vserver 2.0 rc7
[linux-2.6.git] / mm / vmalloc.c
index e6516c2..8ff16a1 100644 (file)
 DEFINE_RWLOCK(vmlist_lock);
 struct vm_struct *vmlist;
 
-static void unmap_area_pte(pmd_t *pmd, unsigned long address,
-                                 unsigned long size)
+static void vunmap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end)
 {
-       unsigned long end;
        pte_t *pte;
 
-       if (pmd_none(*pmd))
-               return;
-       if (pmd_bad(*pmd)) {
-               pmd_ERROR(*pmd);
-               pmd_clear(pmd);
-               return;
-       }
-
-       pte = pte_offset_kernel(pmd, address);
-       address &= ~PMD_MASK;
-       end = address + size;
-       if (end > PMD_SIZE)
-               end = PMD_SIZE;
-
+       pte = pte_offset_kernel(pmd, addr);
        do {
-               pte_t page;
-               page = ptep_get_and_clear(pte);
-               address += PAGE_SIZE;
-               pte++;
-               if (pte_none(page))
-                       continue;
-               if (pte_present(page))
-                       continue;
-               printk(KERN_CRIT "Whee.. Swapped out page in kernel page table\n");
-       } while (address < end);
+               pte_t ptent = ptep_get_and_clear(&init_mm, addr, pte);
+               WARN_ON(!pte_none(ptent) && !pte_present(ptent));
+       } while (pte++, addr += PAGE_SIZE, addr != end);
 }
 
-static void unmap_area_pmd(pud_t *pud, unsigned long address,
-                                 unsigned long size)
+static inline void vunmap_pmd_range(pud_t *pud, unsigned long addr,
+                                               unsigned long end)
 {
-       unsigned long end;
        pmd_t *pmd;
+       unsigned long next;
 
-       if (pud_none(*pud))
-               return;
-       if (pud_bad(*pud)) {
-               pud_ERROR(*pud);
-               pud_clear(pud);
-               return;
-       }
-
-       pmd = pmd_offset(pud, address);
-       address &= ~PUD_MASK;
-       end = address + size;
-       if (end > PUD_SIZE)
-               end = PUD_SIZE;
-
+       pmd = pmd_offset(pud, addr);
        do {
-               unmap_area_pte(pmd, address, end - address);
-               address = (address + PMD_SIZE) & PMD_MASK;
-               pmd++;
-       } while (address < end);
+               next = pmd_addr_end(addr, end);
+               if (pmd_none_or_clear_bad(pmd))
+                       continue;
+               vunmap_pte_range(pmd, addr, next);
+       } while (pmd++, addr = next, addr != end);
 }
 
-static void unmap_area_pud(pgd_t *pgd, unsigned long address,
-                          unsigned long size)
+static inline void vunmap_pud_range(pgd_t *pgd, unsigned long addr,
+                                               unsigned long end)
 {
        pud_t *pud;
-       unsigned long end;
+       unsigned long next;
 
-       if (pgd_none(*pgd))
-               return;
-       if (pgd_bad(*pgd)) {
-               pgd_ERROR(*pgd);
-               pgd_clear(pgd);
-               return;
-       }
+       pud = pud_offset(pgd, addr);
+       do {
+               next = pud_addr_end(addr, end);
+               if (pud_none_or_clear_bad(pud))
+                       continue;
+               vunmap_pmd_range(pud, addr, next);
+       } while (pud++, addr = next, addr != end);
+}
 
-       pud = pud_offset(pgd, address);
-       address &= ~PGDIR_MASK;
-       end = address + size;
-       if (end > PGDIR_SIZE)
-               end = PGDIR_SIZE;
+void unmap_vm_area(struct vm_struct *area)
+{
+       pgd_t *pgd;
+       unsigned long next;
+       unsigned long addr = (unsigned long) area->addr;
+       unsigned long end = addr + area->size;
 
