Back to home page

OSCL-LXR
 
 

     

0001 // SPDX-License-Identifier: GPL-2.0-only
0002 /*
0003  *
0004  * Copyright 2010 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
0005  */
0006 
0007 #include <linux/types.h>
0008 #include <linux/string.h>
0009 #include <linux/kvm.h>
0010 #include <linux/kvm_host.h>
0011 #include <linux/highmem.h>
0012 #include <linux/gfp.h>
0013 #include <linux/slab.h>
0014 #include <linux/hugetlb.h>
0015 #include <linux/vmalloc.h>
0016 #include <linux/srcu.h>
0017 #include <linux/anon_inodes.h>
0018 #include <linux/file.h>
0019 #include <linux/debugfs.h>
0020 
0021 #include <asm/kvm_ppc.h>
0022 #include <asm/kvm_book3s.h>
0023 #include <asm/book3s/64/mmu-hash.h>
0024 #include <asm/hvcall.h>
0025 #include <asm/synch.h>
0026 #include <asm/ppc-opcode.h>
0027 #include <asm/cputable.h>
0028 #include <asm/pte-walk.h>
0029 
0030 #include "book3s.h"
0031 #include "trace_hv.h"
0032 
0033 //#define DEBUG_RESIZE_HPT  1
0034 
0035 #ifdef DEBUG_RESIZE_HPT
0036 #define resize_hpt_debug(resize, ...)               \
0037     do {                            \
0038         printk(KERN_DEBUG "RESIZE HPT %p: ", resize);   \
0039         printk(__VA_ARGS__);                \
0040     } while (0)
0041 #else
0042 #define resize_hpt_debug(resize, ...)               \
0043     do { } while (0)
0044 #endif
0045 
0046 static long kvmppc_virtmode_do_h_enter(struct kvm *kvm, unsigned long flags,
0047                 long pte_index, unsigned long pteh,
0048                 unsigned long ptel, unsigned long *pte_idx_ret);
0049 
0050 struct kvm_resize_hpt {
0051     /* These fields read-only after init */
0052     struct kvm *kvm;
0053     struct work_struct work;
0054     u32 order;
0055 
0056     /* These fields protected by kvm->arch.mmu_setup_lock */
0057 
0058     /* Possible values and their usage:
0059      *  <0     an error occurred during allocation,
0060      *  -EBUSY allocation is in the progress,
0061      *  0      allocation made successfully.
0062      */
0063     int error;
0064 
0065     /* Private to the work thread, until error != -EBUSY,
0066      * then protected by kvm->arch.mmu_setup_lock.
0067      */
0068     struct kvm_hpt_info hpt;
0069 };
0070 
0071 int kvmppc_allocate_hpt(struct kvm_hpt_info *info, u32 order)
0072 {
0073     unsigned long hpt = 0;
0074     int cma = 0;
0075     struct page *page = NULL;
0076     struct revmap_entry *rev;
0077     unsigned long npte;
0078 
0079     if ((order < PPC_MIN_HPT_ORDER) || (order > PPC_MAX_HPT_ORDER))
0080         return -EINVAL;
0081 
0082     page = kvm_alloc_hpt_cma(1ul << (order - PAGE_SHIFT));
0083     if (page) {
0084         hpt = (unsigned long)pfn_to_kaddr(page_to_pfn(page));
0085         memset((void *)hpt, 0, (1ul << order));
0086         cma = 1;
0087     }
0088 
0089     if (!hpt)
0090         hpt = __get_free_pages(GFP_KERNEL|__GFP_ZERO|__GFP_RETRY_MAYFAIL
0091                        |__GFP_NOWARN, order - PAGE_SHIFT);
0092 
0093     if (!hpt)
0094         return -ENOMEM;
0095 
0096     /* HPTEs are 2**4 bytes long */
0097     npte = 1ul << (order - 4);
0098 
0099     /* Allocate reverse map array */
0100     rev = vmalloc(array_size(npte, sizeof(struct revmap_entry)));
0101     if (!rev) {
0102         if (cma)
0103             kvm_free_hpt_cma(page, 1 << (order - PAGE_SHIFT));
0104         else
0105             free_pages(hpt, order - PAGE_SHIFT);
0106         return -ENOMEM;
0107     }
0108 
0109     info->order = order;
0110     info->virt = hpt;
0111     info->cma = cma;
0112     info->rev = rev;
0113 
0114     return 0;
0115 }
0116 
0117 void kvmppc_set_hpt(struct kvm *kvm, struct kvm_hpt_info *info)
0118 {
0119     atomic64_set(&kvm->arch.mmio_update, 0);
0120     kvm->arch.hpt = *info;
0121     kvm->arch.sdr1 = __pa(info->virt) | (info->order - 18);
0122 
0123     pr_debug("KVM guest htab at %lx (order %ld), LPID %x\n",
0124          info->virt, (long)info->order, kvm->arch.lpid);
0125 }
0126 
0127 long kvmppc_alloc_reset_hpt(struct kvm *kvm, int order)
0128 {
0129     long err = -EBUSY;
0130     struct kvm_hpt_info info;
0131 
0132     mutex_lock(&kvm->arch.mmu_setup_lock);
0133     if (kvm->arch.mmu_ready) {
0134         kvm->arch.mmu_ready = 0;
0135         /* order mmu_ready vs. vcpus_running */
0136         smp_mb();
0137         if (atomic_read(&kvm->arch.vcpus_running)) {
0138             kvm->arch.mmu_ready = 1;
0139             goto out;
0140         }
0141     }
0142     if (kvm_is_radix(kvm)) {
0143         err = kvmppc_switch_mmu_to_hpt(kvm);
0144         if (err)
0145             goto out;
0146     }
0147 
0148     if (kvm->arch.hpt.order == order) {
0149         /* We already have a suitable HPT */
0150 
0151         /* Set the entire HPT to 0, i.e. invalid HPTEs */
0152         memset((void *)kvm->arch.hpt.virt, 0, 1ul << order);
0153         /*
0154          * Reset all the reverse-mapping chains for all memslots
0155          */
0156         kvmppc_rmap_reset(kvm);
0157         err = 0;
0158         goto out;
0159     }
0160 
0161     if (kvm->arch.hpt.virt) {
0162         kvmppc_free_hpt(&kvm->arch.hpt);
0163         kvmppc_rmap_reset(kvm);
0164     }
0165 
0166     err = kvmppc_allocate_hpt(&info, order);
0167     if (err < 0)
0168         goto out;
0169     kvmppc_set_hpt(kvm, &info);
0170 
0171 out:
0172     if (err == 0)
0173         /* Ensure that each vcpu will flush its TLB on next entry. */
0174         cpumask_setall(&kvm->arch.need_tlb_flush);
0175 
0176     mutex_unlock(&kvm->arch.mmu_setup_lock);
0177     return err;
0178 }
0179 
0180 void kvmppc_free_hpt(struct kvm_hpt_info *info)
0181 {
0182     vfree(info->rev);
0183     info->rev = NULL;
0184     if (info->cma)
0185         kvm_free_hpt_cma(virt_to_page(info->virt),
0186                  1 << (info->order - PAGE_SHIFT));
0187     else if (info->virt)
0188         free_pages(info->virt, info->order - PAGE_SHIFT);
0189     info->virt = 0;
0190     info->order = 0;
0191 }
0192 
0193 /* Bits in first HPTE dword for pagesize 4k, 64k or 16M */
0194 static inline unsigned long hpte0_pgsize_encoding(unsigned long pgsize)
0195 {
0196     return (pgsize > 0x1000) ? HPTE_V_LARGE : 0;
0197 }
0198 
0199 /* Bits in second HPTE dword for pagesize 4k, 64k or 16M */
0200 static inline unsigned long hpte1_pgsize_encoding(unsigned long pgsize)
0201 {
0202     return (pgsize == 0x10000) ? 0x1000 : 0;
0203 }
0204 
0205 void kvmppc_map_vrma(struct kvm_vcpu *vcpu, struct kvm_memory_slot *memslot,
0206              unsigned long porder)
0207 {
0208     unsigned long i;
0209     unsigned long npages;
0210     unsigned long hp_v, hp_r;
0211     unsigned long addr, hash;
0212     unsigned long psize;
0213     unsigned long hp0, hp1;
0214     unsigned long idx_ret;
0215     long ret;
0216     struct kvm *kvm = vcpu->kvm;
0217 
0218     psize = 1ul << porder;
0219     npages = memslot->npages >> (porder - PAGE_SHIFT);
0220 
0221     /* VRMA can't be > 1TB */
0222     if (npages > 1ul << (40 - porder))
0223         npages = 1ul << (40 - porder);
0224     /* Can't use more than 1 HPTE per HPTEG */
0225     if (npages > kvmppc_hpt_mask(&kvm->arch.hpt) + 1)
0226         npages = kvmppc_hpt_mask(&kvm->arch.hpt) + 1;
0227 
0228     hp0 = HPTE_V_1TB_SEG | (VRMA_VSID << (40 - 16)) |
0229         HPTE_V_BOLTED | hpte0_pgsize_encoding(psize);
0230     hp1 = hpte1_pgsize_encoding(psize) |
0231         HPTE_R_R | HPTE_R_C | HPTE_R_M | PP_RWXX;
0232 
0233     for (i = 0; i < npages; ++i) {
0234         addr = i << porder;
0235         /* can't use hpt_hash since va > 64 bits */
0236         hash = (i ^ (VRMA_VSID ^ (VRMA_VSID << 25)))
0237             & kvmppc_hpt_mask(&kvm->arch.hpt);
0238         /*
0239          * We assume that the hash table is empty and no
0240          * vcpus are using it at this stage.  Since we create
0241          * at most one HPTE per HPTEG, we just assume entry 7
0242          * is available and use it.
0243          */
0244         hash = (hash << 3) + 7;
0245         hp_v = hp0 | ((addr >> 16) & ~0x7fUL);
0246         hp_r = hp1 | addr;
0247         ret = kvmppc_virtmode_do_h_enter(kvm, H_EXACT, hash, hp_v, hp_r,
0248                          &idx_ret);
0249         if (ret != H_SUCCESS) {
0250             pr_err("KVM: map_vrma at %lx failed, ret=%ld\n",
0251                    addr, ret);
0252             break;
0253         }
0254     }
0255 }
0256 
0257 int kvmppc_mmu_hv_init(void)
0258 {
0259     unsigned long nr_lpids;
0260 
0261     if (!mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE))
0262         return -EINVAL;
0263 
0264     if (cpu_has_feature(CPU_FTR_HVMODE)) {
0265         if (WARN_ON(mfspr(SPRN_LPID) != 0))
0266             return -EINVAL;
0267         nr_lpids = 1UL << mmu_lpid_bits;
0268     } else {
0269         nr_lpids = 1UL << KVM_MAX_NESTED_GUESTS_SHIFT;
0270     }
0271 
0272     if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
0273         /* POWER7 has 10-bit LPIDs, POWER8 has 12-bit LPIDs */
0274         if (cpu_has_feature(CPU_FTR_ARCH_207S))
0275             WARN_ON(nr_lpids != 1UL << 12);
0276         else
0277             WARN_ON(nr_lpids != 1UL << 10);
0278 
0279         /*
0280          * Reserve the last implemented LPID use in partition
0281          * switching for POWER7 and POWER8.
