]> rtime.felk.cvut.cz Git - linux-imx.git/commitdiff
KVM: PPC: E500: Split host and guest MMU parts
authorAlexander Graf <agraf@suse.de>
Fri, 11 Jan 2013 14:22:45 +0000 (15:22 +0100)
committerAlexander Graf <agraf@suse.de>
Thu, 24 Jan 2013 18:23:31 +0000 (19:23 +0100)
This patch splits the file e500_tlb.c into e500_mmu.c (guest TLB handling)
and e500_mmu_host.c (host TLB handling).

The main benefit of this split is readability and maintainability. It's
just a lot harder to write dirty code :).

Signed-off-by: Alexander Graf <agraf@suse.de>
arch/powerpc/kvm/Makefile
arch/powerpc/kvm/e500_mmu.c [moved from arch/powerpc/kvm/e500_tlb.c with 56% similarity]
arch/powerpc/kvm/e500_mmu_host.c [new file with mode: 0644]
arch/powerpc/kvm/e500_mmu_host.h [new file with mode: 0644]

index 1e473d46322cd977e5948ec0012ee3563d04ba8b..b772eded8c26b53e1d65d644aa7988bf17e85874 100644 (file)
@@ -10,7 +10,8 @@ common-objs-y = $(addprefix ../../../virt/kvm/, kvm_main.o coalesced_mmio.o \
                                                eventfd.o)
 
 CFLAGS_44x_tlb.o  := -I.
-CFLAGS_e500_tlb.o := -I.
+CFLAGS_e500_mmu.o := -I.
+CFLAGS_e500_mmu_host.o := -I.
 CFLAGS_emulate.o  := -I.
 
 common-objs-y += powerpc.o emulate.o
@@ -35,7 +36,8 @@ kvm-e500-objs := \
        booke_emulate.o \
        booke_interrupts.o \
        e500.o \
-       e500_tlb.o \
+       e500_mmu.o \
+       e500_mmu_host.o \
        e500_emulate.o
 kvm-objs-$(CONFIG_KVM_E500V2) := $(kvm-e500-objs)
 
@@ -45,7 +47,8 @@ kvm-e500mc-objs := \
        booke_emulate.o \
        bookehv_interrupts.o \
        e500mc.o \
-       e500_tlb.o \
+       e500_mmu.o \
+       e500_mmu_host.o \
        e500_emulate.o
 kvm-objs-$(CONFIG_KVM_E500MC) := $(kvm-e500mc-objs)
 
similarity index 56%
rename from arch/powerpc/kvm/e500_tlb.c
rename to arch/powerpc/kvm/e500_mmu.c
index 48d1a4f1f5ff429cb044b55f646e1bcb2cf09a26..c3d1721aa1b8050e75f2a3ceb01d50c1d97f250d 100644 (file)
@@ -1,10 +1,11 @@
 /*
- * Copyright (C) 2008-2011 Freescale Semiconductor, Inc. All rights reserved.
+ * Copyright (C) 2008-2013 Freescale Semiconductor, Inc. All rights reserved.
  *
  * Author: Yu Liu, yu.liu@freescale.com
  *         Scott Wood, scottwood@freescale.com
  *         Ashish Kalra, ashish.kalra@freescale.com
  *         Varun Sethi, varun.sethi@freescale.com
+ *         Alexander Graf, agraf@suse.de
  *
  * Description:
  * This file is based on arch/powerpc/kvm/44x_tlb.c,
 #include "e500.h"
 #include "trace.h"
 #include "timing.h"
-
-#define to_htlb1_esel(esel) (host_tlb_params[1].entries - (esel) - 1)
-
-static struct kvmppc_e500_tlb_params host_tlb_params[E500_TLB_NUM];
+#include "e500_mmu_host.h"
 
 static inline unsigned int gtlb0_get_next_victim(
                struct kvmppc_vcpu_e500 *vcpu_e500)
@@ -50,197 +48,6 @@ static inline unsigned int gtlb0_get_next_victim(
        return victim;
 }
 
