ARM: KVM: rework HYP page table freeing
authorMarc Zyngier <marc.zyngier@arm.com>
Fri, 12 Apr 2013 18:12:05 +0000 (19:12 +0100)
committerChristoffer Dall <cdall@cs.columbia.edu>
Mon, 29 Apr 2013 05:23:10 +0000 (22:23 -0700)
There is no point in freeing HYP page tables differently from Stage-2.
They now have the same requirements, and should be dealt with the same way.

Promote unmap_stage2_range to be The One True Way, and get rid of a number
of nasty bugs in the process (good thing we never actually called free_hyp_pmds
before...).

Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
Signed-off-by: Christoffer Dall <cdall@cs.columbia.edu>
arch/arm/include/asm/kvm_mmu.h
arch/arm/kvm/arm.c
arch/arm/kvm/mmu.c

index 3c71a1d..92eb20d 100644 (file)
@@ -32,7 +32,7 @@
 
 int create_hyp_mappings(void *from, void *to);
 int create_hyp_io_mappings(void *from, void *to, phys_addr_t);
-void free_hyp_pmds(void);
+void free_hyp_pgds(void);
 
 int kvm_alloc_stage2_pgd(struct kvm *kvm);
 void kvm_free_stage2_pgd(struct kvm *kvm);
index 7403f88..16f164a 100644 (file)
@@ -928,7 +928,7 @@ static int init_hyp_mode(void)
 out_free_vfp:
        free_percpu(kvm_host_vfp_state);
 out_free_mappings:
-       free_hyp_pmds();
+       free_hyp_pgds();
 out_free_stack_pages:
        for_each_possible_cpu(cpu)
                free_page(per_cpu(kvm_arm_hyp_stack_page, cpu));
index bfc5927..7464824 100644 (file)
@@ -72,56 +72,104 @@ static void *mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc)
        return p;
 }
 
-static void free_ptes(pmd_t *pmd, unsigned long addr)
+static void clear_pud_entry(pud_t *pud)
 {
-       pte_t *pte;
-       unsigned int i;
+       pmd_t *pmd_table = pmd_offset(pud, 0);
+       pud_clear(pud);
+       pmd_free(NULL, pmd_table);
+       put_page(virt_to_page(pud));
+}
 
-       for (i = 0; i < PTRS_PER_PMD; i++, addr += PMD_SIZE) {
-               if (!pmd_none(*pmd) && pmd_table(*pmd)) {
-                       pte = pte_offset_kernel(pmd, addr);
-                       pte_free_kernel(NULL, pte);
-               }
-               pmd++;
+static void clear_pmd_entry(pmd_t *pmd)
+{
+       pte_t *pte_table = pte_offset_kernel(pmd, 0);
+       pmd_clear(pmd);
+       pte_free_kernel(NULL, pte_table);
+       put_page(virt_to_page(pmd));
+}
+
+static bool pmd_empty(pmd_t *pmd)
+{
+       struct page *pmd_page = virt_to_page(pmd);
+       return page_count(pmd_page) == 1;
+}
+
+static void clear_pte_entry(pte_t *pte)
+{
+       if (pte_present(*pte)) {
+               kvm_set_pte(pte, __pte(0));
+               put_page(virt_to_page(pte));
        }
 }
 
-static void free_hyp_pgd_entry(unsigned long addr)
+static bool pte_empty(pte_t *pte)
+{
+       struct page *pte_page = virt_to_page(pte);
+       return page_count(pte_page) == 1;
+}
+
+static void unmap_range(pgd_t *pgdp, unsigned long long start, u64 size)
 {
        pgd_t *pgd;
        pud_t *pud;
        pmd_t *pmd;
-       unsigned long hyp_addr = KERN_TO_HYP(addr);
+       pte_t *pte;
+       unsigned long long addr = start, end = start + size;
+       u64 range;
 
-       pgd = hyp_pgd + pgd_index(hyp_addr);
-       pud = pud_offset(pgd, hyp_addr);
+       while (addr < end) {
+               pgd = pgdp + pgd_index(addr);
+               pud = pud_offset(pgd, addr);
+               if (pud_none(*pud)) {
+                       addr += PUD_SIZE;
+                       continue;
+               }
 
-       if (pud_none(*pud))
-               return;
-       BUG_ON(pud_bad(*pud));
+               pmd = pmd_offset(pud, addr);
+               if (pmd_none(*pmd)) {
+                       addr += PMD_SIZE;
+                       continue;
+               }
 
-       pmd = pmd_offset(pud, hyp_addr);
-       free_ptes(pmd, addr);
-       pmd_free(NULL, pmd);
-       pud_clear(pud);
+               pte = pte_offset_kernel(pmd, addr);
+               clear_pte_entry(pte);
+               range = PAGE_SIZE;
+
+               /* If we emptied the pte, walk back up the ladder */
+               if (pte_empty(pte)) {
+                       clear_pmd_entry(pmd);
+                       range = PMD_SIZE;
+                       if (pmd_empty(pmd)) {
+                               clear_pud_entry(pud);
+                               range = PUD_SIZE;
+                       }
+               }
+
+               addr += range;
+       }
 }
 