0282          */
0283         nr_lpids -= 1;
0284     }
0285 
0286     kvmppc_init_lpid(nr_lpids);
0287 
0288     return 0;
0289 }
0290 
0291 static long kvmppc_virtmode_do_h_enter(struct kvm *kvm, unsigned long flags,
0292                 long pte_index, unsigned long pteh,
0293                 unsigned long ptel, unsigned long *pte_idx_ret)
0294 {
0295     long ret;
0296 
0297     preempt_disable();
0298     ret = kvmppc_do_h_enter(kvm, flags, pte_index, pteh, ptel,
0299                 kvm->mm->pgd, false, pte_idx_ret);
0300     preempt_enable();
0301     if (ret == H_TOO_HARD) {
0302         /* this can't happen */
0303         pr_err("KVM: Oops, kvmppc_h_enter returned too hard!\n");
0304         ret = H_RESOURCE;   /* or something */
0305     }
0306     return ret;
0307 
0308 }
0309 
0310 static struct kvmppc_slb *kvmppc_mmu_book3s_hv_find_slbe(struct kvm_vcpu *vcpu,
0311                              gva_t eaddr)
0312 {
0313     u64 mask;
0314     int i;
0315 
0316     for (i = 0; i < vcpu->arch.slb_nr; i++) {
0317         if (!(vcpu->arch.slb[i].orige & SLB_ESID_V))
0318             continue;
0319 
0320         if (vcpu->arch.slb[i].origv & SLB_VSID_B_1T)
0321             mask = ESID_MASK_1T;
0322         else
0323             mask = ESID_MASK;
0324 
0325         if (((vcpu->arch.slb[i].orige ^ eaddr) & mask) == 0)
0326             return &vcpu->arch.slb[i];
0327     }
0328     return NULL;
0329 }
0330 
0331 static unsigned long kvmppc_mmu_get_real_addr(unsigned long v, unsigned long r,
0332             unsigned long ea)
0333 {
0334     unsigned long ra_mask;
0335 
0336     ra_mask = kvmppc_actual_pgsz(v, r) - 1;
0337     return (r & HPTE_R_RPN & ~ra_mask) | (ea & ra_mask);
0338 }
0339 
0340 static int kvmppc_mmu_book3s_64_hv_xlate(struct kvm_vcpu *vcpu, gva_t eaddr,
0341             struct kvmppc_pte *gpte, bool data, bool iswrite)
0342 {
0343     struct kvm *kvm = vcpu->kvm;
0344     struct kvmppc_slb *slbe;
0345     unsigned long slb_v;
0346     unsigned long pp, key;
0347     unsigned long v, orig_v, gr;
0348     __be64 *hptep;
0349     long int index;
0350     int virtmode = vcpu->arch.shregs.msr & (data ? MSR_DR : MSR_IR);
0351 
0352     if (kvm_is_radix(vcpu->kvm))
0353         return kvmppc_mmu_radix_xlate(vcpu, eaddr, gpte, data, iswrite);
0354 
0355     /* Get SLB entry */
0356     if (virtmode) {
0357         slbe = kvmppc_mmu_book3s_hv_find_slbe(vcpu, eaddr);
0358         if (!slbe)
0359             return -EINVAL;
0360         slb_v = slbe->origv;
0361     } else {
0362         /* real mode access */
0363         slb_v = vcpu->kvm->arch.vrma_slb_v;
0364     }
0365 
0366     preempt_disable();
0367     /* Find the HPTE in the hash table */
0368     index = kvmppc_hv_find_lock_hpte(kvm, eaddr, slb_v,
0369                      HPTE_V_VALID | HPTE_V_ABSENT);
0370     if (index < 0) {
0371         preempt_enable();
0372         return -ENOENT;
0373     }
0374     hptep = (__be64 *)(kvm->arch.hpt.virt + (index << 4));
0375     v = orig_v = be64_to_cpu(hptep[0]) & ~HPTE_V_HVLOCK;
0376     if (cpu_has_feature(CPU_FTR_ARCH_300))
0377         v = hpte_new_to_old_v(v, be64_to_cpu(hptep[1]));
0378     gr = kvm->arch.hpt.rev[index].guest_rpte;
0379 
0380     unlock_hpte(hptep, orig_v);
0381     preempt_enable();
0382 
0383     gpte->eaddr = eaddr;
0384     gpte->vpage = ((v & HPTE_V_AVPN) << 4) | ((eaddr >> 12) & 0xfff);
0385 
0386     /* Get PP bits and key for permission check */
0387     pp = gr & (HPTE_R_PP0 | HPTE_R_PP);
0388     key = (vcpu->arch.shregs.msr & MSR_PR) ? SLB_VSID_KP : SLB_VSID_KS;
0389     key &= slb_v;
0390 
0391     /* Calculate permissions */
0392     gpte->may_read = hpte_read_permission(pp, key);
0393     gpte->may_write = hpte_write_permission(pp, key);
0394     gpte->may_execute = gpte->may_read && !(gr & (HPTE_R_N | HPTE_R_G));
0395 
0396     /* Storage key permission check for POWER7 */
0397     if (data && virtmode) {
0398         int amrfield = hpte_get_skey_perm(gr, vcpu->arch.amr);
0399         if (amrfield & 1)
0400             gpte->may_read = 0;
0401         if (amrfield & 2)
0402             gpte->may_write = 0;
0403     }
0404 
0405     /* Get the guest physical address */
0406     gpte->raddr = kvmppc_mmu_get_real_addr(v, gr, eaddr);
0407     return 0;
0408 }
0409 
0410 /*
0411  * Quick test for whether an instruction is a load or a store.
0412  * If the instruction is a load or a store, then this will indicate
0413  * which it is, at least on server processors.  (Embedded processors
0414  * have some external PID instructions that don't follow the rule
0415  * embodied here.)  If the instruction isn't a load or store, then
0416  * this doesn't return anything useful.
0417  */
0418 static int instruction_is_store(unsigned int instr)
0419 {
0420     unsigned int mask;
0421 
0422     mask = 0x10000000;
0423     if ((instr & 0xfc000000) == 0x7c000000)
0424         mask = 0x100;       /* major opcode 31 */
0425     return (instr & mask) != 0;
0426 }
0427 
0428 int kvmppc_hv_emulate_mmio(struct kvm_vcpu *vcpu,
0429                unsigned long gpa, gva_t ea, int is_store)
0430 {
0431     u32 last_inst;
0432 
0433     /*
0434      * Fast path - check if the guest physical address corresponds to a
0435      * device on the FAST_MMIO_BUS, if so we can avoid loading the
0436      * instruction all together, then we can just handle it and return.
0437      */
0438     if (is_store) {
0439         int idx, ret;
0440 
0441         idx = srcu_read_lock(&vcpu->kvm->srcu);
0442         ret = kvm_io_bus_write(vcpu, KVM_FAST_MMIO_BUS, (gpa_t) gpa, 0,
0443                        NULL);
0444         srcu_read_unlock(&vcpu->kvm->srcu, idx);
0445         if (!ret) {
0446             kvmppc_set_pc(vcpu, kvmppc_get_pc(vcpu) + 4);
0447             return RESUME_GUEST;
0448         }
0449     }
0450 
0451     /*
0452      * If we fail, we just return to the guest and try executing it again.
0453      */
0454     if (kvmppc_get_last_inst(vcpu, INST_GENERIC, &last_inst) !=
0455         EMULATE_DONE)
0456         return RESUME_GUEST;
0457 
0458     /*
0459      * WARNING: We do not know for sure whether the instruction we just
0460      * read from memory is the same that caused the fault in the first
0461      * place.  If the instruction we read is neither an load or a store,
0462      * then it can't access memory, so we don't need to worry about
0463      * enforcing access permissions.  So, assuming it is a load or
0464      * store, we just check that its direction (load or store) is
0465      * consistent with the original fault, since that's what we
0466      * checked the access permissions against.  If there is a mismatch
0467      * we just return and retry the instruction.
0468      */
0469 
0470     if (instruction_is_store(last_inst) != !!is_store)
0471         return RESUME_GUEST;
0472 
0473     /*
0474      * Emulated accesses are emulated by looking at the hash for
0475      * translation once, then performing the access later. The
0476      * translation could be invalidated in the meantime in which
0477      * point performing the subsequent memory access on the old
0478      * physical address could possibly be a security hole for the
0479      * guest (but not the host).
0480      *
0481      * This is less of an issue for MMIO stores since they aren't
0482      * globally visible. It could be an issue for MMIO loads to
0483      * a certain extent but we'll ignore it for now.
0484      */
0485 
0486     vcpu->arch.paddr_accessed = gpa;
0487     vcpu->arch.vaddr_accessed = ea;
0488     return kvmppc_emulate_mmio(vcpu);
0489 }
0490 
0491 int kvmppc_book3s_hv_page_fault(struct kvm_vcpu *vcpu,
0492                 unsigned long ea, unsigned long dsisr)
0493 {
0494     struct kvm *kvm = vcpu->kvm;
0495     unsigned long hpte[3], r;
0496     unsigned long hnow_v, hnow_r;
0497     __be64 *hptep;
0498     unsigned long mmu_seq, psize, pte_size;
0499     unsigned long gpa_base, gfn_base;
0500     unsigned long gpa, gfn, hva, pfn, hpa;
0501     struct kvm_memory_slot *memslot;
0502     unsigned long *rmap;
0503     struct revmap_entry *rev;
0504     struct page *page;
0505     long index, ret;
0506     bool is_ci;
0507     bool writing, write_ok;
0508     unsigned int shift;
0509     unsigned long rcbits;
0510     long mmio_update;
0511     pte_t pte, *ptep;
0512 
0513     if (kvm_is_radix(kvm))
0514         return kvmppc_book3s_radix_page_fault(vcpu, ea, dsisr);
0515 
0516     /*
0517      * Real-mode code has already searched the HPT and found the
0518      * entry we're interested in.  Lock the entry and check that
0519      * it hasn't changed.  If it has, just return and re-execute the
0520      * instruction.