-static inline unsigned int tlb1_max_shadow_size(void)
-{
-       /* reserve one entry for magic page */
-       return host_tlb_params[1].entries - tlbcam_index - 1;
-}
-
-static inline int tlbe_is_writable(struct kvm_book3e_206_tlb_entry *tlbe)
-{
-       return tlbe->mas7_3 & (MAS3_SW|MAS3_UW);
-}
-
-static inline u32 e500_shadow_mas3_attrib(u32 mas3, int usermode)
-{
-       /* Mask off reserved bits. */
-       mas3 &= MAS3_ATTRIB_MASK;
-
-#ifndef CONFIG_KVM_BOOKE_HV
-       if (!usermode) {
-               /* Guest is in supervisor mode,
-                * so we need to translate guest
-                * supervisor permissions into user permissions. */
-               mas3 &= ~E500_TLB_USER_PERM_MASK;
-               mas3 |= (mas3 & E500_TLB_SUPER_PERM_MASK) << 1;
-       }
-       mas3 |= E500_TLB_SUPER_PERM_MASK;
-#endif
-       return mas3;
-}
-
-static inline u32 e500_shadow_mas2_attrib(u32 mas2, int usermode)
-{
-#ifdef CONFIG_SMP
-       return (mas2 & MAS2_ATTRIB_MASK) | MAS2_M;
-#else
-       return mas2 & MAS2_ATTRIB_MASK;
-#endif
-}
-
-/*
- * writing shadow tlb entry to host TLB
- */
-static inline void __write_host_tlbe(struct kvm_book3e_206_tlb_entry *stlbe,
-                                    uint32_t mas0)
-{
-       unsigned long flags;
-
-       local_irq_save(flags);
-       mtspr(SPRN_MAS0, mas0);
-       mtspr(SPRN_MAS1, stlbe->mas1);
-       mtspr(SPRN_MAS2, (unsigned long)stlbe->mas2);
-       mtspr(SPRN_MAS3, (u32)stlbe->mas7_3);
-       mtspr(SPRN_MAS7, (u32)(stlbe->mas7_3 >> 32));
-#ifdef CONFIG_KVM_BOOKE_HV
-       mtspr(SPRN_MAS8, stlbe->mas8);
-#endif
-       asm volatile("isync; tlbwe" : : : "memory");
-
-#ifdef CONFIG_KVM_BOOKE_HV
-       /* Must clear mas8 for other host tlbwe's */
-       mtspr(SPRN_MAS8, 0);
-       isync();
-#endif
-       local_irq_restore(flags);
-
-       trace_kvm_booke206_stlb_write(mas0, stlbe->mas8, stlbe->mas1,
-                                     stlbe->mas2, stlbe->mas7_3);
-}
-
-/*
- * Acquire a mas0 with victim hint, as if we just took a TLB miss.
- *
- * We don't care about the address we're searching for, other than that it's
- * in the right set and is not present in the TLB.  Using a zero PID and a
- * userspace address means we don't have to set and then restore MAS5, or
- * calculate a proper MAS6 value.
- */
-static u32 get_host_mas0(unsigned long eaddr)
-{
-       unsigned long flags;
-       u32 mas0;
-
-       local_irq_save(flags);
-       mtspr(SPRN_MAS6, 0);
-       asm volatile("tlbsx 0, %0" : : "b" (eaddr & ~CONFIG_PAGE_OFFSET));
-       mas0 = mfspr(SPRN_MAS0);
-       local_irq_restore(flags);
-
-       return mas0;
-}
-
-/* sesel is for tlb1 only */
-static inline void write_host_tlbe(struct kvmppc_vcpu_e500 *vcpu_e500,
-               int tlbsel, int sesel, struct kvm_book3e_206_tlb_entry *stlbe)
-{
-       u32 mas0;
-
-       if (tlbsel == 0) {
-               mas0 = get_host_mas0(stlbe->mas2);
-               __write_host_tlbe(stlbe, mas0);
-       } else {
-               __write_host_tlbe(stlbe,
-                                 MAS0_TLBSEL(1) |
-                                 MAS0_ESEL(to_htlb1_esel(sesel)));
-       }
-}
-
-/* sesel is for tlb1 only */
-static void write_stlbe(struct kvmppc_vcpu_e500 *vcpu_e500,
-                       struct kvm_book3e_206_tlb_entry *gtlbe,
-                       struct kvm_book3e_206_tlb_entry *stlbe,
-                       int stlbsel, int sesel)
-{
-       int stid;
-
-       preempt_disable();
-       stid = kvmppc_e500_get_tlb_stid(&vcpu_e500->vcpu, gtlbe);
-
-       stlbe->mas1 |= MAS1_TID(stid);
-       write_host_tlbe(vcpu_e500, stlbsel, sesel, stlbe);
-       preempt_enable();
-}
-
-#ifdef CONFIG_KVM_E500V2
-void kvmppc_map_magic(struct kvm_vcpu *vcpu)
-{
-       struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
-       struct kvm_book3e_206_tlb_entry magic;
-       ulong shared_page = ((ulong)vcpu->arch.shared) & PAGE_MASK;
-       unsigned int stid;
-       pfn_t pfn;
-
-       pfn = (pfn_t)virt_to_phys((void *)shared_page) >> PAGE_SHIFT;
-       get_page(pfn_to_page(pfn));
-
-       preempt_disable();
-       stid = kvmppc_e500_get_sid(vcpu_e500, 0, 0, 0, 0);
-
-       magic.mas1 = MAS1_VALID | MAS1_TS | MAS1_TID(stid) |
-                    MAS1_TSIZE(BOOK3E_PAGESZ_4K);
-       magic.mas2 = vcpu->arch.magic_page_ea | MAS2_M;
-       magic.mas7_3 = ((u64)pfn << PAGE_SHIFT) |
-                      MAS3_SW | MAS3_SR | MAS3_UW | MAS3_UR;
-       magic.mas8 = 0;
-
-       __write_host_tlbe(&magic, MAS0_TLBSEL(1) | MAS0_ESEL(tlbcam_index));
-       preempt_enable();
-}
-#endif
-
-static void inval_gtlbe_on_host(struct kvmppc_vcpu_e500 *vcpu_e500,
-                               int tlbsel, int esel)
-{
-       struct kvm_book3e_206_tlb_entry *gtlbe =
-               get_entry(vcpu_e500, tlbsel, esel);
-       struct tlbe_ref *ref = &vcpu_e500->gtlb_priv[tlbsel][esel].ref;
-
-       /* Don't bother with unmapped entries */
-       if (!(ref->flags & E500_TLB_VALID))
-               return;
-
-       if (tlbsel == 1 && ref->flags & E500_TLB_BITMAP) {
-               u64 tmp = vcpu_e500->g2h_tlb1_map[esel];
-               int hw_tlb_indx;
-               unsigned long flags;
-
-               local_irq_save(flags);
-               while (tmp) {
-                       hw_tlb_indx = __ilog2_u64(tmp & -tmp);
-                       mtspr(SPRN_MAS0,
-                             MAS0_TLBSEL(1) |
-                             MAS0_ESEL(to_htlb1_esel(hw_tlb_indx)));
-                       mtspr(SPRN_MAS1, 0);
-                       asm volatile("tlbwe");
-                       vcpu_e500->h2g_tlb1_rmap[hw_tlb_indx] = 0;
-                       tmp &= tmp - 1;
-               }
-               mb();
-               vcpu_e500->g2h_tlb1_map[esel] = 0;
-               ref->flags &= ~(E500_TLB_BITMAP | E500_TLB_VALID);
-               local_irq_restore(flags);
-
-               return;
-       }
-
-       /* Guest tlbe is backed by at most one host tlbe per shadow pid. */
-       kvmppc_e500_tlbil_one(vcpu_e500, gtlbe);
-
-       /* Mark the TLB as not backed by the host anymore */
-       ref->flags &= ~E500_TLB_VALID;
-}
-
 static int tlb0_set_base(gva_t addr, int sets, int ways)
 {
        int set_base;
@@ -319,70 +126,6 @@ static int kvmppc_e500_tlb_index(struct kvmppc_vcpu_e500 *vcpu_e500,
        return -1;
 }
 
-static inline void kvmppc_e500_ref_setup(struct tlbe_ref *ref,
-                                        struct kvm_book3e_206_tlb_entry *gtlbe,
-                                        pfn_t pfn)
-{
-       ref->pfn = pfn;
-       ref->flags = E500_TLB_VALID;
-
-       if (tlbe_is_writable(gtlbe))
-               kvm_set_pfn_dirty(pfn);
-}
-
-static inline void kvmppc_e500_ref_release(struct tlbe_ref *ref)
-{
-       if (ref->flags & E500_TLB_VALID) {
-               trace_kvm_booke206_ref_release(ref->pfn, ref->flags);
-               ref->flags = 0;
-       }
-}
-
-static void clear_tlb1_bitmap(struct kvmppc_vcpu_e500 *vcpu_e500)
-{
-       if (vcpu_e500->g2h_tlb1_map)
-               memset(vcpu_e500->g2h_tlb1_map, 0,
-                      sizeof(u64) * vcpu_e500->gtlb_params[1].entries);
-       if (vcpu_e500->h2g_tlb1_rmap)
-               memset(vcpu_e500->h2g_tlb1_rmap, 0,
-                      sizeof(unsigned int) * host_tlb_params[1].entries);
-}
-
-static void clear_tlb_privs(struct kvmppc_vcpu_e500 *vcpu_e500)
-{
-       int tlbsel = 0;
-       int i;
-
-       for (i = 0; i < vcpu_e500->gtlb_params[tlbsel].entries; i++) {
-               struct tlbe_ref *ref =
-                       &vcpu_e500->gtlb_priv[tlbsel][i].ref;
-               kvmppc_e500_ref_release(ref);
-       }
-}
-
-static void clear_tlb_refs(struct kvmppc_vcpu_e500 *vcpu_e500)
-{
-       int stlbsel = 1;
-       int i;
-
-       kvmppc_e500_tlbil_all(vcpu_e500);
-
-       for (i = 0; i < host_tlb_params[stlbsel].entries; i++) {
-               struct tlbe_ref *ref =
-                       &vcpu_e500->tlb_refs[stlbsel][i];
-               kvmppc_e500_ref_release(ref);
-       }
-
-       clear_tlb_privs(vcpu_e500);
-}
-
-void kvmppc_core_flush_tlb(struct kvm_vcpu *vcpu)
-{
-       struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
-       clear_tlb_refs(vcpu_e500);
-       clear_tlb1_bitmap(vcpu_e500);
-}
-
 static inline void kvmppc_e500_deliver_tlb_miss(struct kvm_vcpu *vcpu,
                unsigned int eaddr, int as)
 {
@@ -408,234 +151,6 @@ static inline void kvmppc_e500_deliver_tlb_miss(struct kvm_vcpu *vcpu,
                | (as ? MAS6_SAS : 0);
 }
 
