx *ctx;
1926 /* reject flags we don't recognize */
1927 if (ghf->flags & ~(KVM_GET_HTAB_BOLTED_ONLY | KVM_GET_HTAB_WRITE))
1929 ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
1934 ctx->index = ghf->start_index;
1935 ctx->flags = ghf->flags;
1936 ctx->first_pass = 1;
1938 rwflag = (ghf->flags & KVM_GET_HTAB_WRITE) ? O_WRONLY : O_RDONLY;
1939 ret = anon_inode_getfd("kvm-htab", &kvm_htab_fops, ctx, rwflag | O_CLOEXEC);
1945 if (rwflag == O_RDONLY) {
1946 mutex_lock(&kvm->slots_lock);
1947 atomic_inc(&kvm->arch.hpte_mod_interest);
1948 /* make sure kvmppc_do_h_enter etc. see the increment */
1949 synchronize_srcu_expedited(&kvm->srcu);
1950 mutex_unlock(&kvm->slots_lock);
1956 struct debugfs_htab_state {
1959 unsigned long hpt_index;
1965 static int debugfs_htab_open(struct inode *inode, struct file *file)
1967 struct kvm *kvm = inode->i_private;
1968 struct debugfs_htab_state *p;
1970 p = kzalloc(sizeof(*p), GFP_KERNEL);
1976 mutex_init(&p->mutex);
1977 file->private_data = p;
1979 return nonseekable_open(inode, file);
1982 static int debugfs_htab_release(struct inode *inode, struct file *file)
1984 struct debugfs_htab_state *p = file->private_data;
1986 kvm_put_kvm(p->kvm);
1991 static ssize_t debugfs_htab_read(struct file *file, char __user *buf,
1992 size_t len, loff_t *ppos)
1994 struct debugfs_htab_state *p = file->private_data;
1997 unsigned long v, hr, gr;
2001 ret = mutex_lock_interruptible(&p->mutex);
2005 if (p->chars_left) {
2009 r = copy_to_user(buf, p->buf + p->buf_index, n);
2025 hptp = (__be64 *)(kvm->arch.hpt.virt + (i * HPTE_SIZE));
2026 for (; len != 0 && i < kvmppc_hpt_npte(&kvm->arch.hpt);
2028 if (!(be64_to_cpu(hptp[0]) & (HPTE_V_VALID | HPTE_V_ABSENT)))
2031 /* lock the HPTE so it's stable and read it */
2033 while (!try_lock_hpte(hptp, HPTE_V_HVLOCK))
2035 v = be64_to_cpu(hptp[0]) & ~HPTE_V_HVLOCK;
2036 hr = be64_to_cpu(hptp[1]);
2037 gr = kvm->arch.hpt.rev[i].guest_rpte;
2038 unlock_hpte(hptp, v);
2041 if (!(v & (HPTE_V_VALID | HPTE_V_ABSENT)))
2044 n = scnprintf(p->buf, sizeof(p->buf),
2045 "%6lx %.16lx %.16lx %.16lx\n",
2050 r = copy_to_user(buf, p->buf, n);
2066 mutex_unlock(&p->mutex);
2070 static ssize_t debugfs_htab_write(struct file *file, const char __user *buf,
2071 size_t len, loff_t *ppos)
2076 static const struct file_operations debugfs_htab_fops = {
2077 .owner = THIS_MODULE,
2078 .open = debugfs_htab_open,
2079 .release = debugfs_htab_release,
2080 .read = debugfs_htab_read,
2081 .write = debugfs_htab_write,
2082 .llseek = generic_file_llseek,
2085 void kvmppc_mmu_debugfs_init(struct kvm *kvm)
2087 kvm->arch.htab_dentry = debugfs_create_file("htab", 0400,
2088 kvm->arch.debugfs_dir, kvm,
2089 &debugfs_htab_fops);
2092 void kvmppc_mmu_book3s_hv_init(struct kvm_vcpu *vcpu)
2094 struct kvmppc_mmu *mmu = &vcpu->arch.mmu;
2096 vcpu->arch.slb_nr = 32; /* POWER7/POWER8 */
2098 if (kvm_is_radix(vcpu->kvm))
2099 mmu->xlate = kvmppc_mmu_radix_xlate;
2101 mmu->xlate = kvmppc_mmu_book3s_64_hv_xlate;
2102 mmu->reset_msr = kvmppc_mmu_book3s_64_hv_reset_msr;
2104 vcpu->arch.hflags |= BOOK3S_HFLAG_SLB;