0521      */
0522     if (ea != vcpu->arch.pgfault_addr)
0523         return RESUME_GUEST;
0524 
0525     if (vcpu->arch.pgfault_cache) {
0526         mmio_update = atomic64_read(&kvm->arch.mmio_update);
0527         if (mmio_update == vcpu->arch.pgfault_cache->mmio_update) {
0528             r = vcpu->arch.pgfault_cache->rpte;
0529             psize = kvmppc_actual_pgsz(vcpu->arch.pgfault_hpte[0],
0530                            r);
0531             gpa_base = r & HPTE_R_RPN & ~(psize - 1);
0532             gfn_base = gpa_base >> PAGE_SHIFT;
0533             gpa = gpa_base | (ea & (psize - 1));
0534             return kvmppc_hv_emulate_mmio(vcpu, gpa, ea,
0535                         dsisr & DSISR_ISSTORE);
0536         }
0537     }
0538     index = vcpu->arch.pgfault_index;
0539     hptep = (__be64 *)(kvm->arch.hpt.virt + (index << 4));
0540     rev = &kvm->arch.hpt.rev[index];
0541     preempt_disable();
0542     while (!try_lock_hpte(hptep, HPTE_V_HVLOCK))
0543         cpu_relax();
0544     hpte[0] = be64_to_cpu(hptep[0]) & ~HPTE_V_HVLOCK;
0545     hpte[1] = be64_to_cpu(hptep[1]);
0546     hpte[2] = r = rev->guest_rpte;
0547     unlock_hpte(hptep, hpte[0]);
0548     preempt_enable();
0549 
0550     if (cpu_has_feature(CPU_FTR_ARCH_300)) {
0551         hpte[0] = hpte_new_to_old_v(hpte[0], hpte[1]);
0552         hpte[1] = hpte_new_to_old_r(hpte[1]);
0553     }
0554     if (hpte[0] != vcpu->arch.pgfault_hpte[0] ||
0555         hpte[1] != vcpu->arch.pgfault_hpte[1])
0556         return RESUME_GUEST;
0557 
0558     /* Translate the logical address and get the page */
0559     psize = kvmppc_actual_pgsz(hpte[0], r);
0560     gpa_base = r & HPTE_R_RPN & ~(psize - 1);
0561     gfn_base = gpa_base >> PAGE_SHIFT;
0562     gpa = gpa_base | (ea & (psize - 1));
0563     gfn = gpa >> PAGE_SHIFT;
0564     memslot = gfn_to_memslot(kvm, gfn);
0565 
0566     trace_kvm_page_fault_enter(vcpu, hpte, memslot, ea, dsisr);
0567 
0568     /* No memslot means it's an emulated MMIO region */
0569     if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
0570         return kvmppc_hv_emulate_mmio(vcpu, gpa, ea,
0571                           dsisr & DSISR_ISSTORE);
0572 
0573     /*
0574      * This should never happen, because of the slot_is_aligned()
0575      * check in kvmppc_do_h_enter().
0576      */
0577     if (gfn_base < memslot->base_gfn)
0578         return -EFAULT;
0579 
0580     /* used to check for invalidations in progress */
0581     mmu_seq = kvm->mmu_invalidate_seq;
0582     smp_rmb();
0583 
0584     ret = -EFAULT;
0585     page = NULL;
0586     writing = (dsisr & DSISR_ISSTORE) != 0;
0587     /* If writing != 0, then the HPTE must allow writing, if we get here */
0588     write_ok = writing;
0589     hva = gfn_to_hva_memslot(memslot, gfn);
0590 
0591     /*
0592      * Do a fast check first, since __gfn_to_pfn_memslot doesn't
0593      * do it with !atomic && !async, which is how we call it.
0594      * We always ask for write permission since the common case
0595      * is that the page is writable.
0596      */
0597     if (get_user_page_fast_only(hva, FOLL_WRITE, &page)) {
0598         write_ok = true;
0599     } else {
0600         /* Call KVM generic code to do the slow-path check */
0601         pfn = __gfn_to_pfn_memslot(memslot, gfn, false, NULL,
0602                        writing, &write_ok, NULL);
0603         if (is_error_noslot_pfn(pfn))
0604             return -EFAULT;
0605         page = NULL;
0606         if (pfn_valid(pfn)) {
0607             page = pfn_to_page(pfn);
0608             if (PageReserved(page))
0609                 page = NULL;
0610         }
0611     }
0612 
0613     /*
0614      * Read the PTE from the process' radix tree and use that
0615      * so we get the shift and attribute bits.
0616      */
0617     spin_lock(&kvm->mmu_lock);
0618     ptep = find_kvm_host_pte(kvm, mmu_seq, hva, &shift);
0619     pte = __pte(0);
0620     if (ptep)
0621         pte = READ_ONCE(*ptep);
0622     spin_unlock(&kvm->mmu_lock);
0623     /*
0624      * If the PTE disappeared temporarily due to a THP
0625      * collapse, just return and let the guest try again.
0626      */
0627     if (!pte_present(pte)) {
0628         if (page)
0629             put_page(page);
0630         return RESUME_GUEST;
0631     }
0632     hpa = pte_pfn(pte) << PAGE_SHIFT;
0633     pte_size = PAGE_SIZE;
0634     if (shift)
0635         pte_size = 1ul << shift;
0636     is_ci = pte_ci(pte);
0637 
0638     if (psize > pte_size)
0639         goto out_put;
0640     if (pte_size > psize)
0641         hpa |= hva & (pte_size - psize);
0642 
0643     /* Check WIMG vs. the actual page we're accessing */
0644     if (!hpte_cache_flags_ok(r, is_ci)) {
0645         if (is_ci)
0646             goto out_put;
0647         /*
0648          * Allow guest to map emulated device memory as
0649          * uncacheable, but actually make it cacheable.
0650          */
0651         r = (r & ~(HPTE_R_W|HPTE_R_I|HPTE_R_G)) | HPTE_R_M;
0652     }
0653 
0654     /*
0655      * Set the HPTE to point to hpa.
0656      * Since the hpa is at PAGE_SIZE granularity, make sure we
0657      * don't mask out lower-order bits if psize < PAGE_SIZE.
0658      */
0659     if (psize < PAGE_SIZE)
0660         psize = PAGE_SIZE;
0661     r = (r & HPTE_R_KEY_HI) | (r & ~(HPTE_R_PP0 - psize)) | hpa;
0662     if (hpte_is_writable(r) && !write_ok)
0663         r = hpte_make_readonly(r);
0664     ret = RESUME_GUEST;
0665     preempt_disable();
0666     while (!try_lock_hpte(hptep, HPTE_V_HVLOCK))
0667         cpu_relax();
0668     hnow_v = be64_to_cpu(hptep[0]);
0669     hnow_r = be64_to_cpu(hptep[1]);
0670     if (cpu_has_feature(CPU_FTR_ARCH_300)) {
0671         hnow_v = hpte_new_to_old_v(hnow_v, hnow_r);
0672         hnow_r = hpte_new_to_old_r(hnow_r);
0673     }
0674 
0675     /*
0676      * If the HPT is being resized, don't update the HPTE,
0677      * instead let the guest retry after the resize operation is complete.
0678      * The synchronization for mmu_ready test vs. set is provided
0679      * by the HPTE lock.
0680      */
0681     if (!kvm->arch.mmu_ready)
0682         goto out_unlock;
0683 
0684     if ((hnow_v & ~HPTE_V_HVLOCK) != hpte[0] || hnow_r != hpte[1] ||
0685         rev->guest_rpte != hpte[2])
0686         /* HPTE has been changed under us; let the guest retry */
0687         goto out_unlock;
0688     hpte[0] = (hpte[0] & ~HPTE_V_ABSENT) | HPTE_V_VALID;
0689 
0690     /* Always put the HPTE in the rmap chain for the page base address */
0691     rmap = &memslot->arch.rmap[gfn_base - memslot->base_gfn];
0692     lock_rmap(rmap);
0693 
0694     /* Check if we might have been invalidated; let the guest retry if so */
0695     ret = RESUME_GUEST;
0696     if (mmu_invalidate_retry(vcpu->kvm, mmu_seq)) {
0697         unlock_rmap(rmap);
0698         goto out_unlock;
0699     }
0700 
0701     /* Only set R/C in real HPTE if set in both *rmap and guest_rpte */
0702     rcbits = *rmap >> KVMPPC_RMAP_RC_SHIFT;
0703     r &= rcbits | ~(HPTE_R_R | HPTE_R_C);
0704 
0705     if (be64_to_cpu(hptep[0]) & HPTE_V_VALID) {
0706         /* HPTE was previously valid, so we need to invalidate it */
0707         unlock_rmap(rmap);
0708         hptep[0] |= cpu_to_be64(HPTE_V_ABSENT);
0709         kvmppc_invalidate_hpte(kvm, hptep, index);
0710         /* don't lose previous R and C bits */
0711         r |= be64_to_cpu(hptep[1]) & (HPTE_R_R | HPTE_R_C);
0712     } else {
0713         kvmppc_add_revmap_chain(kvm, rev, rmap, index, 0);
0714     }
0715 
0716     if (cpu_has_feature(CPU_FTR_ARCH_300)) {
0717         r = hpte_old_to_new_r(hpte[0], r);
0718         hpte[0] = hpte_old_to_new_v(hpte[0]);
0719     }
0720     hptep[1] = cpu_to_be64(r);
0721     eieio();
0722     __unlock_hpte(hptep, hpte[0]);
0723     asm volatile("ptesync" : : : "memory");
0724     preempt_enable();
0725     if (page && hpte_is_writable(r))
0726         set_page_dirty_lock(page);
0727 
0728  out_put:
0729     trace_kvm_page_fault_exit(vcpu, hpte, ret);
0730 
0731     if (page)
0732         put_page(page);
0733     return ret;
0734 
0735  out_unlock:
0736     __unlock_hpte(hptep, be64_to_cpu(hptep[0]));
0737     preempt_enable();
0738     goto out_put;
0739 }
0740 
0741 void kvmppc_rmap_reset(struct kvm *kvm)
0742 {
0743     struct kvm_memslots *slots;
0744     struct kvm_memory_slot *memslot;
0745     int srcu_idx, bkt;
0746 
0747     srcu_idx = srcu_read_lock(&kvm->srcu);
0748     slots = kvm_memslots(kvm);
0749     kvm_for_each_memslot(memslot, bkt, slots) {
0750         /* Mutual exclusion with kvm_unmap_hva_range etc. */
0751         spin_lock(&kvm->mmu_lock);
0752         /*
0753          * This assumes it is acceptable to lose reference and
0754          * change bits across a reset.
