ARM: KVM: be more thorough when invalidating TLBs
authorMarc Zyngier <marc.zyngier@arm.com>
Tue, 14 May 2013 11:11:34 +0000 (12:11 +0100)
committerGleb Natapov <gleb@redhat.com>
Mon, 3 Jun 2013 07:58:56 +0000 (10:58 +0300)
The KVM/ARM MMU code doesn't take care of invalidating TLBs before
freeing a {pte,pmd} table. This could cause problems if the page
is reallocated and then speculated into by another CPU.

Reported-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Christoffer Dall <cdall@cs.columbia.edu>
arch/arm/kvm/mmu.c

index 965706578f13bdb0ff062f9df2502608de7a3ae9..84ba67b982c0d32546dae5092ec84ac1d620756b 100644 (file)
@@ -43,7 +43,14 @@ static phys_addr_t hyp_idmap_vector;
 
 static void kvm_tlb_flush_vmid_ipa(struct kvm *kvm, phys_addr_t ipa)
 {
-       kvm_call_hyp(__kvm_tlb_flush_vmid_ipa, kvm, ipa);
+       /*
+        * This function also gets called when dealing with HYP page
+        * tables. As HYP doesn't have an associated struct kvm (and
+        * the HYP page tables are fairly static), we don't do
+        * anything there.
+        */
+       if (kvm)
+               kvm_call_hyp(__kvm_tlb_flush_vmid_ipa, kvm, ipa);
 }
 
 static int mmu_topup_memory_cache(struct kvm_mmu_memory_cache *cache,
@@ -78,18 +85,20 @@ static void *mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc)
        return p;
 }
 
-static void clear_pud_entry(pud_t *pud)
+static void clear_pud_entry(struct kvm *kvm, pud_t *pud, phys_addr_t addr)
 {
        pmd_t *pmd_table = pmd_offset(pud, 0);
        pud_clear(pud);
+       kvm_tlb_flush_vmid_ipa(kvm, addr);
        pmd_free(NULL, pmd_table);
        put_page(virt_to_page(pud));
 }
 
-static void clear_pmd_entry(pmd_t *pmd)
+static void clear_pmd_entry(struct kvm *kvm, pmd_t *pmd, phys_addr_t addr)
 {
        pte_t *pte_table = pte_offset_kernel(pmd, 0);
        pmd_clear(pmd);
+       kvm_tlb_flush_vmid_ipa(kvm, addr);
        pte_free_kernel(NULL, pte_table);
        put_page(virt_to_page(pmd));
 }
@@ -100,11 +109,12 @@ static bool pmd_empty(pmd_t *pmd)
        return page_count(pmd_page) == 1;
 }
 
-static void clear_pte_entry(pte_t *pte)
+static void clear_pte_entry(struct kvm *kvm, pte_t *pte, phys_addr_t addr)
 {
        if (pte_present(*pte)) {
                kvm_set_pte(pte, __pte(0));
                put_page(virt_to_page(pte));
+               kvm_tlb_flush_vmid_ipa(kvm, addr);
        }
 }
 
@@ -114,7 +124,8 @@ static bool pte_empty(pte_t *pte)
        return page_count(pte_page) == 1;
 }
 
-static void unmap_range(pgd_t *pgdp, unsigned long long start, u64 size)
+static void unmap_range(struct kvm *kvm, pgd_t *pgdp,
+                       unsigned long long start, u64 size)
 {
        pgd_t *pgd;
        pud_t *pud;
@@ -138,15 +149,15 @@ static void unmap_range(pgd_t *pgdp, unsigned long long start, u64 size)
                }
 
                pte = pte_offset_kernel(pmd, addr);
-               clear_pte_entry(pte);
+               clear_pte_entry(kvm, pte, addr);
                range = PAGE_SIZE;
 
                /* If we emptied the pte, walk back up the ladder */
                if (pte_empty(pte)) {
-                       clear_pmd_entry(pmd);
+                       clear_pmd_entry(kvm, pmd, addr);
                        range = PMD_SIZE;
                        if (pmd_empty(pmd)) {
-                               clear_pud_entry(pud);
+                               clear_pud_entry(kvm, pud, addr);
                                range = PUD_SIZE;
                        }
                }
@@ -165,14 +176,14 @@ void free_boot_hyp_pgd(void)
        mutex_lock(&kvm_hyp_pgd_mutex);
 
        if (boot_hyp_pgd) {
-               unmap_range(boot_hyp_pgd, hyp_idmap_start, PAGE_SIZE);
-               unmap_range(boot_hyp_pgd, TRAMPOLINE_VA, PAGE_SIZE);
+               unmap_range(NULL, boot_hyp_pgd, hyp_idmap_start, PAGE_SIZE);
+               unmap_range(NULL, boot_hyp_pgd, TRAMPOLINE_VA, PAGE_SIZE);
                kfree(boot_hyp_pgd);
                boot_hyp_pgd = NULL;
        }
 
        if (hyp_pgd)
-               unmap_range(hyp_pgd, TRAMPOLINE_VA, PAGE_SIZE);
+               unmap_range(NULL, hyp_pgd, TRAMPOLINE_VA, PAGE_SIZE);
 
        kfree(init_bounce_page);
        init_bounce_page = NULL;
@@ -200,9 +211,10 @@ void free_hyp_pgds(void)
 
        if (hyp_pgd) {
                for (addr = PAGE_OFFSET; virt_addr_valid(addr); addr += PGDIR_SIZE)
-                       unmap_range(hyp_pgd, KERN_TO_HYP(addr), PGDIR_SIZE);
+                       unmap_range(NULL, hyp_pgd, KERN_TO_HYP(addr), PGDIR_SIZE);
                for (addr = VMALLOC_START; is_vmalloc_addr((void*)addr); addr += PGDIR_SIZE)
-                       unmap_range(hyp_pgd, KERN_TO_HYP(addr), PGDIR_SIZE);
+                       unmap_range(NULL, hyp_pgd, KERN_TO_HYP(addr), PGDIR_SIZE);
+
                kfree(hyp_pgd);
                hyp_pgd = NULL;
        }
@@ -393,7 +405,7 @@ int kvm_alloc_stage2_pgd(struct kvm *kvm)
  */
 static void unmap_stage2_range(struct kvm *kvm, phys_addr_t start, u64 size)
 {
-       unmap_range(kvm->arch.pgd, start, size);
+       unmap_range(kvm, kvm->arch.pgd, start, size);
 }
 
 /**
@@ -675,7 +687,6 @@ static void handle_hva_to_gpa(struct kvm *kvm,
 static void kvm_unmap_hva_handler(struct kvm *kvm, gpa_t gpa, void *data)
 {
        unmap_stage2_range(kvm, gpa, PAGE_SIZE);
-       kvm_tlb_flush_vmid_ipa(kvm, gpa);
 }
 
 int kvm_unmap_hva(struct kvm *kvm, unsigned long hva)