 /**
- * free_hyp_pmds - free a Hyp-mode level-2 tables and child level-3 tables
+ * free_hyp_pgds - free Hyp-mode page tables
  *
- * Assumes this is a page table used strictly in Hyp-mode and therefore contains
+ * Assumes hyp_pgd is a page table used strictly in Hyp-mode and therefore contains
  * either mappings in the kernel memory area (above PAGE_OFFSET), or
  * device mappings in the vmalloc range (from VMALLOC_START to VMALLOC_END).
  */
-void free_hyp_pmds(void)
+void free_hyp_pgds(void)
 {
        unsigned long addr;
 
        mutex_lock(&kvm_hyp_pgd_mutex);
-       for (addr = PAGE_OFFSET; virt_addr_valid(addr); addr += PGDIR_SIZE)
-               free_hyp_pgd_entry(addr);
-       for (addr = VMALLOC_START; is_vmalloc_addr((void*)addr); addr += PGDIR_SIZE)
-               free_hyp_pgd_entry(addr);
+
+       if (hyp_pgd) {
+               for (addr = PAGE_OFFSET; virt_addr_valid(addr); addr += PGDIR_SIZE)
+                       unmap_range(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);
+               kfree(hyp_pgd);
+       }
+
        mutex_unlock(&kvm_hyp_pgd_mutex);
 }
 
@@ -136,6 +184,7 @@ static void create_hyp_pte_mappings(pmd_t *pmd, unsigned long start,
        do {
                pte = pte_offset_kernel(pmd, addr);
                kvm_set_pte(pte, pfn_pte(pfn, prot));
+               get_page(virt_to_page(pte));
                pfn++;
        } while (addr += PAGE_SIZE, addr != end);
 }
@@ -161,6 +210,7 @@ static int create_hyp_pmd_mappings(pud_t *pud, unsigned long start,
                                return -ENOMEM;
                        }
                        pmd_populate_kernel(NULL, pmd, pte);
+                       get_page(virt_to_page(pmd));
                }
 
                next = pmd_addr_end(addr, end);
@@ -197,6 +247,7 @@ static int __create_hyp_mappings(pgd_t *pgdp,
                                goto out;
                        }
                        pud_populate(NULL, pud, pmd);
+                       get_page(virt_to_page(pud));
                }
 
                next = pgd_addr_end(addr, end);
@@ -289,42 +340,6 @@ int kvm_alloc_stage2_pgd(struct kvm *kvm)
        return 0;
 }
 
-static void clear_pud_entry(pud_t *pud)
-{
-       pmd_t *pmd_table = pmd_offset(pud, 0);
-       pud_clear(pud);
-       pmd_free(NULL, pmd_table);
-       put_page(virt_to_page(pud));
-}
-
-static void clear_pmd_entry(pmd_t *pmd)
-{
-       pte_t *pte_table = pte_offset_kernel(pmd, 0);
-       pmd_clear(pmd);
-       pte_free_kernel(NULL, pte_table);
-       put_page(virt_to_page(pmd));
-}
-
-static bool pmd_empty(pmd_t *pmd)
-{
-       struct page *pmd_page = virt_to_page(pmd);
-       return page_count(pmd_page) == 1;
-}
-
-static void clear_pte_entry(pte_t *pte)
-{
-       if (pte_present(*pte)) {
-               kvm_set_pte(pte, __pte(0));
-               put_page(virt_to_page(pte));
-       }
-}
-
-static bool pte_empty(pte_t *pte)
-{
-       struct page *pte_page = virt_to_page(pte);
-       return page_count(pte_page) == 1;
-}
-
 /**
  * unmap_stage2_range -- Clear stage2 page table entries to unmap a range
  * @kvm:   The VM pointer
@@ -338,43 +353,7 @@ static bool pte_empty(pte_t *pte)
  */
 static void unmap_stage2_range(struct kvm *kvm, phys_addr_t start, u64 size)
 {
-       pgd_t *pgd;
-       pud_t *pud;
-       pmd_t *pmd;
-       pte_t *pte;
-       phys_addr_t addr = start, end = start + size;
-       u64 range;
-
-       while (addr < end) {
-               pgd = kvm->arch.pgd + pgd_index(addr);
-               pud = pud_offset(pgd, addr);
-               if (pud_none(*pud)) {
-                       addr += PUD_SIZE;
-                       continue;
-               }
-
-               pmd = pmd_offset(pud, addr);
-               if (pmd_none(*pmd)) {
-                       addr += PMD_SIZE;
-                       continue;
-               }
-
-               pte = pte_offset_kernel(pmd, addr);
-               clear_pte_entry(pte);
-               range = PAGE_SIZE;
-
-               /* If we emptied the pte, walk back up the ladder */
-               if (pte_empty(pte)) {
-                       clear_pmd_entry(pmd);
-                       range = PMD_SIZE;
-                       if (pmd_empty(pmd)) {
-                               clear_pud_entry(pud);
-                               range = PUD_SIZE;
-                       }
-               }
-
-               addr += range;
-       }
+       unmap_range(kvm->arch.pgd, start, size);
 }
 
 /**
@@ -741,7 +720,7 @@ int kvm_mmu_init(void)
 
        return 0;
 out:
-       kfree(hyp_pgd);
+       free_hyp_pgds();
        return err;
 }