0755          */
0756         memset(memslot->arch.rmap, 0,
0757                memslot->npages * sizeof(*memslot->arch.rmap));
0758         spin_unlock(&kvm->mmu_lock);
0759     }
0760     srcu_read_unlock(&kvm->srcu, srcu_idx);
0761 }
0762 
0763 /* Must be called with both HPTE and rmap locked */
0764 static void kvmppc_unmap_hpte(struct kvm *kvm, unsigned long i,
0765                   struct kvm_memory_slot *memslot,
0766                   unsigned long *rmapp, unsigned long gfn)
0767 {
0768     __be64 *hptep = (__be64 *) (kvm->arch.hpt.virt + (i << 4));
0769     struct revmap_entry *rev = kvm->arch.hpt.rev;
0770     unsigned long j, h;
0771     unsigned long ptel, psize, rcbits;
0772 
0773     j = rev[i].forw;
0774     if (j == i) {
0775         /* chain is now empty */
0776         *rmapp &= ~(KVMPPC_RMAP_PRESENT | KVMPPC_RMAP_INDEX);
0777     } else {
0778         /* remove i from chain */
0779         h = rev[i].back;
0780         rev[h].forw = j;
0781         rev[j].back = h;
0782         rev[i].forw = rev[i].back = i;
0783         *rmapp = (*rmapp & ~KVMPPC_RMAP_INDEX) | j;
0784     }
0785 
0786     /* Now check and modify the HPTE */
0787     ptel = rev[i].guest_rpte;
0788     psize = kvmppc_actual_pgsz(be64_to_cpu(hptep[0]), ptel);
0789     if ((be64_to_cpu(hptep[0]) & HPTE_V_VALID) &&
0790         hpte_rpn(ptel, psize) == gfn) {
0791         hptep[0] |= cpu_to_be64(HPTE_V_ABSENT);
0792         kvmppc_invalidate_hpte(kvm, hptep, i);
0793         hptep[1] &= ~cpu_to_be64(HPTE_R_KEY_HI | HPTE_R_KEY_LO);
0794         /* Harvest R and C */
0795         rcbits = be64_to_cpu(hptep[1]) & (HPTE_R_R | HPTE_R_C);
0796         *rmapp |= rcbits << KVMPPC_RMAP_RC_SHIFT;
0797         if ((rcbits & HPTE_R_C) && memslot->dirty_bitmap)
0798             kvmppc_update_dirty_map(memslot, gfn, psize);
0799         if (rcbits & ~rev[i].guest_rpte) {
0800             rev[i].guest_rpte = ptel | rcbits;
0801             note_hpte_modification(kvm, &rev[i]);
0802         }
0803     }
0804 }
0805 
0806 static void kvm_unmap_rmapp(struct kvm *kvm, struct kvm_memory_slot *memslot,
0807                 unsigned long gfn)
0808 {
0809     unsigned long i;
0810     __be64 *hptep;
0811     unsigned long *rmapp;
0812 
0813     rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn];
0814     for (;;) {
0815         lock_rmap(rmapp);
0816         if (!(*rmapp & KVMPPC_RMAP_PRESENT)) {
0817             unlock_rmap(rmapp);
0818             break;
0819         }
0820 
0821         /*
0822          * To avoid an ABBA deadlock with the HPTE lock bit,
0823          * we can't spin on the HPTE lock while holding the
0824          * rmap chain lock.
0825          */
0826         i = *rmapp & KVMPPC_RMAP_INDEX;
0827         hptep = (__be64 *) (kvm->arch.hpt.virt + (i << 4));
0828         if (!try_lock_hpte(hptep, HPTE_V_HVLOCK)) {
0829             /* unlock rmap before spinning on the HPTE lock */
0830             unlock_rmap(rmapp);
0831             while (be64_to_cpu(hptep[0]) & HPTE_V_HVLOCK)
0832                 cpu_relax();
0833             continue;
0834         }
0835 
0836         kvmppc_unmap_hpte(kvm, i, memslot, rmapp, gfn);
0837         unlock_rmap(rmapp);
0838         __unlock_hpte(hptep, be64_to_cpu(hptep[0]));
0839     }
0840 }
0841 
0842 bool kvm_unmap_gfn_range_hv(struct kvm *kvm, struct kvm_gfn_range *range)
0843 {
0844     gfn_t gfn;
0845 
0846     if (kvm_is_radix(kvm)) {
0847         for (gfn = range->start; gfn < range->end; gfn++)
0848             kvm_unmap_radix(kvm, range->slot, gfn);
0849     } else {
0850         for (gfn = range->start; gfn < range->end; gfn++)
0851             kvm_unmap_rmapp(kvm, range->slot, gfn);
0852     }
0853 
0854     return false;
0855 }
0856 
0857 void kvmppc_core_flush_memslot_hv(struct kvm *kvm,
0858                   struct kvm_memory_slot *memslot)
0859 {
0860     unsigned long gfn;
0861     unsigned long n;
0862     unsigned long *rmapp;
0863 
0864     gfn = memslot->base_gfn;
0865     rmapp = memslot->arch.rmap;
0866     if (kvm_is_radix(kvm)) {
0867         kvmppc_radix_flush_memslot(kvm, memslot);
0868         return;
0869     }
0870 
0871     for (n = memslot->npages; n; --n, ++gfn) {
0872         /*
0873          * Testing the present bit without locking is OK because
0874          * the memslot has been marked invalid already, and hence
0875          * no new HPTEs referencing this page can be created,
0876          * thus the present bit can't go from 0 to 1.
0877          */
0878         if (*rmapp & KVMPPC_RMAP_PRESENT)
0879             kvm_unmap_rmapp(kvm, memslot, gfn);
0880         ++rmapp;
0881     }
0882 }
0883 
0884 static bool kvm_age_rmapp(struct kvm *kvm, struct kvm_memory_slot *memslot,
0885               unsigned long gfn)
0886 {
0887     struct revmap_entry *rev = kvm->arch.hpt.rev;
0888     unsigned long head, i, j;
0889     __be64 *hptep;
0890     bool ret = false;
0891     unsigned long *rmapp;
0892 
0893     rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn];
0894  retry:
0895     lock_rmap(rmapp);
0896     if (*rmapp & KVMPPC_RMAP_REFERENCED) {
0897         *rmapp &= ~KVMPPC_RMAP_REFERENCED;
0898         ret = true;
0899     }
0900     if (!(*rmapp & KVMPPC_RMAP_PRESENT)) {
0901         unlock_rmap(rmapp);
0902         return ret;
0903     }
0904 
0905     i = head = *rmapp & KVMPPC_RMAP_INDEX;
0906     do {
0907         hptep = (__be64 *) (kvm->arch.hpt.virt + (i << 4));
0908         j = rev[i].forw;
0909 
0910         /* If this HPTE isn't referenced, ignore it */
0911         if (!(be64_to_cpu(hptep[1]) & HPTE_R_R))
0912             continue;
0913 
0914         if (!try_lock_hpte(hptep, HPTE_V_HVLOCK)) {
0915             /* unlock rmap before spinning on the HPTE lock */
0916             unlock_rmap(rmapp);
0917             while (be64_to_cpu(hptep[0]) & HPTE_V_HVLOCK)
0918                 cpu_relax();
0919             goto retry;
0920         }
0921 
0922         /* Now check and modify the HPTE */
0923         if ((be64_to_cpu(hptep[0]) & HPTE_V_VALID) &&
0924             (be64_to_cpu(hptep[1]) & HPTE_R_R)) {
0925             kvmppc_clear_ref_hpte(kvm, hptep, i);
0926             if (!(rev[i].guest_rpte & HPTE_R_R)) {
0927                 rev[i].guest_rpte |= HPTE_R_R;
0928                 note_hpte_modification(kvm, &rev[i]);
0929             }
0930             ret = true;
0931         }
0932         __unlock_hpte(hptep, be64_to_cpu(hptep[0]));
0933     } while ((i = j) != head);
0934 
0935     unlock_rmap(rmapp);
0936     return ret;
0937 }
0938 
0939 bool kvm_age_gfn_hv(struct kvm *kvm, struct kvm_gfn_range *range)
0940 {
0941     gfn_t gfn;
0942     bool ret = false;
0943 
0944     if (kvm_is_radix(kvm)) {
0945         for (gfn = range->start; gfn < range->end; gfn++)
0946             ret |= kvm_age_radix(kvm, range->slot, gfn);
0947     } else {
0948         for (gfn = range->start; gfn < range->end; gfn++)
0949             ret |= kvm_age_rmapp(kvm, range->slot, gfn);
0950     }
0951 
0952     return ret;
0953 }
0954 
0955 static bool kvm_test_age_rmapp(struct kvm *kvm, struct kvm_memory_slot *memslot,
0956                    unsigned long gfn)
0957 {
0958     struct revmap_entry *rev = kvm->arch.hpt.rev;
0959     unsigned long head, i, j;
0960     unsigned long *hp;
0961     bool ret = true;
0962     unsigned long *rmapp;
0963 
0964     rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn];
0965     if (*rmapp & KVMPPC_RMAP_REFERENCED)
0966         return true;
0967 
0968     lock_rmap(rmapp);
0969     if (*rmapp & KVMPPC_RMAP_REFERENCED)
0970         goto out;
0971 
0972     if (*rmapp & KVMPPC_RMAP_PRESENT) {
0973         i = head = *rmapp & KVMPPC_RMAP_INDEX;
0974         do {
0975             hp = (unsigned long *)(kvm->arch.hpt.virt + (i << 4));
0976             j = rev[i].forw;
0977             if (be64_to_cpu(hp[1]) & HPTE_R_R)
0978                 goto out;
0979         } while ((i = j) != head);
0980     }
0981     ret = false;
0982 
0983  out:
0984     unlock_rmap(rmapp);
0985     return ret;
0986 }
0987 
0988 bool kvm_test_age_gfn_hv(struct kvm *kvm, struct kvm_gfn_range *range)
0989 {
0990     WARN_ON(range->start + 1 != range->end);
0991 
0992     if (kvm_is_radix(kvm))
0993         return kvm_test_age_radix(kvm, range->slot, range->start);
0994     else
0995         return kvm_test_age_rmapp(kvm, range->slot, range->start);
0996 }
0997 
0998 bool kvm_set_spte_gfn_hv(struct kvm *kvm, struct kvm_gfn_range *range)
0999 {
1000     WARN_ON(range->start + 1 != range->end);
1001 
1002     if (kvm_is_radix(kvm))
1003         kvm_unmap_radix(kvm, range->slot, range->start);
1004     else
1005         kvm_unmap_rmapp(kvm, range->slot, range->start);
1006 
1007     return false;
1008 }
1009 
1010 static int vcpus_running(struct kvm *kvm)
1011 {
1012     return atomic_read(&kvm->arch.vcpus_running) != 0;
1013 }
1014 
1015 /*
1016  * Returns the number of system pages that are dirty.
1017  * This can be more than 1 if we find a huge-page HPTE.
1018  */
1019 static int kvm_test_clear_dirty_npages(struct kvm *kvm, unsigned long *rmapp)
1020 {
1021     struct revmap_entry *rev = kvm->arch.hpt.rev;
1022     unsigned long head, i, j;
1023     unsigned long n;
1024     unsigned long v, r;
1025     __be64 *hptep;
1026     int npages_dirty = 0;
1027 
1028  retry:
1029     lock_rmap(rmapp);
1030     if (!(*rmapp & KVMPPC_RMAP_PRESENT)) {
1031         unlock_rmap(rmapp);
1032         return npages_dirty;
1033     }
1034 
1035     i = head = *rmapp & KVMPPC_RMAP_INDEX;
1036     do {
1037         unsigned long hptep1;
1038         hptep = (__be64 *) (kvm->arch.hpt.virt + (i << 4));
1039         j = rev[i].forw;
1040 
1041         /*
1042          * Checking the C (changed) bit here is racy since there
1043          * is no guarantee about when the hardware writes it back.