-/* TID must be supplied by the caller */
-static inline void kvmppc_e500_setup_stlbe(
-       struct kvm_vcpu *vcpu,
-       struct kvm_book3e_206_tlb_entry *gtlbe,
-       int tsize, struct tlbe_ref *ref, u64 gvaddr,
-       struct kvm_book3e_206_tlb_entry *stlbe)
-{
-       pfn_t pfn = ref->pfn;
-       u32 pr = vcpu->arch.shared->msr & MSR_PR;
-
-       BUG_ON(!(ref->flags & E500_TLB_VALID));
-
-       /* Force IPROT=0 for all guest mappings. */
-       stlbe->mas1 = MAS1_TSIZE(tsize) | get_tlb_sts(gtlbe) | MAS1_VALID;
-       stlbe->mas2 = (gvaddr & MAS2_EPN) |
-                     e500_shadow_mas2_attrib(gtlbe->mas2, pr);
-       stlbe->mas7_3 = ((u64)pfn << PAGE_SHIFT) |
-                       e500_shadow_mas3_attrib(gtlbe->mas7_3, pr);
-
-#ifdef CONFIG_KVM_BOOKE_HV
-       stlbe->mas8 = MAS8_TGS | vcpu->kvm->arch.lpid;
-#endif
-}
-
-static inline int kvmppc_e500_shadow_map(struct kvmppc_vcpu_e500 *vcpu_e500,
-       u64 gvaddr, gfn_t gfn, struct kvm_book3e_206_tlb_entry *gtlbe,
-       int tlbsel, struct kvm_book3e_206_tlb_entry *stlbe,
-       struct tlbe_ref *ref)
-{
-       struct kvm_memory_slot *slot;
-       unsigned long pfn = 0; /* silence GCC warning */
-       unsigned long hva;
-       int pfnmap = 0;
-       int tsize = BOOK3E_PAGESZ_4K;
-
-       /*
-        * Translate guest physical to true physical, acquiring
-        * a page reference if it is normal, non-reserved memory.
-        *
-        * gfn_to_memslot() must succeed because otherwise we wouldn't
-        * have gotten this far.  Eventually we should just pass the slot
-        * pointer through from the first lookup.
-        */
-       slot = gfn_to_memslot(vcpu_e500->vcpu.kvm, gfn);
-       hva = gfn_to_hva_memslot(slot, gfn);
-
-       if (tlbsel == 1) {
-               struct vm_area_struct *vma;
-               down_read(&current->mm->mmap_sem);
-
-               vma = find_vma(current->mm, hva);
-               if (vma && hva >= vma->vm_start &&
-                   (vma->vm_flags & VM_PFNMAP)) {
-                       /*
-                        * This VMA is a physically contiguous region (e.g.
-                        * /dev/mem) that bypasses normal Linux page
-                        * management.  Find the overlap between the
-                        * vma and the memslot.
-                        */
-
-                       unsigned long start, end;
-                       unsigned long slot_start, slot_end;
-
-                       pfnmap = 1;
-
-                       start = vma->vm_pgoff;
-                       end = start +
-                             ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT);
-
-                       pfn = start + ((hva - vma->vm_start) >> PAGE_SHIFT);
-
-                       slot_start = pfn - (gfn - slot->base_gfn);
-                       slot_end = slot_start + slot->npages;
-
-                       if (start < slot_start)
-                               start = slot_start;
-                       if (end > slot_end)
-                               end = slot_end;
-
-                       tsize = (gtlbe->mas1 & MAS1_TSIZE_MASK) >>
-                               MAS1_TSIZE_SHIFT;
-
-                       /*
-                        * e500 doesn't implement the lowest tsize bit,
-                        * or 1K pages.
-                        */
-                       tsize = max(BOOK3E_PAGESZ_4K, tsize & ~1);
-
-                       /*
-                        * Now find the largest tsize (up to what the guest
-                        * requested) that will cover gfn, stay within the
-                        * range, and for which gfn and pfn are mutually
-                        * aligned.
-                        */
-
-                       for (; tsize > BOOK3E_PAGESZ_4K; tsize -= 2) {
-                               unsigned long gfn_start, gfn_end, tsize_pages;
-                               tsize_pages = 1 << (tsize - 2);
-
-                               gfn_start = gfn & ~(tsize_pages - 1);
-                               gfn_end = gfn_start + tsize_pages;
-
-                               if (gfn_start + pfn - gfn < start)
-                                       continue;
-                               if (gfn_end + pfn - gfn > end)
-                                       continue;
-                               if ((gfn & (tsize_pages - 1)) !=
-                                   (pfn & (tsize_pages - 1)))
-                                       continue;
-
-                               gvaddr &= ~((tsize_pages << PAGE_SHIFT) - 1);
-                               pfn &= ~(tsize_pages - 1);
-                               break;
-                       }
-               } else if (vma && hva >= vma->vm_start &&
-                          (vma->vm_flags & VM_HUGETLB)) {
-                       unsigned long psize = vma_kernel_pagesize(vma);
-
-                       tsize = (gtlbe->mas1 & MAS1_TSIZE_MASK) >>
-                               MAS1_TSIZE_SHIFT;
-
-                       /*
-                        * Take the largest page size that satisfies both host
-                        * and guest mapping
-                        */
-                       tsize = min(__ilog2(psize) - 10, tsize);
-
-                       /*
-                        * e500 doesn't implement the lowest tsize bit,
-                        * or 1K pages.
-                        */
-                       tsize = max(BOOK3E_PAGESZ_4K, tsize & ~1);
-               }
-
-               up_read(&current->mm->mmap_sem);
-       }
-
-       if (likely(!pfnmap)) {
-               unsigned long tsize_pages = 1 << (tsize + 10 - PAGE_SHIFT);
-               pfn = gfn_to_pfn_memslot(slot, gfn);
-               if (is_error_noslot_pfn(pfn)) {
-                       printk(KERN_ERR "Couldn't get real page for gfn %lx!\n",
-                                       (long)gfn);
-                       return -EINVAL;
-               }
-
-               /* Align guest and physical address to page map boundaries */
-               pfn &= ~(tsize_pages - 1);
-               gvaddr &= ~((tsize_pages << PAGE_SHIFT) - 1);
-       }
-
-       /* Drop old ref and setup new one. */
-       kvmppc_e500_ref_release(ref);
-       kvmppc_e500_ref_setup(ref, gtlbe, pfn);
-
-       kvmppc_e500_setup_stlbe(&vcpu_e500->vcpu, gtlbe, tsize,
-                               ref, gvaddr, stlbe);
-
-       /* Clear i-cache for new pages */
-       kvmppc_mmu_flush_icache(pfn);
-
-       /* Drop refcount on page, so that mmu notifiers can clear it */
-       kvm_release_pfn_clean(pfn);
-
-       return 0;
-}
-
-/* XXX only map the one-one case, for now use TLB0 */
-static int kvmppc_e500_tlb0_map(struct kvmppc_vcpu_e500 *vcpu_e500,
-                               int esel,
-                               struct kvm_book3e_206_tlb_entry *stlbe)
-{
-       struct kvm_book3e_206_tlb_entry *gtlbe;
-       struct tlbe_ref *ref;
-       int stlbsel = 0;
-       int sesel = 0;
-       int r;
-
-       gtlbe = get_entry(vcpu_e500, 0, esel);
-       ref = &vcpu_e500->gtlb_priv[0][esel].ref;
-
-       r = kvmppc_e500_shadow_map(vcpu_e500, get_tlb_eaddr(gtlbe),
-                       get_tlb_raddr(gtlbe) >> PAGE_SHIFT,
-                       gtlbe, 0, stlbe, ref);
-       if (r)
-               return r;
-
-       write_stlbe(vcpu_e500, gtlbe, stlbe, stlbsel, sesel);
-
-       return 0;
-}
-
-/* Caller must ensure that the specified guest TLB entry is safe to insert into
- * the shadow TLB. */
-/* XXX for both one-one and one-to-many , for now use TLB1 */
-static int kvmppc_e500_tlb1_map(struct kvmppc_vcpu_e500 *vcpu_e500,
-               u64 gvaddr, gfn_t gfn, struct kvm_book3e_206_tlb_entry *gtlbe,
-               struct kvm_book3e_206_tlb_entry *stlbe, int esel)
-{
-       struct tlbe_ref *ref;
-       unsigned int sesel;
-       int r;
-       int stlbsel = 1;
-
-       sesel = vcpu_e500->host_tlb1_nv++;
-
-       if (unlikely(vcpu_e500->host_tlb1_nv >= tlb1_max_shadow_size()))
-               vcpu_e500->host_tlb1_nv = 0;
-
-       ref = &vcpu_e500->tlb_refs[1][sesel];
-       r = kvmppc_e500_shadow_map(vcpu_e500, gvaddr, gfn, gtlbe, 1, stlbe,
-                                  ref);
-       if (r)
-               return r;
-
-       vcpu_e500->g2h_tlb1_map[esel] |= (u64)1 << sesel;
-       vcpu_e500->gtlb_priv[1][esel].ref.flags |= E500_TLB_BITMAP;
-       if (vcpu_e500->h2g_tlb1_rmap[sesel]) {
-               unsigned int idx = vcpu_e500->h2g_tlb1_rmap[sesel];
-               vcpu_e500->g2h_tlb1_map[idx] &= ~(1ULL << sesel);
-       }
-       vcpu_e500->h2g_tlb1_rmap[sesel] = esel;
-
-       write_stlbe(vcpu_e500, gtlbe, stlbe, stlbsel, sesel);
-
-       return 0;
-}
-
 static void kvmppc_recalc_tlb1map_range(struct kvmppc_vcpu_e500 *vcpu_e500)
 {
        int size = vcpu_e500->gtlb_params[1].entries;
@@ -1020,85 +535,6 @@ void kvmppc_mmu_destroy(struct kvm_vcpu *vcpu)
 {
 }
 
