KVM: Fix unused but set warnings
[linux-2.6.git] / arch / x86 / kvm / paging_tmpl.h
index a601713..9308be2 100644 (file)
@@ -7,6 +7,7 @@
  * MMU support
  *
  * Copyright (C) 2006 Qumranet, Inc.
+ * Copyright 2010 Red Hat, Inc. and/or its affilates.
  *
  * Authors:
  *   Yaniv Kamay  <yaniv@qumranet.com>
@@ -150,7 +151,9 @@ walk:
                walker->table_gfn[walker->level - 1] = table_gfn;
                walker->pte_gpa[walker->level - 1] = pte_gpa;
 
-               kvm_read_guest(vcpu->kvm, pte_gpa, &pte, sizeof(pte));
+               if (kvm_read_guest(vcpu->kvm, pte_gpa, &pte, sizeof(pte)))
+                       goto not_present;
+
                trace_kvm_mmu_paging_element(pte, walker->level);
 
                if (!is_present_gpte(pte))
@@ -160,7 +163,7 @@ walk:
                if (rsvd_fault)
                        goto access_error;
 
-               if (write_fault && !is_writeble_pte(pte))
+               if (write_fault && !is_writable_pte(pte))
                        if (user_fault || is_write_protection(vcpu))
                                goto access_error;
 
@@ -168,17 +171,17 @@ walk:
                        goto access_error;
 
 #if PTTYPE == 64
-               if (fetch_fault && is_nx(vcpu) && (pte & PT64_NX_MASK))
+               if (fetch_fault && (pte & PT64_NX_MASK))
                        goto access_error;
 #endif
 
                if (!(pte & PT_ACCESSED_MASK)) {
                        trace_kvm_mmu_set_accessed_bit(table_gfn, index,
                                                       sizeof(pte));
-                       mark_page_dirty(vcpu->kvm, table_gfn);
                        if (FNAME(cmpxchg_gpte)(vcpu->kvm, table_gfn,
                            index, pte, pte|PT_ACCESSED_MASK))
                                goto walk;
+                       mark_page_dirty(vcpu->kvm, table_gfn);
                        pte |= PT_ACCESSED_MASK;
                }
 
@@ -188,10 +191,10 @@ walk:
 
                if ((walker->level == PT_PAGE_TABLE_LEVEL) ||
                    ((walker->level == PT_DIRECTORY_LEVEL) &&
-                               (pte & PT_PAGE_SIZE_MASK)  &&
+                               is_large_pte(pte) &&
                                (PTTYPE == 64 || is_pse(vcpu))) ||
                    ((walker->level == PT_PDPE_LEVEL) &&
-                               (pte & PT_PAGE_SIZE_MASK)  &&
+                               is_large_pte(pte) &&
                                is_long_mode(vcpu))) {
                        int lvl = walker->level;
 
@@ -215,11 +218,11 @@ walk:
                bool ret;
 
                trace_kvm_mmu_set_dirty_bit(table_gfn, index, sizeof(pte));
-               mark_page_dirty(vcpu->kvm, table_gfn);
                ret = FNAME(cmpxchg_gpte)(vcpu->kvm, table_gfn, index, pte,
                            pte|PT_DIRTY_MASK);
                if (ret)
                        goto walk;
+               mark_page_dirty(vcpu->kvm, table_gfn);
                pte |= PT_DIRTY_MASK;
                walker->ptes[walker->level - 1] = pte;
        }
@@ -227,7 +230,7 @@ walk:
        walker->pt_access = pt_access;
        walker->pte_access = pte_access;
        pgprintk("%s: pte %llx pte_access %x pt_access %x\n",
-                __func__, (u64)pte, pt_access, pte_access);
+                __func__, (u64)pte, pte_access, pt_access);
        return 1;
 
 not_present:
@@ -256,11 +259,17 @@ static void FNAME(update_pte)(struct kvm_vcpu *vcpu, struct kvm_mmu_page *page,
        pt_element_t gpte;
        unsigned pte_access;
        pfn_t pfn;
+       u64 new_spte;
 