1044          * If the HPTE is not writable then it is stable since the
1045          * page can't be written to, and we would have done a tlbie
1046          * (which forces the hardware to complete any writeback)
1047          * when making the HPTE read-only.
1048          * If vcpus are running then this call is racy anyway
1049          * since the page could get dirtied subsequently, so we
1050          * expect there to be a further call which would pick up
1051          * any delayed C bit writeback.
1052          * Otherwise we need to do the tlbie even if C==0 in
1053          * order to pick up any delayed writeback of C.
1054          */
1055         hptep1 = be64_to_cpu(hptep[1]);
1056         if (!(hptep1 & HPTE_R_C) &&
1057             (!hpte_is_writable(hptep1) || vcpus_running(kvm)))
1058             continue;
1059 
1060         if (!try_lock_hpte(hptep, HPTE_V_HVLOCK)) {
1061             /* unlock rmap before spinning on the HPTE lock */
1062             unlock_rmap(rmapp);
1063             while (hptep[0] & cpu_to_be64(HPTE_V_HVLOCK))
1064                 cpu_relax();
1065             goto retry;
1066         }
1067 
1068         /* Now check and modify the HPTE */
1069         if (!(hptep[0] & cpu_to_be64(HPTE_V_VALID))) {
1070             __unlock_hpte(hptep, be64_to_cpu(hptep[0]));
1071             continue;
1072         }
1073 
1074         /* need to make it temporarily absent so C is stable */
1075         hptep[0] |= cpu_to_be64(HPTE_V_ABSENT);
1076         kvmppc_invalidate_hpte(kvm, hptep, i);
1077         v = be64_to_cpu(hptep[0]);
1078         r = be64_to_cpu(hptep[1]);
1079         if (r & HPTE_R_C) {
1080             hptep[1] = cpu_to_be64(r & ~HPTE_R_C);
1081             if (!(rev[i].guest_rpte & HPTE_R_C)) {
1082                 rev[i].guest_rpte |= HPTE_R_C;
1083                 note_hpte_modification(kvm, &rev[i]);
1084             }
1085             n = kvmppc_actual_pgsz(v, r);
1086             n = (n + PAGE_SIZE - 1) >> PAGE_SHIFT;
1087             if (n > npages_dirty)
1088                 npages_dirty = n;
1089             eieio();
1090         }
1091         v &= ~HPTE_V_ABSENT;
1092         v |= HPTE_V_VALID;
1093         __unlock_hpte(hptep, v);
1094     } while ((i = j) != head);
1095 
1096     unlock_rmap(rmapp);
1097     return npages_dirty;
1098 }
1099 
1100 void kvmppc_harvest_vpa_dirty(struct kvmppc_vpa *vpa,
1101                   struct kvm_memory_slot *memslot,
1102                   unsigned long *map)
1103 {
1104     unsigned long gfn;
1105 
1106     if (!vpa->dirty || !vpa->pinned_addr)
1107         return;
1108     gfn = vpa->gpa >> PAGE_SHIFT;
1109     if (gfn < memslot->base_gfn ||
1110         gfn >= memslot->base_gfn + memslot->npages)
1111         return;
1112 
1113     vpa->dirty = false;
1114     if (map)
1115         __set_bit_le(gfn - memslot->base_gfn, map);
1116 }
1117 
1118 long kvmppc_hv_get_dirty_log_hpt(struct kvm *kvm,
1119             struct kvm_memory_slot *memslot, unsigned long *map)
1120 {
1121     unsigned long i;
1122     unsigned long *rmapp;
1123 
1124     preempt_disable();
1125     rmapp = memslot->arch.rmap;
1126     for (i = 0; i < memslot->npages; ++i) {
1127         int npages = kvm_test_clear_dirty_npages(kvm, rmapp);
1128         /*
1129          * Note that if npages > 0 then i must be a multiple of npages,
1130          * since we always put huge-page HPTEs in the rmap chain
1131          * corresponding to their page base address.
1132          */
1133         if (npages)
1134             set_dirty_bits(map, i, npages);
1135         ++rmapp;
1136     }
1137     preempt_enable();
1138     return 0;
1139 }
1140 
1141 void *kvmppc_pin_guest_page(struct kvm *kvm, unsigned long gpa,
1142                 unsigned long *nb_ret)
1143 {
1144     struct kvm_memory_slot *memslot;
1145     unsigned long gfn = gpa >> PAGE_SHIFT;
1146     struct page *page, *pages[1];
1147     int npages;
1148     unsigned long hva, offset;
1149     int srcu_idx;
1150 
1151     srcu_idx = srcu_read_lock(&kvm->srcu);
1152     memslot = gfn_to_memslot(kvm, gfn);
1153     if (!memslot || (memslot->flags & KVM_MEMSLOT_INVALID))
1154         goto err;
1155     hva = gfn_to_hva_memslot(memslot, gfn);
1156     npages = get_user_pages_fast(hva, 1, FOLL_WRITE, pages);
1157     if (npages < 1)
1158         goto err;
1159     page = pages[0];
1160     srcu_read_unlock(&kvm->srcu, srcu_idx);
1161 
1162     offset = gpa & (PAGE_SIZE - 1);
1163     if (nb_ret)
1164         *nb_ret = PAGE_SIZE - offset;
1165     return page_address(page) + offset;
1166 
1167  err:
1168     srcu_read_unlock(&kvm->srcu, srcu_idx);
1169     return NULL;
1170 }
1171 
1172 void kvmppc_unpin_guest_page(struct kvm *kvm, void *va, unsigned long gpa,
1173                  bool dirty)
1174 {
1175     struct page *page = virt_to_page(va);
1176     struct kvm_memory_slot *memslot;
1177     unsigned long gfn;
1178     int srcu_idx;
1179 
1180     put_page(page);
1181 
1182     if (!dirty)
1183         return;
1184 
1185     /* We need to mark this page dirty in the memslot dirty_bitmap, if any */
1186     gfn = gpa >> PAGE_SHIFT;
1187     srcu_idx = srcu_read_lock(&kvm->srcu);
1188     memslot = gfn_to_memslot(kvm, gfn);
1189     if (memslot && memslot->dirty_bitmap)
1190         set_bit_le(gfn - memslot->base_gfn, memslot->dirty_bitmap);
1191     srcu_read_unlock(&kvm->srcu, srcu_idx);
1192 }
1193 
1194 /*
1195  * HPT resizing
1196  */
1197 static int resize_hpt_allocate(struct kvm_resize_hpt *resize)
1198 {
1199     int rc;
1200 
1201     rc = kvmppc_allocate_hpt(&resize->hpt, resize->order);
1202     if (rc < 0)
1203         return rc;
1204 
1205     resize_hpt_debug(resize, "resize_hpt_allocate(): HPT @ 0x%lx\n",
1206              resize->hpt.virt);
1207 
1208     return 0;
1209 }
1210 
1211 static unsigned long resize_hpt_rehash_hpte(struct kvm_resize_hpt *resize,
1212                         unsigned long idx)
1213 {
1214     struct kvm *kvm = resize->kvm;
1215     struct kvm_hpt_info *old = &kvm->arch.hpt;
1216     struct kvm_hpt_info *new = &resize->hpt;
1217     unsigned long old_hash_mask = (1ULL << (old->order - 7)) - 1;
1218     unsigned long new_hash_mask = (1ULL << (new->order - 7)) - 1;
1219     __be64 *hptep, *new_hptep;
1220     unsigned long vpte, rpte, guest_rpte;
1221     int ret;
1222     struct revmap_entry *rev;
1223     unsigned long apsize, avpn, pteg, hash;
1224     unsigned long new_idx, new_pteg, replace_vpte;
1225     int pshift;
1226 
1227     hptep = (__be64 *)(old->virt + (idx << 4));
1228 
1229     /* Guest is stopped, so new HPTEs can't be added or faulted
1230      * in, only unmapped or altered by host actions.  So, it's
1231      * safe to check this before we take the HPTE lock */
1232     vpte = be64_to_cpu(hptep[0]);
1233     if (!(vpte & HPTE_V_VALID) && !(vpte & HPTE_V_ABSENT))
1234         return 0; /* nothing to do */
1235 
1236     while (!try_lock_hpte(hptep, HPTE_V_HVLOCK))
1237         cpu_relax();
1238 
1239     vpte = be64_to_cpu(hptep[0]);
1240 
1241     ret = 0;
1242     if (!(vpte & HPTE_V_VALID) && !(vpte & HPTE_V_ABSENT))
1243         /* Nothing to do */
1244         goto out;
1245 
1246     if (cpu_has_feature(CPU_FTR_ARCH_300)) {
1247         rpte = be64_to_cpu(hptep[1]);
1248         vpte = hpte_new_to_old_v(vpte, rpte);
1249     }
1250 
1251     /* Unmap */
1252     rev = &old->rev[idx];
1253     guest_rpte = rev->guest_rpte;
1254 
1255     ret = -EIO;
1256     apsize = kvmppc_actual_pgsz(vpte, guest_rpte);
1257     if (!apsize)
1258         goto out;
1259 
1260     if (vpte & HPTE_V_VALID) {
1261         unsigned long gfn = hpte_rpn(guest_rpte, apsize);
1262         int srcu_idx = srcu_read_lock(&kvm->srcu);
1263         struct kvm_memory_slot *memslot =
1264             __gfn_to_memslot(kvm_memslots(kvm), gfn);
1265 
1266         if (memslot) {
1267             unsigned long *rmapp;
1268             rmapp = &memslot->arch.rmap[gfn - memslot->base_gfn];
1269 
1270             lock_rmap(rmapp);
1271             kvmppc_unmap_hpte(kvm, idx, memslot, rmapp, gfn);
1272             unlock_rmap(rmapp);
1273         }
1274 
1275         srcu_read_unlock(&kvm->srcu, srcu_idx);
1276     }
1277 
1278     /* Reload PTE after unmap */
1279     vpte = be64_to_cpu(hptep[0]);
1280     BUG_ON(vpte & HPTE_V_VALID);
1281     BUG_ON(!(vpte & HPTE_V_ABSENT));
1282 
1283     ret = 0;
1284     if (!(vpte & HPTE_V_BOLTED))
1285         goto out;
1286 
1287     rpte = be64_to_cpu(hptep[1]);
1288 
1289     if (cpu_has_feature(CPU_FTR_ARCH_300)) {
1290         vpte = hpte_new_to_old_v(vpte, rpte);
1291         rpte = hpte_new_to_old_r(rpte);
1292     }
1293 
1294     pshift = kvmppc_hpte_base_page_shift(vpte, rpte);
1295     avpn = HPTE_V_AVPN_VAL(vpte) & ~(((1ul << pshift) - 1) >> 23);
1296     pteg = idx / HPTES_PER_GROUP;
1297     if (vpte & HPTE_V_SECONDARY)
1298         pteg = ~pteg;
1299 
1300     if (!(vpte & HPTE_V_1TB_SEG)) {