-void kvmppc_mmu_map(struct kvm_vcpu *vcpu, u64 eaddr, gpa_t gpaddr,
-                       unsigned int index)
-{
-       struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
-       struct tlbe_priv *priv;
-       struct kvm_book3e_206_tlb_entry *gtlbe, stlbe;
-       int tlbsel = tlbsel_of(index);
-       int esel = esel_of(index);
-
-       gtlbe = get_entry(vcpu_e500, tlbsel, esel);
-
-       switch (tlbsel) {
-       case 0:
-               priv = &vcpu_e500->gtlb_priv[tlbsel][esel];
-
-               /* Triggers after clear_tlb_refs or on initial mapping */
-               if (!(priv->ref.flags & E500_TLB_VALID)) {
-                       kvmppc_e500_tlb0_map(vcpu_e500, esel, &stlbe);
-               } else {
-                       kvmppc_e500_setup_stlbe(vcpu, gtlbe, BOOK3E_PAGESZ_4K,
-                                               &priv->ref, eaddr, &stlbe);
-                       write_stlbe(vcpu_e500, gtlbe, &stlbe, 0, 0);
-               }
-               break;
-
-       case 1: {
-               gfn_t gfn = gpaddr >> PAGE_SHIFT;
-               kvmppc_e500_tlb1_map(vcpu_e500, eaddr, gfn, gtlbe, &stlbe,
-                                    esel);
-               break;
-       }
-
-       default:
-               BUG();
-               break;
-       }
-}
-
-/************* MMU Notifiers *************/
-
-int kvm_unmap_hva(struct kvm *kvm, unsigned long hva)
-{
-       trace_kvm_unmap_hva(hva);
-
-       /*
-        * Flush all shadow tlb entries everywhere. This is slow, but
-        * we are 100% sure that we catch the to be unmapped page
-        */
-       kvm_flush_remote_tlbs(kvm);
-
-       return 0;
-}
-
-int kvm_unmap_hva_range(struct kvm *kvm, unsigned long start, unsigned long end)
-{
-       /* kvm_unmap_hva flushes everything anyways */
-       kvm_unmap_hva(kvm, start);
-
-       return 0;
-}
-
-int kvm_age_hva(struct kvm *kvm, unsigned long hva)
-{
-       /* XXX could be more clever ;) */
-       return 0;
-}
-
-int kvm_test_age_hva(struct kvm *kvm, unsigned long hva)
-{
-       /* XXX could be more clever ;) */
-       return 0;
-}
-
-void kvm_set_spte_hva(struct kvm *kvm, unsigned long hva, pte_t pte)
-{
-       /* The page will get remapped properly on its next fault */
-       kvm_unmap_hva(kvm, hva);
-}
-
 /*****************************************/
 
 static void free_gtlb(struct kvmppc_vcpu_e500 *vcpu_e500)
@@ -1309,37 +745,8 @@ int kvmppc_e500_tlb_init(struct kvmppc_vcpu_e500 *vcpu_e500)
        int entry_size = sizeof(struct kvm_book3e_206_tlb_entry);
        int entries = KVM_E500_TLB0_SIZE + KVM_E500_TLB1_SIZE;
 
-       host_tlb_params[0].entries = mfspr(SPRN_TLB0CFG) & TLBnCFG_N_ENTRY;
-       host_tlb_params[1].entries = mfspr(SPRN_TLB1CFG) & TLBnCFG_N_ENTRY;
-
-       /*
-        * This should never happen on real e500 hardware, but is
-        * architecturally possible -- e.g. in some weird nested
-        * virtualization case.
-        */
-       if (host_tlb_params[0].entries == 0 ||
-           host_tlb_params[1].entries == 0) {
-               pr_err("%s: need to know host tlb size\n", __func__);
-               return -ENODEV;
-       }
-
-       host_tlb_params[0].ways = (mfspr(SPRN_TLB0CFG) & TLBnCFG_ASSOC) >>
-                                 TLBnCFG_ASSOC_SHIFT;
-       host_tlb_params[1].ways = host_tlb_params[1].entries;
-
-       if (!is_power_of_2(host_tlb_params[0].entries) ||
-           !is_power_of_2(host_tlb_params[0].ways) ||
-           host_tlb_params[0].entries < host_tlb_params[0].ways ||
-           host_tlb_params[0].ways == 0) {
-               pr_err("%s: bad tlb0 host config: %u entries %u ways\n",
-                      __func__, host_tlb_params[0].entries,
-                      host_tlb_params[0].ways);
-               return -ENODEV;
-       }
-
-       host_tlb_params[0].sets =
-               host_tlb_params[0].entries / host_tlb_params[0].ways;
-       host_tlb_params[1].sets = 1;
+       if (e500_mmu_host_init(vcpu_e500))
+               goto err;
 
        vcpu_e500->gtlb_params[0].entries = KVM_E500_TLB0_SIZE;
        vcpu_e500->gtlb_params[1].entries = KVM_E500_TLB1_SIZE;
@@ -1358,18 +765,6 @@ int kvmppc_e500_tlb_init(struct kvmppc_vcpu_e500 *vcpu_e500)
        vcpu_e500->gtlb_offset[0] = 0;
        vcpu_e500->gtlb_offset[1] = KVM_E500_TLB0_SIZE;
 