+       BUG_ON(addr >= end);
+       pgd = pgd_offset_k(addr);
+       flush_cache_vunmap(addr, end);
        do {
-               unmap_area_pmd(pud, address, end - address);
-               address = (address + PUD_SIZE) & PUD_MASK;
-               pud++;
-       } while (address && (address < end));
+               next = pgd_addr_end(addr, end);
+               if (pgd_none_or_clear_bad(pgd))
+                       continue;
+               vunmap_pud_range(pgd, addr, next);
+       } while (pgd++, addr = next, addr != end);
+       flush_tlb_kernel_range((unsigned long) area->addr, end);
 }
 
-static int map_area_pte(pte_t *pte, unsigned long address,
-                              unsigned long size, pgprot_t prot,
-                              struct page ***pages)
+static int vmap_pte_range(pmd_t *pmd, unsigned long addr,
+                       unsigned long end, pgprot_t prot, struct page ***pages)
 {
-       unsigned long end;
-
-       address &= ~PMD_MASK;
-       end = address + size;
-       if (end > PMD_SIZE)
-               end = PMD_SIZE;
+       pte_t *pte;
 
+       pte = pte_alloc_kernel(&init_mm, pmd, addr);
+       if (!pte)
+               return -ENOMEM;
        do {
                struct page *page = **pages;
                WARN_ON(!pte_none(*pte));
                if (!page)
                        return -ENOMEM;
-
-               set_pte(pte, mk_pte(page, prot));
-               address += PAGE_SIZE;
-               pte++;
+               set_pte_at(&init_mm, addr, pte, mk_pte(page, prot));
                (*pages)++;
-       } while (address < end);
+       } while (pte++, addr += PAGE_SIZE, addr != end);
        return 0;
 }
 
-static int map_area_pmd(pmd_t *pmd, unsigned long address,
-                              unsigned long size, pgprot_t prot,
-                              struct page ***pages)
+static inline int vmap_pmd_range(pud_t *pud, unsigned long addr,
+                       unsigned long end, pgprot_t prot, struct page ***pages)
 {
-       unsigned long base, end;
-
-       base = address & PUD_MASK;
-       address &= ~PUD_MASK;
-       end = address + size;
-       if (end > PUD_SIZE)
-               end = PUD_SIZE;
+       pmd_t *pmd;
+       unsigned long next;
 
+       pmd = pmd_alloc(&init_mm, pud, addr);
+       if (!pmd)
+               return -ENOMEM;
        do {
-               pte_t * pte = pte_alloc_kernel(&init_mm, pmd, base + address);
-               if (!pte)
-                       return -ENOMEM;
-               if (map_area_pte(pte, address, end - address, prot, pages))
+               next = pmd_addr_end(addr, end);
+               if (vmap_pte_range(pmd, addr, next, prot, pages))
                        return -ENOMEM;
-               address = (address + PMD_SIZE) & PMD_MASK;
-               pmd++;
-       } while (address < end);
-
+       } while (pmd++, addr = next, addr != end);
        return 0;
 }
 
-static int map_area_pud(pud_t *pud, unsigned long address,
-                              unsigned long end, pgprot_t prot,
-                              struct page ***pages)
+static inline int vmap_pud_range(pgd_t *pgd, unsigned long addr,
+                       unsigned long end, pgprot_t prot, struct page ***pages)
 {
+       pud_t *pud;
+       unsigned long next;
+
+       pud = pud_alloc(&init_mm, pgd, addr);
+       if (!pud)
+               return -ENOMEM;
        do {
-               pmd_t *pmd = pmd_alloc(&init_mm, pud, address);
-               if (!pmd)
+               next = pud_addr_end(addr, end);
+               if (vmap_pmd_range(pud, addr, next, prot, pages))
                        return -ENOMEM;
-               if (map_area_pmd(pmd, address, end - address, prot, pages))
-                       return -ENOMEM;
-               address = (address + PUD_SIZE) & PUD_MASK;
-               pud++;
-       } while (address && address < end);
-
+       } while (pud++, addr = next, addr != end);
        return 0;
 }
 