1301         unsigned long offset, vsid;
1302 
1303         /* We only have 28 - 23 bits of offset in avpn */
1304         offset = (avpn & 0x1f) << 23;
1305         vsid = avpn >> 5;
1306         /* We can find more bits from the pteg value */
1307         if (pshift < 23)
1308             offset |= ((vsid ^ pteg) & old_hash_mask) << pshift;
1309 
1310         hash = vsid ^ (offset >> pshift);
1311     } else {
1312         unsigned long offset, vsid;
1313 
1314         /* We only have 40 - 23 bits of seg_off in avpn */
1315         offset = (avpn & 0x1ffff) << 23;
1316         vsid = avpn >> 17;
1317         if (pshift < 23)
1318             offset |= ((vsid ^ (vsid << 25) ^ pteg) & old_hash_mask) << pshift;
1319 
1320         hash = vsid ^ (vsid << 25) ^ (offset >> pshift);
1321     }
1322 
1323     new_pteg = hash & new_hash_mask;
1324     if (vpte & HPTE_V_SECONDARY)
1325         new_pteg = ~hash & new_hash_mask;
1326 
1327     new_idx = new_pteg * HPTES_PER_GROUP + (idx % HPTES_PER_GROUP);
1328     new_hptep = (__be64 *)(new->virt + (new_idx << 4));
1329 
1330     replace_vpte = be64_to_cpu(new_hptep[0]);
1331     if (cpu_has_feature(CPU_FTR_ARCH_300)) {
1332         unsigned long replace_rpte = be64_to_cpu(new_hptep[1]);
1333         replace_vpte = hpte_new_to_old_v(replace_vpte, replace_rpte);
1334     }
1335 
1336     if (replace_vpte & (HPTE_V_VALID | HPTE_V_ABSENT)) {
1337         BUG_ON(new->order >= old->order);
1338 
1339         if (replace_vpte & HPTE_V_BOLTED) {
1340             if (vpte & HPTE_V_BOLTED)
1341                 /* Bolted collision, nothing we can do */
1342                 ret = -ENOSPC;
1343             /* Discard the new HPTE */
1344             goto out;
1345         }
1346 
1347         /* Discard the previous HPTE */
1348     }
1349 
1350     if (cpu_has_feature(CPU_FTR_ARCH_300)) {
1351         rpte = hpte_old_to_new_r(vpte, rpte);
1352         vpte = hpte_old_to_new_v(vpte);
1353     }
1354 
1355     new_hptep[1] = cpu_to_be64(rpte);
1356     new->rev[new_idx].guest_rpte = guest_rpte;
1357     /* No need for a barrier, since new HPT isn't active */
1358     new_hptep[0] = cpu_to_be64(vpte);
1359     unlock_hpte(new_hptep, vpte);
1360 
1361 out:
1362     unlock_hpte(hptep, vpte);
1363     return ret;
1364 }
1365 
1366 static int resize_hpt_rehash(struct kvm_resize_hpt *resize)
1367 {
1368     struct kvm *kvm = resize->kvm;
1369     unsigned  long i;
1370     int rc;
1371 
1372     for (i = 0; i < kvmppc_hpt_npte(&kvm->arch.hpt); i++) {
1373         rc = resize_hpt_rehash_hpte(resize, i);
1374         if (rc != 0)
1375             return rc;
1376     }
1377 
1378     return 0;
1379 }
1380 
1381 static void resize_hpt_pivot(struct kvm_resize_hpt *resize)
1382 {
1383     struct kvm *kvm = resize->kvm;
1384     struct kvm_hpt_info hpt_tmp;
1385 
1386     /* Exchange the pending tables in the resize structure with
1387      * the active tables */
1388 
1389     resize_hpt_debug(resize, "resize_hpt_pivot()\n");
1390 
1391     spin_lock(&kvm->mmu_lock);
1392     asm volatile("ptesync" : : : "memory");
1393 
1394     hpt_tmp = kvm->arch.hpt;
1395     kvmppc_set_hpt(kvm, &resize->hpt);
1396     resize->hpt = hpt_tmp;
1397 
1398     spin_unlock(&kvm->mmu_lock);
1399 
1400     synchronize_srcu_expedited(&kvm->srcu);
1401 
1402     if (cpu_has_feature(CPU_FTR_ARCH_300))
1403         kvmppc_setup_partition_table(kvm);
1404 
1405     resize_hpt_debug(resize, "resize_hpt_pivot() done\n");
1406 }
1407 
1408 static void resize_hpt_release(struct kvm *kvm, struct kvm_resize_hpt *resize)
1409 {
1410     if (WARN_ON(!mutex_is_locked(&kvm->arch.mmu_setup_lock)))
1411         return;
1412 
1413     if (!resize)
1414         return;
1415 
1416     if (resize->error != -EBUSY) {
1417         if (resize->hpt.virt)
1418             kvmppc_free_hpt(&resize->hpt);
1419         kfree(resize);
1420     }
1421 
1422     if (kvm->arch.resize_hpt == resize)
1423         kvm->arch.resize_hpt = NULL;
1424 }
1425 
1426 static void resize_hpt_prepare_work(struct work_struct *work)
1427 {
1428     struct kvm_resize_hpt *resize = container_of(work,
1429                              struct kvm_resize_hpt,
1430                              work);
1431     struct kvm *kvm = resize->kvm;
1432     int err = 0;
1433 
1434     if (WARN_ON(resize->error != -EBUSY))
1435         return;
1436 
1437     mutex_lock(&kvm->arch.mmu_setup_lock);
1438 
1439     /* Request is still current? */
1440     if (kvm->arch.resize_hpt == resize) {
1441         /* We may request large allocations here:
1442          * do not sleep with kvm->arch.mmu_setup_lock held for a while.
1443          */
1444         mutex_unlock(&kvm->arch.mmu_setup_lock);
1445 
1446         resize_hpt_debug(resize, "resize_hpt_prepare_work(): order = %d\n",
1447                  resize->order);
1448 
1449         err = resize_hpt_allocate(resize);
1450 
1451         /* We have strict assumption about -EBUSY
1452          * when preparing for HPT resize.
1453          */
1454         if (WARN_ON(err == -EBUSY))
1455             err = -EINPROGRESS;
1456 
1457         mutex_lock(&kvm->arch.mmu_setup_lock);
1458         /* It is possible that kvm->arch.resize_hpt != resize
1459          * after we grab kvm->arch.mmu_setup_lock again.
1460          */
1461     }
1462 
1463     resize->error = err;
1464 
1465     if (kvm->arch.resize_hpt != resize)
1466         resize_hpt_release(kvm, resize);
1467 
1468     mutex_unlock(&kvm->arch.mmu_setup_lock);
1469 }
1470 
1471 long kvm_vm_ioctl_resize_hpt_prepare(struct kvm *kvm,
1472                      struct kvm_ppc_resize_hpt *rhpt)
1473 {
1474     unsigned long flags = rhpt->flags;
1475     unsigned long shift = rhpt->shift;
1476     struct kvm_resize_hpt *resize;
1477     int ret;
1478 
1479     if (flags != 0 || kvm_is_radix(kvm))
1480         return -EINVAL;
1481 
1482     if (shift && ((shift < 18) || (shift > 46)))
1483         return -EINVAL;
1484 
1485     mutex_lock(&kvm->arch.mmu_setup_lock);
1486 
1487     resize = kvm->arch.resize_hpt;
1488 
1489     if (resize) {
1490         if (resize->order == shift) {
1491             /* Suitable resize in progress? */
1492             ret = resize->error;
1493             if (ret == -EBUSY)
1494                 ret = 100; /* estimated time in ms */
1495             else if (ret)
1496                 resize_hpt_release(kvm, resize);
1497 
1498             goto out;
1499         }
1500 
1501         /* not suitable, cancel it */
1502         resize_hpt_release(kvm, resize);
1503     }
1504 
1505     ret = 0;
1506     if (!shift)
1507         goto out; /* nothing to do */
1508 
1509     /* start new resize */
1510 
1511     resize = kzalloc(sizeof(*resize), GFP_KERNEL);
1512     if (!resize) {
1513         ret = -ENOMEM;
1514         goto out;
1515     }
1516 
1517     resize->error = -EBUSY;
1518     resize->order = shift;
1519     resize->kvm = kvm;
1520     INIT_WORK(&resize->work, resize_hpt_prepare_work);
1521     kvm->arch.resize_hpt = resize;
1522 
1523     schedule_work(&resize->work);
1524 
1525     ret = 100; /* estimated time in ms */
1526 
1527 out:
1528     mutex_unlock(&kvm->arch.mmu_setup_lock);
1529     return ret;
1530 }
1531 
1532 static void resize_hpt_boot_vcpu(void *opaque)
1533 {
1534     /* Nothing to do, just force a KVM exit */
1535 }
1536 
1537 long kvm_vm_ioctl_resize_hpt_commit(struct kvm *kvm,
1538                     struct kvm_ppc_resize_hpt *rhpt)
1539 {
1540     unsigned long flags = rhpt->flags;
1541     unsigned long shift = rhpt->shift;
1542     struct kvm_resize_hpt *resize;
1543     long ret;
1544 
1545     if (flags != 0 || kvm_is_radix(kvm))
1546         return -EINVAL;
1547 
1548     if (shift && ((shift < 18) || (shift > 46)))
1549         return -EINVAL;
1550 
1551     mutex_lock(&kvm->arch.mmu_setup_lock);
1552 
1553     resize = kvm->arch.resize_hpt;
1554 
1555     /* This shouldn't be possible */
1556     ret = -EIO;
1557     if (WARN_ON(!kvm->arch.mmu_ready))
1558         goto out_no_hpt;
1559 
1560     /* Stop VCPUs from running while we mess with the HPT */
1561     kvm->arch.mmu_ready = 0;
1562     smp_mb();
1563 
1564     /* Boot all CPUs out of the guest so they re-read
1565      * mmu_ready */
1566     on_each_cpu(resize_hpt_boot_vcpu, NULL, 1);
1567 
1568     ret = -ENXIO;
1569     if (!resize || (resize->order != shift))
1570         goto out;
1571 
1572     ret = resize->error;
1573     if (ret)
1574         goto out;
1575 
1576     ret = resize_hpt_rehash(resize);
1577     if (ret)
1578         goto out;
1579 
1580     resize_hpt_pivot(resize);
1581 
1582 out:
1583     /* Let VCPUs run again */
1584     kvm->arch.mmu_ready = 1;
1585     smp_mb();
1586 out_no_hpt:
1587     resize_hpt_release(kvm, resize);
1588     mutex_unlock(&kvm->arch.mmu_setup_lock);
1589     return ret;
1590 }
1591 
1592 /*
1593  * Functions for reading and writing the hash table via reads and
1594  * writes on a file descriptor.