-       vcpu_e500->tlb_refs[0] =
-               kzalloc(sizeof(struct tlbe_ref) * host_tlb_params[0].entries,
-                       GFP_KERNEL);
-       if (!vcpu_e500->tlb_refs[0])
-               goto err;
-
-       vcpu_e500->tlb_refs[1] =
-               kzalloc(sizeof(struct tlbe_ref) * host_tlb_params[1].entries,
-                       GFP_KERNEL);
-       if (!vcpu_e500->tlb_refs[1])
-               goto err;
-
        vcpu_e500->gtlb_priv[0] = kzalloc(sizeof(struct tlbe_ref) *
                                          vcpu_e500->gtlb_params[0].entries,
                                          GFP_KERNEL);
@@ -1388,12 +783,6 @@ int kvmppc_e500_tlb_init(struct kvmppc_vcpu_e500 *vcpu_e500)
        if (!vcpu_e500->g2h_tlb1_map)
                goto err;
 
-       vcpu_e500->h2g_tlb1_rmap = kzalloc(sizeof(unsigned int) *
-                                          host_tlb_params[1].entries,
-                                          GFP_KERNEL);
-       if (!vcpu_e500->h2g_tlb1_rmap)
-               goto err;
-
        /* Init TLB configuration register */
        vcpu->arch.tlbcfg[0] = mfspr(SPRN_TLB0CFG) &
                             ~(TLBnCFG_N_ENTRY | TLBnCFG_ASSOC);
@@ -1412,15 +801,11 @@ int kvmppc_e500_tlb_init(struct kvmppc_vcpu_e500 *vcpu_e500)
 
 err:
        free_gtlb(vcpu_e500);
-       kfree(vcpu_e500->tlb_refs[0]);
-       kfree(vcpu_e500->tlb_refs[1]);
        return -1;
 }
 