-void unmap_vm_area(struct vm_struct *area)
-{
-       unsigned long address = (unsigned long) area->addr;
-       unsigned long end = (address + area->size);
-       unsigned long next;
-       pgd_t *pgd;
-       int i;
-
-       pgd = pgd_offset_k(address);
-       flush_cache_vunmap(address, end);
-       for (i = pgd_index(address); i <= pgd_index(end-1); i++) {
-               next = (address + PGDIR_SIZE) & PGDIR_MASK;
-               if (next <= address || next > end)
-                       next = end;
-               unmap_area_pud(pgd, address, next - address);
-               address = next;
-               pgd++;
-       }
-       flush_tlb_kernel_range((unsigned long) area->addr, end);
-}
-
 int map_vm_area(struct vm_struct *area, pgprot_t prot, struct page ***pages)
 {
-       unsigned long address = (unsigned long) area->addr;
-       unsigned long end = address + (area->size-PAGE_SIZE);
-       unsigned long next;
        pgd_t *pgd;
-       int err = 0;
-       int i;
+       unsigned long next;
+       unsigned long addr = (unsigned long) area->addr;
+       unsigned long end = addr + area->size - PAGE_SIZE;
+       int err;
 
-       pgd = pgd_offset_k(address);
+       BUG_ON(addr >= end);
+       pgd = pgd_offset_k(addr);
        spin_lock(&init_mm.page_table_lock);
-       for (i = pgd_index(address); i <= pgd_index(end-1); i++) {
-               pud_t *pud = pud_alloc(&init_mm, pgd, address);
-               if (!pud) {
-                       err = -ENOMEM;
-                       break;
-               }
-               next = (address + PGDIR_SIZE) & PGDIR_MASK;
-               if (next < address || next > end)
-                       next = end;
-               if (map_area_pud(pud, address, next, prot, pages)) {
-                       err = -ENOMEM;
+       do {
+               next = pgd_addr_end(addr, end);
+               err = vmap_pud_range(pgd, addr, next, prot, pages);
+               if (err)
                        break;
-               }
-
-               address = next;
-               pgd++;
-       }
-
+       } while (pgd++, addr = next, addr != end);
        spin_unlock(&init_mm.page_table_lock);
        flush_cache_vmap((unsigned long) area->addr, end);
        return err;
@@ -252,20 +178,22 @@ struct vm_struct *__get_vm_area(unsigned long size, unsigned long flags,
                align = 1ul << bit;
        }
        addr = ALIGN(start, align);
+       size = PAGE_ALIGN(size);
 
        area = kmalloc(sizeof(*area), GFP_KERNEL);
        if (unlikely(!area))
                return NULL;
 
-       /*
-        * We always allocate a guard page.
-        */
-       size += PAGE_SIZE;
        if (unlikely(!size)) {
                kfree (area);
                return NULL;
        }
 
+       /*
+        * We always allocate a guard page.
+        */
+       size += PAGE_SIZE;
+
        write_lock(&vmlist_lock);
        for (p = &vmlist; (tmp = *p) != NULL ;p = &tmp->next) {
                if ((unsigned long)tmp->addr < addr) {
@@ -320,34 +248,46 @@ struct vm_struct *get_vm_area(unsigned long size, unsigned long flags)
        return __get_vm_area(size, flags, VMALLOC_START, VMALLOC_END);
 }
 
-/**
- *     remove_vm_area  -  find and remove a contingous kernel virtual area
- *
- *     @addr:          base address
- *
- *     Search for the kernel VM area starting at @addr, and remove it.
- *     This function returns the found VM area, but using it is NOT safe
- *     on SMP machines.
- */
-struct vm_struct *remove_vm_area(void *addr)
+/* Caller must hold vmlist_lock */
+struct vm_struct *__remove_vm_area(void *addr)
 {
        struct vm_struct **p, *tmp;
 
-       write_lock(&vmlist_lock);
        for (p = &vmlist ; (tmp = *p) != NULL ;p = &tmp->next) {
                 if (tmp->addr == addr)
                         goto found;
        }
-       write_unlock(&vmlist_lock);
        return NULL;
 
 found:
        unmap_vm_area(tmp);
        *p = tmp->next;
-       write_unlock(&vmlist_lock);
+
+       /*
+        * Remove the guard page.
+        */
+       tmp->size -= PAGE_SIZE;
        return tmp;
 }
 