1595  *
1596  * Reads return the guest view of the hash table, which has to be
1597  * pieced together from the real hash table and the guest_rpte
1598  * values in the revmap array.
1599  *
1600  * On writes, each HPTE written is considered in turn, and if it
1601  * is valid, it is written to the HPT as if an H_ENTER with the
1602  * exact flag set was done.  When the invalid count is non-zero
1603  * in the header written to the stream, the kernel will make
1604  * sure that that many HPTEs are invalid, and invalidate them
1605  * if not.
1606  */
1607 
1608 struct kvm_htab_ctx {
1609     unsigned long   index;
1610     unsigned long   flags;
1611     struct kvm  *kvm;
1612     int     first_pass;
1613 };
1614 
1615 #define HPTE_SIZE   (2 * sizeof(unsigned long))
1616 
1617 /*
1618  * Returns 1 if this HPT entry has been modified or has pending
1619  * R/C bit changes.
1620  */
1621 static int hpte_dirty(struct revmap_entry *revp, __be64 *hptp)
1622 {
1623     unsigned long rcbits_unset;
1624 
1625     if (revp->guest_rpte & HPTE_GR_MODIFIED)
1626         return 1;
1627 
1628     /* Also need to consider changes in reference and changed bits */
1629     rcbits_unset = ~revp->guest_rpte & (HPTE_R_R | HPTE_R_C);
1630     if ((be64_to_cpu(hptp[0]) & HPTE_V_VALID) &&
1631         (be64_to_cpu(hptp[1]) & rcbits_unset))
1632         return 1;
1633 
1634     return 0;
1635 }
1636 
1637 static long record_hpte(unsigned long flags, __be64 *hptp,
1638             unsigned long *hpte, struct revmap_entry *revp,
1639             int want_valid, int first_pass)
1640 {
1641     unsigned long v, r, hr;
1642     unsigned long rcbits_unset;
1643     int ok = 1;
1644     int valid, dirty;
1645 
1646     /* Unmodified entries are uninteresting except on the first pass */
1647     dirty = hpte_dirty(revp, hptp);
1648     if (!first_pass && !dirty)
1649         return 0;
1650 
1651     valid = 0;
1652     if (be64_to_cpu(hptp[0]) & (HPTE_V_VALID | HPTE_V_ABSENT)) {
1653         valid = 1;
1654         if ((flags & KVM_GET_HTAB_BOLTED_ONLY) &&
1655             !(be64_to_cpu(hptp[0]) & HPTE_V_BOLTED))
1656             valid = 0;
1657     }
1658     if (valid != want_valid)
1659         return 0;
1660 
1661     v = r = 0;
1662     if (valid || dirty) {
1663         /* lock the HPTE so it's stable and read it */
1664         preempt_disable();
1665         while (!try_lock_hpte(hptp, HPTE_V_HVLOCK))
1666             cpu_relax();
1667         v = be64_to_cpu(hptp[0]);
1668         hr = be64_to_cpu(hptp[1]);
1669         if (cpu_has_feature(CPU_FTR_ARCH_300)) {
1670             v = hpte_new_to_old_v(v, hr);
1671             hr = hpte_new_to_old_r(hr);
1672         }
1673 
1674         /* re-evaluate valid and dirty from synchronized HPTE value */
1675         valid = !!(v & HPTE_V_VALID);
1676         dirty = !!(revp->guest_rpte & HPTE_GR_MODIFIED);
1677 
1678         /* Harvest R and C into guest view if necessary */
1679         rcbits_unset = ~revp->guest_rpte & (HPTE_R_R | HPTE_R_C);
1680         if (valid && (rcbits_unset & hr)) {
1681             revp->guest_rpte |= (hr &
1682                 (HPTE_R_R | HPTE_R_C)) | HPTE_GR_MODIFIED;
1683             dirty = 1;
1684         }
1685 
1686         if (v & HPTE_V_ABSENT) {
1687             v &= ~HPTE_V_ABSENT;
1688             v |= HPTE_V_VALID;
1689             valid = 1;
1690         }
1691         if ((flags & KVM_GET_HTAB_BOLTED_ONLY) && !(v & HPTE_V_BOLTED))
1692             valid = 0;
1693 
1694         r = revp->guest_rpte;
1695         /* only clear modified if this is the right sort of entry */
1696         if (valid == want_valid && dirty) {
1697             r &= ~HPTE_GR_MODIFIED;
1698             revp->guest_rpte = r;
1699         }
1700         unlock_hpte(hptp, be64_to_cpu(hptp[0]));
1701         preempt_enable();
1702         if (!(valid == want_valid && (first_pass || dirty)))
1703             ok = 0;
1704     }
1705     hpte[0] = cpu_to_be64(v);
1706     hpte[1] = cpu_to_be64(r);
1707     return ok;
1708 }
1709 
1710 static ssize_t kvm_htab_read(struct file *file, char __user *buf,
1711                  size_t count, loff_t *ppos)
1712 {
1713     struct kvm_htab_ctx *ctx = file->private_data;
1714     struct kvm *kvm = ctx->kvm;
1715     struct kvm_get_htab_header hdr;
1716     __be64 *hptp;
1717     struct revmap_entry *revp;
1718     unsigned long i, nb, nw;
1719     unsigned long __user *lbuf;
1720     struct kvm_get_htab_header __user *hptr;
1721     unsigned long flags;
1722     int first_pass;
1723     unsigned long hpte[2];
1724 
1725     if (!access_ok(buf, count))
1726         return -EFAULT;
1727     if (kvm_is_radix(kvm))
1728         return 0;
1729 
1730     first_pass = ctx->first_pass;
1731     flags = ctx->flags;
1732 
1733     i = ctx->index;
1734     hptp = (__be64 *)(kvm->arch.hpt.virt + (i * HPTE_SIZE));
1735     revp = kvm->arch.hpt.rev + i;
1736     lbuf = (unsigned long __user *)buf;
1737 
1738     nb = 0;
1739     while (nb + sizeof(hdr) + HPTE_SIZE < count) {
1740         /* Initialize header */
1741         hptr = (struct kvm_get_htab_header __user *)buf;
1742         hdr.n_valid = 0;
1743         hdr.n_invalid = 0;
1744         nw = nb;
1745         nb += sizeof(hdr);
1746         lbuf = (unsigned long __user *)(buf + sizeof(hdr));
1747 
1748         /* Skip uninteresting entries, i.e. clean on not-first pass */
1749         if (!first_pass) {
1750             while (i < kvmppc_hpt_npte(&kvm->arch.hpt) &&
1751                    !hpte_dirty(revp, hptp)) {
1752                 ++i;
1753                 hptp += 2;
1754                 ++revp;
1755             }
1756         }
1757         hdr.index = i;
1758 
1759         /* Grab a series of valid entries */
1760         while (i < kvmppc_hpt_npte(&kvm->arch.hpt) &&
1761                hdr.n_valid < 0xffff &&
1762                nb + HPTE_SIZE < count &&
1763                record_hpte(flags, hptp, hpte, revp, 1, first_pass)) {
1764             /* valid entry, write it out */
1765             ++hdr.n_valid;
1766             if (__put_user(hpte[0], lbuf) ||
1767                 __put_user(hpte[1], lbuf + 1))
1768                 return -EFAULT;
1769             nb += HPTE_SIZE;
1770             lbuf += 2;
1771             ++i;
1772             hptp += 2;
1773             ++revp;
1774         }
1775         /* Now skip invalid entries while we can */
1776         while (i < kvmppc_hpt_npte(&kvm->arch.hpt) &&
1777                hdr.n_invalid < 0xffff &&
1778                record_hpte(flags, hptp, hpte, revp, 0, first_pass)) {
1779             /* found an invalid entry */
1780             ++hdr.n_invalid;
1781             ++i;
1782             hptp += 2;
1783             ++revp;
1784         }
1785 
1786         if (hdr.n_valid || hdr.n_invalid) {
1787             /* write back the header */
1788             if (__copy_to_user(hptr, &hdr, sizeof(hdr)))
1789                 return -EFAULT;
1790             nw = nb;
1791             buf = (char __user *)lbuf;
1792         } else {
1793             nb = nw;
1794         }
1795 
1796         /* Check if we've wrapped around the hash table */
1797         if (i >= kvmppc_hpt_npte(&kvm->arch.hpt)) {
1798             i = 0;
1799             ctx->first_pass = 0;
1800             break;
1801         }
1802     }
1803 
1804     ctx->index = i;
1805 
1806     return nb;
1807 }
1808 
1809 static ssize_t kvm_htab_write(struct file *file, const char __user *buf,
1810                   size_t count, loff_t *ppos)
1811 {
1812     struct kvm_htab_ctx *ctx = file->private_data;
1813     struct kvm *kvm = ctx->kvm;
1814     struct kvm_get_htab_header hdr;
1815     unsigned long i, j;
1816     unsigned long v, r;
1817     unsigned long __user *lbuf;
1818     __be64 *hptp;
1819     unsigned long tmp[2];
1820     ssize_t nb;
1821     long int err, ret;
1822     int mmu_ready;
1823     int pshift;
1824 
1825     if (!access_ok(buf, count))
1826         return -EFAULT;
1827     if (kvm_is_radix(kvm))
1828         return -EINVAL;
1829 
1830     /* lock out vcpus from running while we're doing this */
1831     mutex_lock(&kvm->arch.mmu_setup_lock);
1832     mmu_ready = kvm->arch.mmu_ready;
1833     if (mmu_ready) {
1834         kvm->arch.mmu_ready = 0;    /* temporarily */
1835         /* order mmu_ready vs. vcpus_running */
1836         smp_mb();
1837         if (atomic_read(&kvm->arch.vcpus_running)) {
1838             kvm->arch.mmu_ready = 1;
1839             mutex_unlock(&kvm->arch.mmu_setup_lock);
1840             return -EBUSY;
1841         }
1842     }
1843 
1844     err = 0;
1845     for (nb = 0; nb + sizeof(hdr) <= count; ) {
1846         err = -EFAULT;
1847         if (__copy_from_user(&hdr, buf, sizeof(hdr)))
1848             break;
1849 
1850         err = 0;
1851         if (nb + hdr.n_valid * HPTE_SIZE > count)
1852             break;
1853 
1854         nb += sizeof(hdr);
1855         buf += sizeof(hdr);
1856 
1857         err = -EINVAL;
1858         i = hdr.index;
1859         if (i >= kvmppc_hpt_npte(&kvm->arch.hpt) ||