        gpte = *(const pt_element_t *)pte;
        if (~gpte & (PT_PRESENT_MASK | PT_ACCESSED_MASK)) {
-               if (!is_present_gpte(gpte))
-                       __set_spte(spte, shadow_notrap_nonpresent_pte);
+               if (!is_present_gpte(gpte)) {
+                       if (page->unsync)
+                               new_spte = shadow_trap_nonpresent_pte;
+                       else
+                               new_spte = shadow_notrap_nonpresent_pte;
+                       __set_spte(spte, new_spte);
+               }
                return;
        }
        pgprintk("%s: gpte %llx spte %p\n", __func__, (u64)gpte, spte);
@@ -330,10 +339,14 @@ static u64 *FNAME(fetch)(struct kvm_vcpu *vcpu, gva_t addr,
                        direct = 1;
                        if (!is_dirty_gpte(gw->ptes[level - delta]))
                                access &= ~ACC_WRITE_MASK;
-                       table_gfn = gpte_to_gfn(gw->ptes[level - delta]);
-                       /* advance table_gfn when emulating 1gb pages with 4k */
-                       if (delta == 0)
-                               table_gfn += PT_INDEX(addr, level);
+                       /*
+                        * It is a large guest pages backed by small host pages,
+                        * So we set @direct(@shadow_page->role.direct)=1, and
+                        * set @table_gfn(@shadow_page->gfn)=the base page frame
+                        * for linear translations.
+                        */
+                       table_gfn = gw->gfn & ~(KVM_PAGES_PER_HPAGE(level) - 1);
+                       access &= gw->pte_access;
                } else {
                        direct = 0;
                        table_gfn = gw->table_gfn[level - 2];
@@ -422,11 +435,8 @@ static int FNAME(page_fault)(struct kvm_vcpu *vcpu, gva_t addr,
        pfn = gfn_to_pfn(vcpu->kvm, walker.gfn);
 
        /* mmio */
-       if (is_error_pfn(pfn)) {
-               pgprintk("gfn %lx is mmio\n", walker.gfn);
-               kvm_release_pfn_clean(pfn);
-               return 1;
-       }
+       if (is_error_pfn(pfn))
+               return kvm_handle_bad_page(vcpu->kvm, walker.gfn, pfn);
 
        spin_lock(&vcpu->kvm->mmu_lock);
        if (mmu_notifier_retry(vcpu, mmu_seq))
@@ -434,6 +444,7 @@ static int FNAME(page_fault)(struct kvm_vcpu *vcpu, gva_t addr,
        kvm_mmu_free_some_pages(vcpu);
        sptep = FNAME(fetch)(vcpu, addr, &walker, user_fault, write_fault,
                             level, &write_pt, pfn);
+       (void)sptep;
        pgprintk("%s: shadow pte %p %llx ptwrite %d\n", __func__,
                 sptep, *sptep, write_pt);
 
@@ -455,7 +466,7 @@ out_unlock:
 static void FNAME(invlpg)(struct kvm_vcpu *vcpu, gva_t gva)
 {
        struct kvm_shadow_walk_iterator iterator;
-       pt_element_t gpte;
+       struct kvm_mmu_page *sp;
        gpa_t pte_gpa = -1;
        int level;
        u64 *sptep;
@@ -467,12 +478,18 @@ static void FNAME(invlpg)(struct kvm_vcpu *vcpu, gva_t gva)
                level = iterator.level;
                sptep = iterator.sptep;
 
-               if (level == PT_PAGE_TABLE_LEVEL  ||
-                   ((level == PT_DIRECTORY_LEVEL && is_large_pte(*sptep))) ||
-                   ((level == PT_PDPE_LEVEL && is_large_pte(*sptep)))) {
-                       struct kvm_mmu_page *sp = page_header(__pa(sptep));
+               sp = page_header(__pa(sptep));
+               if (is_last_spte(*sptep, level)) {
+                       int offset, shift;
 
-                       pte_gpa = (sp->gfn << PAGE_SHIFT);
+                       if (!sp->unsync)
+                               break;
+
+                       shift = PAGE_SHIFT -
+                                 (PT_LEVEL_BITS - PT64_LEVEL_BITS) * level;
+                       offset = sp->role.quadrant << shift;
+
+                       pte_gpa = (sp->gfn << PAGE_SHIFT) + offset;
                        pte_gpa += (sptep - sp->spt) * sizeof(pt_element_t);
 