 void kvmppc_e500_tlb_uninit(struct kvmppc_vcpu_e500 *vcpu_e500)
 {
        free_gtlb(vcpu_e500);
-       kfree(vcpu_e500->h2g_tlb1_rmap);
-       kfree(vcpu_e500->tlb_refs[0]);
-       kfree(vcpu_e500->tlb_refs[1]);
+       e500_mmu_host_uninit(vcpu_e500);
 }
diff --git a/arch/powerpc/kvm/e500_mmu_host.c b/arch/powerpc/kvm/e500_mmu_host.c
new file mode 100644 (file)
index 0000000..4c32d65
--- /dev/null
@@ -0,0 +1,672 @@
+/*
+ * Copyright (C) 2008-2013 Freescale Semiconductor, Inc. All rights reserved.
+ *
+ * Author: Yu Liu, yu.liu@freescale.com
+ *         Scott Wood, scottwood@freescale.com
+ *         Ashish Kalra, ashish.kalra@freescale.com
+ *         Varun Sethi, varun.sethi@freescale.com
+ *         Alexander Graf, agraf@suse.de
+ *
+ * Description:
+ * This file is based on arch/powerpc/kvm/44x_tlb.c,
+ * by Hollis Blanchard <hollisb@us.ibm.com>.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License, version 2, as
+ * published by the Free Software Foundation.
+ */
+
+#include <linux/kernel.h>
+#include <linux/types.h>
+#include <linux/slab.h>
+#include <linux/string.h>
+#include <linux/kvm.h>
+#include <linux/kvm_host.h>
+#include <linux/highmem.h>
+#include <linux/log2.h>
+#include <linux/uaccess.h>
+#include <linux/sched.h>
+#include <linux/rwsem.h>
+#include <linux/vmalloc.h>
+#include <linux/hugetlb.h>
+#include <asm/kvm_ppc.h>
+
+#include "e500.h"
+#include "trace.h"
+#include "timing.h"
+#include "e500_mmu_host.h"
+
+#define to_htlb1_esel(esel) (host_tlb_params[1].entries - (esel) - 1)
+
+static struct kvmppc_e500_tlb_params host_tlb_params[E500_TLB_NUM];
+
+static inline unsigned int tlb1_max_shadow_size(void)
+{
+       /* reserve one entry for magic page */
+       return host_tlb_params[1].entries - tlbcam_index - 1;
+}
+
+static inline u32 e500_shadow_mas3_attrib(u32 mas3, int usermode)
+{
+       /* Mask off reserved bits. */
+       mas3 &= MAS3_ATTRIB_MASK;
+
+#ifndef CONFIG_KVM_BOOKE_HV
+       if (!usermode) {
+               /* Guest is in supervisor mode,
+                * so we need to translate guest
+                * supervisor permissions into user permissions. */
+               mas3 &= ~E500_TLB_USER_PERM_MASK;
+               mas3 |= (mas3 & E500_TLB_SUPER_PERM_MASK) << 1;
+       }
+       mas3 |= E500_TLB_SUPER_PERM_MASK;
+#endif
+       return mas3;
+}
+
+static inline u32 e500_shadow_mas2_attrib(u32 mas2, int usermode)
+{
+#ifdef CONFIG_SMP
+       return (mas2 & MAS2_ATTRIB_MASK) | MAS2_M;
+#else
+       return mas2 & MAS2_ATTRIB_MASK;
+#endif
+}
+
+/*
+ * writing shadow tlb entry to host TLB
+ */
+static inline void __write_host_tlbe(struct kvm_book3e_206_tlb_entry *stlbe,
+                                    uint32_t mas0)
+{
+       unsigned long flags;
+
+       local_irq_save(flags);
+       mtspr(SPRN_MAS0, mas0);
+       mtspr(SPRN_MAS1, stlbe->mas1);
+       mtspr(SPRN_MAS2, (unsigned long)stlbe->mas2);
+       mtspr(SPRN_MAS3, (u32)stlbe->mas7_3);
+       mtspr(SPRN_MAS7, (u32)(stlbe->mas7_3 >> 32));
+#ifdef CONFIG_KVM_BOOKE_HV
+       mtspr(SPRN_MAS8, stlbe->mas8);
+#endif
+       asm volatile("isync; tlbwe" : : : "memory");
+
+#ifdef CONFIG_KVM_BOOKE_HV
+       /* Must clear mas8 for other host tlbwe's */
+       mtspr(SPRN_MAS8, 0);
+       isync();
+#endif
+       local_irq_restore(flags);
+
+       trace_kvm_booke206_stlb_write(mas0, stlbe->mas8, stlbe->mas1,
+                                     stlbe->mas2, stlbe->mas7_3);
+}
+
+/*
+ * Acquire a mas0 with victim hint, as if we just took a TLB miss.
+ *
+ * We don't care about the address we're searching for, other than that it's
+ * in the right set and is not present in the TLB.  Using a zero PID and a
+ * userspace address means we don't have to set and then restore MAS5, or
+ * calculate a proper MAS6 value.
+ */
+static u32 get_host_mas0(unsigned long eaddr)
+{
+       unsigned long flags;
+       u32 mas0;
+
+       local_irq_save(flags);
+       mtspr(SPRN_MAS6, 0);
+       asm volatile("tlbsx 0, %0" : : "b" (eaddr & ~CONFIG_PAGE_OFFSET));
+       mas0 = mfspr(SPRN_MAS0);
+       local_irq_restore(flags);
+
+       return mas0;
+}
+
+/* sesel is for tlb1 only */
+static inline void write_host_tlbe(struct kvmppc_vcpu_e500 *vcpu_e500,
+               int tlbsel, int sesel, struct kvm_book3e_206_tlb_entry *stlbe)
+{
+       u32 mas0;
+
+       if (tlbsel == 0) {
+               mas0 = get_host_mas0(stlbe->mas2);
+               __write_host_tlbe(stlbe, mas0);
+       } else {
+               __write_host_tlbe(stlbe,
+                                 MAS0_TLBSEL(1) |
+                                 MAS0_ESEL(to_htlb1_esel(sesel)));
+       }
+}
+
+/* sesel is for tlb1 only */
+static void write_stlbe(struct kvmppc_vcpu_e500 *vcpu_e500,
+                       struct kvm_book3e_206_tlb_entry *gtlbe,
+                       struct kvm_book3e_206_tlb_entry *stlbe,
+                       int stlbsel, int sesel)
+{
+       int stid;
+
+       preempt_disable();
+       stid = kvmppc_e500_get_tlb_stid(&vcpu_e500->vcpu, gtlbe);
+
+       stlbe->mas1 |= MAS1_TID(stid);
+       write_host_tlbe(vcpu_e500, stlbsel, sesel, stlbe);
+       preempt_enable();
+}
+
+#ifdef CONFIG_KVM_E500V2
+/* XXX should be a hook in the gva2hpa translation */
+void kvmppc_map_magic(struct kvm_vcpu *vcpu)
+{
+       struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
+       struct kvm_book3e_206_tlb_entry magic;
+       ulong shared_page = ((ulong)vcpu->arch.shared) & PAGE_MASK;
+       unsigned int stid;
+       pfn_t pfn;
+
+       pfn = (pfn_t)virt_to_phys((void *)shared_page) >> PAGE_SHIFT;
+       get_page(pfn_to_page(pfn));
+
+       preempt_disable();
+       stid = kvmppc_e500_get_sid(vcpu_e500, 0, 0, 0, 0);
+
+       magic.mas1 = MAS1_VALID | MAS1_TS | MAS1_TID(stid) |
+                    MAS1_TSIZE(BOOK3E_PAGESZ_4K);
+       magic.mas2 = vcpu->arch.magic_page_ea | MAS2_M;
+       magic.mas7_3 = ((u64)pfn << PAGE_SHIFT) |
+                      MAS3_SW | MAS3_SR | MAS3_UW | MAS3_UR;
+       magic.mas8 = 0;
+
+       __write_host_tlbe(&magic, MAS0_TLBSEL(1) | MAS0_ESEL(tlbcam_index));
+       preempt_enable();
+}
+#endif
+
+void inval_gtlbe_on_host(struct kvmppc_vcpu_e500 *vcpu_e500, int tlbsel,
+                        int esel)
+{
+       struct kvm_book3e_206_tlb_entry *gtlbe =
+               get_entry(vcpu_e500, tlbsel, esel);
+       struct tlbe_ref *ref = &vcpu_e500->gtlb_priv[tlbsel][esel].ref;
+
+       /* Don't bother with unmapped entries */
+       if (!(ref->flags & E500_TLB_VALID))
+               return;
+
+       if (tlbsel == 1 && ref->flags & E500_TLB_BITMAP) {
+               u64 tmp = vcpu_e500->g2h_tlb1_map[esel];
+               int hw_tlb_indx;
+               unsigned long flags;
+
+               local_irq_save(flags);
+               while (tmp) {
+                       hw_tlb_indx = __ilog2_u64(tmp & -tmp);
+                       mtspr(SPRN_MAS0,
+                             MAS0_TLBSEL(1) |
+                             MAS0_ESEL(to_htlb1_esel(hw_tlb_indx)));
+                       mtspr(SPRN_MAS1, 0);
+                       asm volatile("tlbwe");
+                       vcpu_e500->h2g_tlb1_rmap[hw_tlb_indx] = 0;
+                       tmp &= tmp - 1;
+               }
+               mb();
+               vcpu_e500->g2h_tlb1_map[esel] = 0;
+               ref->flags &= ~(E500_TLB_BITMAP | E500_TLB_VALID);
+               local_irq_restore(flags);
+
+               return;
+       }
+
+       /* Guest tlbe is backed by at most one host tlbe per shadow pid. */
+       kvmppc_e500_tlbil_one(vcpu_e500, gtlbe);
+
+       /* Mark the TLB as not backed by the host anymore */
+       ref->flags &= ~E500_TLB_VALID;
+}
+
+static inline int tlbe_is_writable(struct kvm_book3e_206_tlb_entry *tlbe)
+{
+       return tlbe->mas7_3 & (MAS3_SW|MAS3_UW);
+}
+
+static inline void kvmppc_e500_ref_setup(struct tlbe_ref *ref,
+                                        struct kvm_book3e_206_tlb_entry *gtlbe,