1860             i + hdr.n_valid + hdr.n_invalid > kvmppc_hpt_npte(&kvm->arch.hpt))
1861             break;
1862 
1863         hptp = (__be64 *)(kvm->arch.hpt.virt + (i * HPTE_SIZE));
1864         lbuf = (unsigned long __user *)buf;
1865         for (j = 0; j < hdr.n_valid; ++j) {
1866             __be64 hpte_v;
1867             __be64 hpte_r;
1868 
1869             err = -EFAULT;
1870             if (__get_user(hpte_v, lbuf) ||
1871                 __get_user(hpte_r, lbuf + 1))
1872                 goto out;
1873             v = be64_to_cpu(hpte_v);
1874             r = be64_to_cpu(hpte_r);
1875             err = -EINVAL;
1876             if (!(v & HPTE_V_VALID))
1877                 goto out;
1878             pshift = kvmppc_hpte_base_page_shift(v, r);
1879             if (pshift <= 0)
1880                 goto out;
1881             lbuf += 2;
1882             nb += HPTE_SIZE;
1883 
1884             if (be64_to_cpu(hptp[0]) & (HPTE_V_VALID | HPTE_V_ABSENT))
1885                 kvmppc_do_h_remove(kvm, 0, i, 0, tmp);
1886             err = -EIO;
1887             ret = kvmppc_virtmode_do_h_enter(kvm, H_EXACT, i, v, r,
1888                              tmp);
1889             if (ret != H_SUCCESS) {
1890                 pr_err("kvm_htab_write ret %ld i=%ld v=%lx "
1891                        "r=%lx\n", ret, i, v, r);
1892                 goto out;
1893             }
1894             if (!mmu_ready && is_vrma_hpte(v)) {
1895                 unsigned long senc, lpcr;
1896 
1897                 senc = slb_pgsize_encoding(1ul << pshift);
1898                 kvm->arch.vrma_slb_v = senc | SLB_VSID_B_1T |
1899                     (VRMA_VSID << SLB_VSID_SHIFT_1T);
1900                 if (!cpu_has_feature(CPU_FTR_ARCH_300)) {
1901                     lpcr = senc << (LPCR_VRMASD_SH - 4);
1902                     kvmppc_update_lpcr(kvm, lpcr,
1903                                LPCR_VRMASD);
1904                 } else {
1905                     kvmppc_setup_partition_table(kvm);
1906                 }
1907                 mmu_ready = 1;
1908             }
1909             ++i;
1910             hptp += 2;
1911         }
1912 
1913         for (j = 0; j < hdr.n_invalid; ++j) {
1914             if (be64_to_cpu(hptp[0]) & (HPTE_V_VALID | HPTE_V_ABSENT))
1915                 kvmppc_do_h_remove(kvm, 0, i, 0, tmp);
1916             ++i;
1917             hptp += 2;
1918         }
1919         err = 0;
1920     }
1921 
1922  out:
1923     /* Order HPTE updates vs. mmu_ready */
1924     smp_wmb();
1925     kvm->arch.mmu_ready = mmu_ready;
1926     mutex_unlock(&kvm->arch.mmu_setup_lock);
1927 
1928     if (err)
1929         return err;
1930     return nb;
1931 }
1932 
1933 static int kvm_htab_release(struct inode *inode, struct file *filp)
1934 {
1935     struct kvm_htab_ctx *ctx = filp->private_data;
1936 
1937     filp->private_data = NULL;
1938     if (!(ctx->flags & KVM_GET_HTAB_WRITE))
1939         atomic_dec(&ctx->kvm->arch.hpte_mod_interest);
1940     kvm_put_kvm(ctx->kvm);
1941     kfree(ctx);
1942     return 0;
1943 }
1944 
1945 static const struct file_operations kvm_htab_fops = {
1946     .read       = kvm_htab_read,
1947     .write      = kvm_htab_write,
1948     .llseek     = default_llseek,
1949     .release    = kvm_htab_release,
1950 };
1951 
1952 int kvm_vm_ioctl_get_htab_fd(struct kvm *kvm, struct kvm_get_htab_fd *ghf)
1953 {
1954     int ret;
1955     struct kvm_htab_ctx *ctx;
1956     int rwflag;
1957 
1958     /* reject flags we don't recognize */
1959     if (ghf->flags & ~(KVM_GET_HTAB_BOLTED_ONLY | KVM_GET_HTAB_WRITE))
1960         return -EINVAL;
1961     ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
1962     if (!ctx)
1963         return -ENOMEM;
1964     kvm_get_kvm(kvm);
1965     ctx->kvm = kvm;
1966     ctx->index = ghf->start_index;
1967     ctx->flags = ghf->flags;
1968     ctx->first_pass = 1;
1969 
1970     rwflag = (ghf->flags & KVM_GET_HTAB_WRITE) ? O_WRONLY : O_RDONLY;
1971     ret = anon_inode_getfd("kvm-htab", &kvm_htab_fops, ctx, rwflag | O_CLOEXEC);
1972     if (ret < 0) {
1973         kfree(ctx);
1974         kvm_put_kvm_no_destroy(kvm);
1975         return ret;
1976     }
1977 
1978     if (rwflag == O_RDONLY) {
1979         mutex_lock(&kvm->slots_lock);
1980         atomic_inc(&kvm->arch.hpte_mod_interest);
1981         /* make sure kvmppc_do_h_enter etc. see the increment */
1982         synchronize_srcu_expedited(&kvm->srcu);
1983         mutex_unlock(&kvm->slots_lock);
1984     }
1985 
1986     return ret;
1987 }
1988 
1989 struct debugfs_htab_state {
1990     struct kvm  *kvm;
1991     struct mutex    mutex;
1992     unsigned long   hpt_index;
1993     int     chars_left;
1994     int     buf_index;
1995     char        buf[64];
1996 };
1997 
1998 static int debugfs_htab_open(struct inode *inode, struct file *file)
1999 {
2000     struct kvm *kvm = inode->i_private;
2001     struct debugfs_htab_state *p;
2002 
2003     p = kzalloc(sizeof(*p), GFP_KERNEL);
2004     if (!p)
2005         return -ENOMEM;
2006 
2007     kvm_get_kvm(kvm);
2008     p->kvm = kvm;
2009     mutex_init(&p->mutex);
2010     file->private_data = p;
2011 
2012     return nonseekable_open(inode, file);
2013 }
2014 
2015 static int debugfs_htab_release(struct inode *inode, struct file *file)
2016 {
2017     struct debugfs_htab_state *p = file->private_data;
2018 
2019     kvm_put_kvm(p->kvm);
2020     kfree(p);
2021     return 0;
2022 }
2023 
2024 static ssize_t debugfs_htab_read(struct file *file, char __user *buf,
2025                  size_t len, loff_t *ppos)
2026 {
2027     struct debugfs_htab_state *p = file->private_data;
2028     ssize_t ret, r;
2029     unsigned long i, n;
2030     unsigned long v, hr, gr;
2031     struct kvm *kvm;
2032     __be64 *hptp;
2033 
2034     kvm = p->kvm;
2035     if (kvm_is_radix(kvm))
2036         return 0;
2037 
2038     ret = mutex_lock_interruptible(&p->mutex);
2039     if (ret)
2040         return ret;
2041 
2042     if (p->chars_left) {
2043         n = p->chars_left;
2044         if (n > len)
2045             n = len;
2046         r = copy_to_user(buf, p->buf + p->buf_index, n);
2047         n -= r;
2048         p->chars_left -= n;
2049         p->buf_index += n;
2050         buf += n;
2051         len -= n;
2052         ret = n;
2053         if (r) {
2054             if (!n)
2055                 ret = -EFAULT;
2056             goto out;
2057         }
2058     }
2059 
2060     i = p->hpt_index;
2061     hptp = (__be64 *)(kvm->arch.hpt.virt + (i * HPTE_SIZE));
2062     for (; len != 0 && i < kvmppc_hpt_npte(&kvm->arch.hpt);
2063          ++i, hptp += 2) {
2064         if (!(be64_to_cpu(hptp[0]) & (HPTE_V_VALID | HPTE_V_ABSENT)))
2065             continue;
2066 
2067         /* lock the HPTE so it's stable and read it */
2068         preempt_disable();
2069         while (!try_lock_hpte(hptp, HPTE_V_HVLOCK))
2070             cpu_relax();
2071         v = be64_to_cpu(hptp[0]) & ~HPTE_V_HVLOCK;
2072         hr = be64_to_cpu(hptp[1]);
2073         gr = kvm->arch.hpt.rev[i].guest_rpte;
2074         unlock_hpte(hptp, v);
2075         preempt_enable();
2076 
2077         if (!(v & (HPTE_V_VALID | HPTE_V_ABSENT)))
2078             continue;
2079 
2080         n = scnprintf(p->buf, sizeof(p->buf),
2081                   "%6lx %.16lx %.16lx %.16lx\n",
2082                   i, v, hr, gr);
2083         p->chars_left = n;
2084         if (n > len)
2085             n = len;
2086         r = copy_to_user(buf, p->buf, n);
2087         n -= r;
2088         p->chars_left -= n;
2089         p->buf_index = n;
2090         buf += n;
2091         len -= n;
2092         ret += n;
2093         if (r) {
2094             if (!ret)
2095                 ret = -EFAULT;
2096             goto out;
2097         }
2098     }
2099     p->hpt_index = i;
2100 
2101  out:
2102     mutex_unlock(&p->mutex);
2103     return ret;
2104 }
2105 
2106 static ssize_t debugfs_htab_write(struct file *file, const char __user *buf,
2107                size_t len, loff_t *ppos)
2108 {
2109     return -EACCES;
2110 }
2111 
2112 static const struct file_operations debugfs_htab_fops = {
2113     .owner   = THIS_MODULE,
2114     .open    = debugfs_htab_open,
2115     .release = debugfs_htab_release,
2116     .read    = debugfs_htab_read,
2117     .write   = debugfs_htab_write,
2118     .llseek  = generic_file_llseek,
2119 };
2120 
2121 void kvmppc_mmu_debugfs_init(struct kvm *kvm)
2122 {
2123     debugfs_create_file("htab", 0400, kvm->debugfs_dentry, kvm,
2124                 &debugfs_htab_fops);
2125 }
2126 
2127 void kvmppc_mmu_book3s_hv_init(struct kvm_vcpu *vcpu)
2128 {
2129     struct kvmppc_mmu *mmu = &vcpu->arch.mmu;
2130 
2131     vcpu->arch.slb_nr = 32;     /* POWER7/POWER8 */
2132 
2133     mmu->xlate = kvmppc_mmu_book3s_64_hv_xlate;
2134 
2135     vcpu->arch.hflags |= BOOK3S_HFLAG_SLB;
2136 }
[ Source navigation ] [ Diff markup ] [ Identifier search ] [ general search ]

This page was automatically generated by the 2.1.0 LXR engine.

The LXR team
Valid CSS 2.1! Valid HTML 4.01!