                        if (is_shadow_present_pte(*sptep)) {
@@ -485,39 +502,42 @@ static void FNAME(invlpg)(struct kvm_vcpu *vcpu, gva_t gva)
                        break;
                }
 
-               if (!is_shadow_present_pte(*sptep))
+               if (!is_shadow_present_pte(*sptep) || !sp->unsync_children)
                        break;
        }
 
        if (need_flush)
                kvm_flush_remote_tlbs(vcpu->kvm);
+
+       atomic_inc(&vcpu->kvm->arch.invlpg_counter);
+
        spin_unlock(&vcpu->kvm->mmu_lock);
 
        if (pte_gpa == -1)
                return;
-       if (kvm_read_guest_atomic(vcpu->kvm, pte_gpa, &gpte,
-                                 sizeof(pt_element_t)))
+
+       if (mmu_topup_memory_caches(vcpu))
                return;
-       if (is_present_gpte(gpte) && (gpte & PT_ACCESSED_MASK)) {
-               if (mmu_topup_memory_caches(vcpu))
-                       return;
-               kvm_mmu_pte_write(vcpu, pte_gpa, (const u8 *)&gpte,
-                                 sizeof(pt_element_t), 0);
-       }
+       kvm_mmu_pte_write(vcpu, pte_gpa, NULL, sizeof(pt_element_t), 0);
 }
 
-static gpa_t FNAME(gva_to_gpa)(struct kvm_vcpu *vcpu, gva_t vaddr)
+static gpa_t FNAME(gva_to_gpa)(struct kvm_vcpu *vcpu, gva_t vaddr, u32 access,
+                              u32 *error)
 {
        struct guest_walker walker;
        gpa_t gpa = UNMAPPED_GVA;
        int r;
 
-       r = FNAME(walk_addr)(&walker, vcpu, vaddr, 0, 0, 0);
+       r = FNAME(walk_addr)(&walker, vcpu, vaddr,
+                            !!(access & PFERR_WRITE_MASK),
+                            !!(access & PFERR_USER_MASK),
+                            !!(access & PFERR_FETCH_MASK));
 
        if (r) {
                gpa = gfn_to_gpa(walker.gfn);
                gpa |= vaddr & ~PAGE_MASK;
-       }
+       } else if (error)
+               *error = walker.error_code;
 
        return gpa;
 }
@@ -562,30 +582,36 @@ static int FNAME(sync_page)(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp)
 {
        int i, offset, nr_present;
        bool reset_host_protection;
+       gpa_t first_pte_gpa;
 
        offset = nr_present = 0;
 
+       /* direct kvm_mmu_page can not be unsync. */
+       BUG_ON(sp->role.direct);
+
        if (PTTYPE == 32)
                offset = sp->role.quadrant << PT64_LEVEL_BITS;
 
+       first_pte_gpa = gfn_to_gpa(sp->gfn) + offset * sizeof(pt_element_t);
+
        for (i = 0; i < PT64_ENT_PER_PAGE; i++) {
                unsigned pte_access;
                pt_element_t gpte;
                gpa_t pte_gpa;
-               gfn_t gfn = sp->gfns[i];
+               gfn_t gfn;
 
                if (!is_shadow_present_pte(sp->spt[i]))
                        continue;
 
-               pte_gpa = gfn_to_gpa(sp->gfn);
-               pte_gpa += (i+offset) * sizeof(pt_element_t);
+               pte_gpa = first_pte_gpa + i * sizeof(pt_element_t);
 
                if (kvm_read_guest_atomic(vcpu->kvm, pte_gpa, &gpte,
                                          sizeof(pt_element_t)))
                        return -EINVAL;
 
-               if (gpte_to_gfn(gpte) != gfn || !is_present_gpte(gpte) ||
-                   !(gpte & PT_ACCESSED_MASK)) {
+               gfn = gpte_to_gfn(gpte);
+               if (unalias_gfn(vcpu->kvm, gfn) != sp->gfns[i] ||
+                     !is_present_gpte(gpte) || !(gpte & PT_ACCESSED_MASK)) {
                        u64 nonpresent;
 
                        rmap_remove(vcpu->kvm, &sp->spt[i]);