+                                        pfn_t pfn)
+{
+       ref->pfn = pfn;
+       ref->flags = E500_TLB_VALID;
+
+       if (tlbe_is_writable(gtlbe))
+               kvm_set_pfn_dirty(pfn);
+}
+
+static inline void kvmppc_e500_ref_release(struct tlbe_ref *ref)
+{
+       if (ref->flags & E500_TLB_VALID) {
+               trace_kvm_booke206_ref_release(ref->pfn, ref->flags);
+               ref->flags = 0;
+       }
+}
+
+void clear_tlb1_bitmap(struct kvmppc_vcpu_e500 *vcpu_e500)
+{
+       if (vcpu_e500->g2h_tlb1_map)
+               memset(vcpu_e500->g2h_tlb1_map, 0,
+                      sizeof(u64) * vcpu_e500->gtlb_params[1].entries);
+       if (vcpu_e500->h2g_tlb1_rmap)
+               memset(vcpu_e500->h2g_tlb1_rmap, 0,
+                      sizeof(unsigned int) * host_tlb_params[1].entries);
+}
+
+static void clear_tlb_privs(struct kvmppc_vcpu_e500 *vcpu_e500)
+{
+       int tlbsel = 0;
+       int i;
+
+       for (i = 0; i < vcpu_e500->gtlb_params[tlbsel].entries; i++) {
+               struct tlbe_ref *ref =
+                       &vcpu_e500->gtlb_priv[tlbsel][i].ref;
+               kvmppc_e500_ref_release(ref);
+       }
+}
+
+void clear_tlb_refs(struct kvmppc_vcpu_e500 *vcpu_e500)
+{
+       int stlbsel = 1;
+       int i;
+
+       kvmppc_e500_tlbil_all(vcpu_e500);
+
+       for (i = 0; i < host_tlb_params[stlbsel].entries; i++) {
+               struct tlbe_ref *ref =
+                       &vcpu_e500->tlb_refs[stlbsel][i];
+               kvmppc_e500_ref_release(ref);
+       }
+
+       clear_tlb_privs(vcpu_e500);
+}
+
+void kvmppc_core_flush_tlb(struct kvm_vcpu *vcpu)
+{
+       struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
+       clear_tlb_refs(vcpu_e500);
+       clear_tlb1_bitmap(vcpu_e500);
+}
+
+/* TID must be supplied by the caller */
+static void kvmppc_e500_setup_stlbe(
+       struct kvm_vcpu *vcpu,
+       struct kvm_book3e_206_tlb_entry *gtlbe,
+       int tsize, struct tlbe_ref *ref, u64 gvaddr,
+       struct kvm_book3e_206_tlb_entry *stlbe)
+{
+       pfn_t pfn = ref->pfn;
+       u32 pr = vcpu->arch.shared->msr & MSR_PR;
+
+       BUG_ON(!(ref->flags & E500_TLB_VALID));
+
+       /* Force IPROT=0 for all guest mappings. */
+       stlbe->mas1 = MAS1_TSIZE(tsize) | get_tlb_sts(gtlbe) | MAS1_VALID;
+       stlbe->mas2 = (gvaddr & MAS2_EPN) |
+                     e500_shadow_mas2_attrib(gtlbe->mas2, pr);
+       stlbe->mas7_3 = ((u64)pfn << PAGE_SHIFT) |
+                       e500_shadow_mas3_attrib(gtlbe->mas7_3, pr);
+
+#ifdef CONFIG_KVM_BOOKE_HV
+       stlbe->mas8 = MAS8_TGS | vcpu->kvm->arch.lpid;
+#endif
+}
+
+static inline int kvmppc_e500_shadow_map(struct kvmppc_vcpu_e500 *vcpu_e500,
+       u64 gvaddr, gfn_t gfn, struct kvm_book3e_206_tlb_entry *gtlbe,
+       int tlbsel, struct kvm_book3e_206_tlb_entry *stlbe,
+       struct tlbe_ref *ref)
+{
+       struct kvm_memory_slot *slot;
+       unsigned long pfn = 0; /* silence GCC warning */
+       unsigned long hva;
+       int pfnmap = 0;
+       int tsize = BOOK3E_PAGESZ_4K;
+
+       /*
+        * Translate guest physical to true physical, acquiring
+        * a page reference if it is normal, non-reserved memory.
+        *
+        * gfn_to_memslot() must succeed because otherwise we wouldn't
+        * have gotten this far.  Eventually we should just pass the slot
+        * pointer through from the first lookup.
+        */
+       slot = gfn_to_memslot(vcpu_e500->vcpu.kvm, gfn);
+       hva = gfn_to_hva_memslot(slot, gfn);
+
+       if (tlbsel == 1) {
+               struct vm_area_struct *vma;
+               down_read(&current->mm->mmap_sem);
+
+               vma = find_vma(current->mm, hva);
+               if (vma && hva >= vma->vm_start &&
+                   (vma->vm_flags & VM_PFNMAP)) {
+                       /*
+                        * This VMA is a physically contiguous region (e.g.
+                        * /dev/mem) that bypasses normal Linux page
+                        * management.  Find the overlap between the
+                        * vma and the memslot.
+                        */
+
+                       unsigned long start, end;
+                       unsigned long slot_start, slot_end;
+
+                       pfnmap = 1;
+
+                       start = vma->vm_pgoff;
+                       end = start +
+                             ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT);
+
+                       pfn = start + ((hva - vma->vm_start) >> PAGE_SHIFT);
+
+                       slot_start = pfn - (gfn - slot->base_gfn);
+                       slot_end = slot_start + slot->npages;
+
+                       if (start < slot_start)
+                               start = slot_start;
+                       if (end > slot_end)
+                               end = slot_end;
+
+                       tsize = (gtlbe->mas1 & MAS1_TSIZE_MASK) >>
+                               MAS1_TSIZE_SHIFT;
+
+                       /*
+                        * e500 doesn't implement the lowest tsize bit,
+                        * or 1K pages.
+                        */
+                       tsize = max(BOOK3E_PAGESZ_4K, tsize & ~1);
+
+                       /*
+                        * Now find the largest tsize (up to what the guest
+                        * requested) that will cover gfn, stay within the
+                        * range, and for which gfn and pfn are mutually
+                        * aligned.
+                        */
+
+                       for (; tsize > BOOK3E_PAGESZ_4K; tsize -= 2) {
+                               unsigned long gfn_start, gfn_end, tsize_pages;
+                               tsize_pages = 1 << (tsize - 2);
+
+                               gfn_start = gfn & ~(tsize_pages - 1);
+                               gfn_end = gfn_start + tsize_pages;
+
+                               if (gfn_start + pfn - gfn < start)
+                                       continue;
+                               if (gfn_end + pfn - gfn > end)
+                                       continue;
+                               if ((gfn & (tsize_pages - 1)) !=
+                                   (pfn & (tsize_pages - 1)))
+                                       continue;
+
+                               gvaddr &= ~((tsize_pages << PAGE_SHIFT) - 1);
+                               pfn &= ~(tsize_pages - 1);
+                               break;
+                       }
+               } else if (vma && hva >= vma->vm_start &&
+                          (vma->vm_flags & VM_HUGETLB)) {
+                       unsigned long psize = vma_kernel_pagesize(vma);
+
+                       tsize = (gtlbe->mas1 & MAS1_TSIZE_MASK) >>
+                               MAS1_TSIZE_SHIFT;
+
+                       /*
+                        * Take the largest page size that satisfies both host
+                        * and guest mapping
+                        */
+                       tsize = min(__ilog2(psize) - 10, tsize);
+
+                       /*
+                        * e500 doesn't implement the lowest tsize bit,
+                        * or 1K pages.
+                        */
+                       tsize = max(BOOK3E_PAGESZ_4K, tsize & ~1);
+               }
+
+               up_read(&current->mm->mmap_sem);
+       }
+
+       if (likely(!pfnmap)) {
+               unsigned long tsize_pages = 1 << (tsize + 10 - PAGE_SHIFT);
+               pfn = gfn_to_pfn_memslot(slot, gfn);
+               if (is_error_noslot_pfn(pfn)) {
+                       printk(KERN_ERR "Couldn't get real page for gfn %lx!\n",
+                                       (long)gfn);
+                       return -EINVAL;
+               }
+
+               /* Align guest and physical address to page map boundaries */
+               pfn &= ~(tsize_pages - 1);
+               gvaddr &= ~((tsize_pages << PAGE_SHIFT) - 1);
+       }
+
+       /* Drop old ref and setup new one. */
+       kvmppc_e500_ref_release(ref);