+/**
+ *     remove_vm_area  -  find and remove a contingous kernel virtual area
+ *
+ *     @addr:          base address
+ *
+ *     Search for the kernel VM area starting at @addr, and remove it.
+ *     This function returns the found VM area, but using it is NOT safe
+ *     on SMP machines, except for its size or flags.
+ */
+struct vm_struct *remove_vm_area(void *addr)
+{
+       struct vm_struct *v;
+       write_lock(&vmlist_lock);
+       v = __remove_vm_area(addr);
+       write_unlock(&vmlist_lock);
+       return v;
+}
+
 void __vunmap(void *addr, int deallocate_pages)
 {
        struct vm_struct *area;
@@ -368,7 +308,7 @@ void __vunmap(void *addr, int deallocate_pages)
                WARN_ON(1);
                return;
        }
-       
+
        if (deallocate_pages) {
                int i;
 
@@ -456,32 +396,12 @@ void *vmap(struct page **pages, unsigned int count,
 
 EXPORT_SYMBOL(vmap);
 
-/**
- *     __vmalloc  -  allocate virtually contiguous memory
- *
- *     @size:          allocation size
- *     @gfp_mask:      flags for the page level allocator
- *     @prot:          protection mask for the allocated pages
- *
- *     Allocate enough pages to cover @size from the page level
- *     allocator with @gfp_mask flags.  Map them into contiguous
- *     kernel virtual space, using a pagetable protection of @prot.
- */
-void *__vmalloc(unsigned long size, int gfp_mask, pgprot_t prot)
+void *__vmalloc_area(struct vm_struct *area, unsigned int __nocast gfp_mask, pgprot_t prot)
 {
-       struct vm_struct *area;
        struct page **pages;
        unsigned int nr_pages, array_size, i;
 
-       size = PAGE_ALIGN(size);
-       if (!size || (size >> PAGE_SHIFT) > num_physpages)
-               return NULL;
-
-       area = get_vm_area(size, VM_ALLOC);
-       if (!area)
-               return NULL;
-
-       nr_pages = size >> PAGE_SHIFT;
+       nr_pages = (area->size - PAGE_SIZE) >> PAGE_SHIFT;
        array_size = (nr_pages * sizeof(struct page *));
 
        area->nr_pages = nr_pages;
@@ -506,7 +426,7 @@ void *__vmalloc(unsigned long size, int gfp_mask, pgprot_t prot)
                        goto fail;
                }
        }
-       
+
        if (map_vm_area(area, prot, &pages))
                goto fail;
        return area->addr;
@@ -516,6 +436,32 @@ fail:
        return NULL;
 }
 
+/**
+ *     __vmalloc  -  allocate virtually contiguous memory
+ *
+ *     @size:          allocation size
+ *     @gfp_mask:      flags for the page level allocator
+ *     @prot:          protection mask for the allocated pages
+ *
+ *     Allocate enough pages to cover @size from the page level
+ *     allocator with @gfp_mask flags.  Map them into contiguous
+ *     kernel virtual space, using a pagetable protection of @prot.
+ */
+void *__vmalloc(unsigned long size, unsigned int __nocast gfp_mask, pgprot_t prot)
+{
+       struct vm_struct *area;
+
+       size = PAGE_ALIGN(size);
+       if (!size || (size >> PAGE_SHIFT) > num_physpages)
+               return NULL;
+
+       area = get_vm_area(size, VM_ALLOC);
+       if (!area)
+               return NULL;
+
+       return __vmalloc_area(area, gfp_mask, prot);
+}
+
 EXPORT_SYMBOL(__vmalloc);
 
 /**
@@ -536,6 +482,10 @@ void *vmalloc(unsigned long size)
 
 EXPORT_SYMBOL(vmalloc);
 
+#ifndef PAGE_KERNEL_EXEC
+# define PAGE_KERNEL_EXEC PAGE_KERNEL
+#endif
+
 /**
  *     vmalloc_exec  -  allocate virtually contiguous, executable memory
  *
@@ -549,10 +499,6 @@ EXPORT_SYMBOL(vmalloc);
  *     use __vmalloc() instead.
  */
 
-#ifndef PAGE_KERNEL_EXEC
-# define PAGE_KERNEL_EXEC PAGE_KERNEL
-#endif
-
 void *vmalloc_exec(unsigned long size)
 {
        return __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM, PAGE_KERNEL_EXEC);