+       kvmppc_e500_ref_setup(ref, gtlbe, pfn);
+
+       kvmppc_e500_setup_stlbe(&vcpu_e500->vcpu, gtlbe, tsize,
+                               ref, gvaddr, stlbe);
+
+       /* Clear i-cache for new pages */
+       kvmppc_mmu_flush_icache(pfn);
+
+       /* Drop refcount on page, so that mmu notifiers can clear it */
+       kvm_release_pfn_clean(pfn);
+
+       return 0;
+}
+
+/* XXX only map the one-one case, for now use TLB0 */
+static int kvmppc_e500_tlb0_map(struct kvmppc_vcpu_e500 *vcpu_e500, int esel,
+                               struct kvm_book3e_206_tlb_entry *stlbe)
+{
+       struct kvm_book3e_206_tlb_entry *gtlbe;
+       struct tlbe_ref *ref;
+       int stlbsel = 0;
+       int sesel = 0;
+       int r;
+
+       gtlbe = get_entry(vcpu_e500, 0, esel);
+       ref = &vcpu_e500->gtlb_priv[0][esel].ref;
+
+       r = kvmppc_e500_shadow_map(vcpu_e500, get_tlb_eaddr(gtlbe),
+                       get_tlb_raddr(gtlbe) >> PAGE_SHIFT,
+                       gtlbe, 0, stlbe, ref);
+       if (r)
+               return r;
+
+       write_stlbe(vcpu_e500, gtlbe, stlbe, stlbsel, sesel);
+
+       return 0;
+}
+
+/* Caller must ensure that the specified guest TLB entry is safe to insert into
+ * the shadow TLB. */
+/* XXX for both one-one and one-to-many , for now use TLB1 */
+static int kvmppc_e500_tlb1_map(struct kvmppc_vcpu_e500 *vcpu_e500,
+               u64 gvaddr, gfn_t gfn, struct kvm_book3e_206_tlb_entry *gtlbe,
+               struct kvm_book3e_206_tlb_entry *stlbe, int esel)
+{
+       struct tlbe_ref *ref;
+       unsigned int sesel;
+       int r;
+       int stlbsel = 1;
+
+       sesel = vcpu_e500->host_tlb1_nv++;
+
+       if (unlikely(vcpu_e500->host_tlb1_nv >= tlb1_max_shadow_size()))
+               vcpu_e500->host_tlb1_nv = 0;
+
+       ref = &vcpu_e500->tlb_refs[1][sesel];
+       r = kvmppc_e500_shadow_map(vcpu_e500, gvaddr, gfn, gtlbe, 1, stlbe,
+                                  ref);
+       if (r)
+               return r;
+
+       vcpu_e500->g2h_tlb1_map[esel] |= (u64)1 << sesel;
+       vcpu_e500->gtlb_priv[1][esel].ref.flags |= E500_TLB_BITMAP;
+       if (vcpu_e500->h2g_tlb1_rmap[sesel]) {
+               unsigned int idx = vcpu_e500->h2g_tlb1_rmap[sesel];
+               vcpu_e500->g2h_tlb1_map[idx] &= ~(1ULL << sesel);
+       }
+       vcpu_e500->h2g_tlb1_rmap[sesel] = esel;
+
+       write_stlbe(vcpu_e500, gtlbe, stlbe, stlbsel, sesel);
+
+       return 0;
+}
+
+void kvmppc_mmu_map(struct kvm_vcpu *vcpu, u64 eaddr, gpa_t gpaddr,
+                   unsigned int index)
+{
+       struct kvmppc_vcpu_e500 *vcpu_e500 = to_e500(vcpu);
+       struct tlbe_priv *priv;
+       struct kvm_book3e_206_tlb_entry *gtlbe, stlbe;
+       int tlbsel = tlbsel_of(index);
+       int esel = esel_of(index);
+
+       gtlbe = get_entry(vcpu_e500, tlbsel, esel);
+
+       switch (tlbsel) {
+       case 0:
+               priv = &vcpu_e500->gtlb_priv[tlbsel][esel];
+
+               /* Triggers after clear_tlb_refs or on initial mapping */
+               if (!(priv->ref.flags & E500_TLB_VALID)) {
+                       kvmppc_e500_tlb0_map(vcpu_e500, esel, &stlbe);
+               } else {
+                       kvmppc_e500_setup_stlbe(vcpu, gtlbe, BOOK3E_PAGESZ_4K,
+                                               &priv->ref, eaddr, &stlbe);
+                       write_stlbe(vcpu_e500, gtlbe, &stlbe, 0, 0);
+               }
+               break;
+
+       case 1: {
+               gfn_t gfn = gpaddr >> PAGE_SHIFT;
+               kvmppc_e500_tlb1_map(vcpu_e500, eaddr, gfn, gtlbe, &stlbe,
+                                    esel);
+               break;
+       }
+
+       default:
+               BUG();
+               break;
+       }
+}
+
+/************* MMU Notifiers *************/
+
+int kvm_unmap_hva(struct kvm *kvm, unsigned long hva)
+{
+       trace_kvm_unmap_hva(hva);
+
+       /*
+        * Flush all shadow tlb entries everywhere. This is slow, but
+        * we are 100% sure that we catch the to be unmapped page
+        */
+       kvm_flush_remote_tlbs(kvm);
+
+       return 0;
+}
+
+int kvm_unmap_hva_range(struct kvm *kvm, unsigned long start, unsigned long end)
+{
+       /* kvm_unmap_hva flushes everything anyways */
+       kvm_unmap_hva(kvm, start);
+
+       return 0;
+}
+
+int kvm_age_hva(struct kvm *kvm, unsigned long hva)
+{
+       /* XXX could be more clever ;) */
+       return 0;
+}
+
+int kvm_test_age_hva(struct kvm *kvm, unsigned long hva)
+{
+       /* XXX could be more clever ;) */
+       return 0;
+}
+
+void kvm_set_spte_hva(struct kvm *kvm, unsigned long hva, pte_t pte)
+{
+       /* The page will get remapped properly on its next fault */
+       kvm_unmap_hva(kvm, hva);
+}
+
+/*****************************************/
+
+int e500_mmu_host_init(struct kvmppc_vcpu_e500 *vcpu_e500)
+{
+       host_tlb_params[0].entries = mfspr(SPRN_TLB0CFG) & TLBnCFG_N_ENTRY;
+       host_tlb_params[1].entries = mfspr(SPRN_TLB1CFG) & TLBnCFG_N_ENTRY;
+
+       /*
+        * This should never happen on real e500 hardware, but is
+        * architecturally possible -- e.g. in some weird nested
+        * virtualization case.
+        */
+       if (host_tlb_params[0].entries == 0 ||
+           host_tlb_params[1].entries == 0) {
+               pr_err("%s: need to know host tlb size\n", __func__);
+               return -ENODEV;
+       }
+
+       host_tlb_params[0].ways = (mfspr(SPRN_TLB0CFG) & TLBnCFG_ASSOC) >>
+                                 TLBnCFG_ASSOC_SHIFT;
+       host_tlb_params[1].ways = host_tlb_params[1].entries;
+
+       if (!is_power_of_2(host_tlb_params[0].entries) ||
+           !is_power_of_2(host_tlb_params[0].ways) ||
+           host_tlb_params[0].entries < host_tlb_params[0].ways ||
+           host_tlb_params[0].ways == 0) {
+               pr_err("%s: bad tlb0 host config: %u entries %u ways\n",
+                      __func__, host_tlb_params[0].entries,
+                      host_tlb_params[0].ways);
+               return -ENODEV;
+       }
+
+       host_tlb_params[0].sets =
+               host_tlb_params[0].entries / host_tlb_params[0].ways;
+       host_tlb_params[1].sets = 1;
+
+       vcpu_e500->tlb_refs[0] =
+               kzalloc(sizeof(struct tlbe_ref) * host_tlb_params[0].entries,
+                       GFP_KERNEL);
+       if (!vcpu_e500->tlb_refs[0])
+               goto err;
+
+       vcpu_e500->tlb_refs[1] =
+               kzalloc(sizeof(struct tlbe_ref) * host_tlb_params[1].entries,
+                       GFP_KERNEL);
+       if (!vcpu_e500->tlb_refs[1])
+               goto err;
+
+       vcpu_e500->h2g_tlb1_rmap = kzalloc(sizeof(unsigned int) *
+                                          host_tlb_params[1].entries,
+                                          GFP_KERNEL);
+       if (!vcpu_e500->h2g_tlb1_rmap)
+               goto err;
+
+       return 0;
+
+err:
+       kfree(vcpu_e500->tlb_refs[0]);
+       kfree(vcpu_e500->tlb_refs[1]);
+       return -EINVAL;
+}
+
+void e500_mmu_host_uninit(struct kvmppc_vcpu_e500 *vcpu_e500)
+{
+       kfree(vcpu_e500->h2g_tlb1_rmap);
+       kfree(vcpu_e500->tlb_refs[0]);
+       kfree(vcpu_e500->tlb_refs[1]);
+}
diff --git a/arch/powerpc/kvm/e500_mmu_host.h b/arch/powerpc/kvm/e500_mmu_host.h
new file mode 100644 (file)
index 0000000..9e4d4a2
--- /dev/null
@@ -0,0 +1,20 @@
+/*
+ * Copyright (C) 2008-2013 Freescale Semiconductor, Inc. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License, version 2, as
+ * published by the Free Software Foundation.
+ */
+
+#ifndef KVM_E500_MMU_HOST_H
+#define KVM_E500_MMU_HOST_H
+
+void inval_gtlbe_on_host(struct kvmppc_vcpu_e500 *vcpu_e500, int tlbsel,
+                        int esel);
+
+void clear_tlb1_bitmap(struct kvmppc_vcpu_e500 *vcpu_e500);
+void clear_tlb_refs(struct kvmppc_vcpu_e500 *vcpu_e500);
+int e500_mmu_host_init(struct kvmppc_vcpu_e500 *vcpu_e500);
+void e500_mmu_host_uninit(struct kvmppc_vcpu_e500 *vcpu_e500);
+
+#endif /* KVM_E500_MMU_HOST_H */