i386: prepare shared mm/pageattr.c
authorThomas Gleixner <tglx@linutronix.de>
Thu, 11 Oct 2007 09:13:56 +0000 (11:13 +0200)
committerThomas Gleixner <tglx@linutronix.de>
Thu, 11 Oct 2007 09:13:56 +0000 (11:13 +0200)
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
arch/i386/mm/Makefile
arch/i386/mm/pageattr.c [deleted file]
arch/i386/mm/pageattr_32.c [new file with mode: 0644]

index 2c81ad14a62bd22e3d5b3b1f5106afbc37a80da4..4d7c5af4765d1eef065e7be2cf1b3b123bb64f06 100644 (file)
@@ -2,7 +2,7 @@
 # Makefile for the linux i386-specific parts of the memory manager.
 #
 
-obj-y  := init_32.o pgtable_32.o fault.o ioremap_32.o extable_32.o pageattr.o mmap_32.o
+obj-y  := init_32.o pgtable_32.o fault.o ioremap_32.o extable_32.o pageattr_32.o mmap_32.o
 
 obj-$(CONFIG_NUMA) += discontig_32.o
 obj-$(CONFIG_HUGETLB_PAGE) += hugetlbpage.o
diff --git a/arch/i386/mm/pageattr.c b/arch/i386/mm/pageattr.c
deleted file mode 100644 (file)
index 4241a74..0000000
+++ /dev/null
@@ -1,278 +0,0 @@
-/* 
- * Copyright 2002 Andi Kleen, SuSE Labs. 
- * Thanks to Ben LaHaise for precious feedback.
- */ 
-
-#include <linux/mm.h>
-#include <linux/sched.h>
-#include <linux/highmem.h>
-#include <linux/module.h>
-#include <linux/slab.h>
-#include <asm/uaccess.h>
-#include <asm/processor.h>
-#include <asm/tlbflush.h>
-#include <asm/pgalloc.h>
-#include <asm/sections.h>
-
-static DEFINE_SPINLOCK(cpa_lock);
-static struct list_head df_list = LIST_HEAD_INIT(df_list);
-
-
-pte_t *lookup_address(unsigned long address) 
-{ 
-       pgd_t *pgd = pgd_offset_k(address);
-       pud_t *pud;
-       pmd_t *pmd;
-       if (pgd_none(*pgd))
-               return NULL;
-       pud = pud_offset(pgd, address);
-       if (pud_none(*pud))
-               return NULL;
-       pmd = pmd_offset(pud, address);
-       if (pmd_none(*pmd))
-               return NULL;
-       if (pmd_large(*pmd))
-               return (pte_t *)pmd;
-        return pte_offset_kernel(pmd, address);
-} 
-
-static struct page *split_large_page(unsigned long address, pgprot_t prot,
-                                       pgprot_t ref_prot)
-{ 
-       int i; 
-       unsigned long addr;
-       struct page *base;
-       pte_t *pbase;
-
-       spin_unlock_irq(&cpa_lock);
-       base = alloc_pages(GFP_KERNEL, 0);
-       spin_lock_irq(&cpa_lock);
-       if (!base) 
-               return NULL;
-
-       /*
-        * page_private is used to track the number of entries in
-        * the page table page that have non standard attributes.
-        */
-       SetPagePrivate(base);
-       page_private(base) = 0;
-
-       address = __pa(address);
-       addr = address & LARGE_PAGE_MASK; 
-       pbase = (pte_t *)page_address(base);
-       paravirt_alloc_pt(&init_mm, page_to_pfn(base));
-       for (i = 0; i < PTRS_PER_PTE; i++, addr += PAGE_SIZE) {
-               set_pte(&pbase[i], pfn_pte(addr >> PAGE_SHIFT,
-                                          addr == address ? prot : ref_prot));
-       }
-       return base;
-} 
-
-static void cache_flush_page(struct page *p)
-{ 
-       unsigned long adr = (unsigned long)page_address(p);
-       int i;
-       for (i = 0; i < PAGE_SIZE; i += boot_cpu_data.x86_clflush_size)
-               asm volatile("clflush (%0)" :: "r" (adr + i));
-}
-
-static void flush_kernel_map(void *arg)
-{
-       struct list_head *lh = (struct list_head *)arg;
-       struct page *p;
-
-       /* High level code is not ready for clflush yet */
-       if (0 && cpu_has_clflush) {
-               list_for_each_entry (p, lh, lru)
-                       cache_flush_page(p);
-       } else if (boot_cpu_data.x86_model >= 4)
-               wbinvd();
-
-       /* Flush all to work around Errata in early athlons regarding 
-        * large page flushing. 
-        */
-       __flush_tlb_all();      
-}
-
-static void set_pmd_pte(pte_t *kpte, unsigned long address, pte_t pte) 
-{ 
-       struct page *page;
-       unsigned long flags;
-
-       set_pte_atomic(kpte, pte);      /* change init_mm */
-       if (SHARED_KERNEL_PMD)
-               return;
-
-       spin_lock_irqsave(&pgd_lock, flags);
-       for (page = pgd_list; page; page = (struct page *)page->index) {
-               pgd_t *pgd;
-               pud_t *pud;
-               pmd_t *pmd;
-               pgd = (pgd_t *)page_address(page) + pgd_index(address);
-               pud = pud_offset(pgd, address);
-               pmd = pmd_offset(pud, address);
-               set_pte_atomic((pte_t *)pmd, pte);
-       }
-       spin_unlock_irqrestore(&pgd_lock, flags);
-}
-
-/* 
- * No more special protections in this 2/4MB area - revert to a
- * large page again. 
- */
-static inline void revert_page(struct page *kpte_page, unsigned long address)
-{
-       pgprot_t ref_prot;
-       pte_t *linear;
-
-       ref_prot =
-       ((address & LARGE_PAGE_MASK) < (unsigned long)&_etext)
-               ? PAGE_KERNEL_LARGE_EXEC : PAGE_KERNEL_LARGE;
-
-       linear = (pte_t *)
-               pmd_offset(pud_offset(pgd_offset_k(address), address), address);
-       set_pmd_pte(linear,  address,
-                   pfn_pte((__pa(address) & LARGE_PAGE_MASK) >> PAGE_SHIFT,
-                           ref_prot));
-}
-
-static inline void save_page(struct page *kpte_page)
-{
-       if (!test_and_set_bit(PG_arch_1, &kpte_page->flags))
-               list_add(&kpte_page->lru, &df_list);
-}
-
-static int
-__change_page_attr(struct page *page, pgprot_t prot)
-{ 
-       pte_t *kpte; 
-       unsigned long address;
-       struct page *kpte_page;
-
-       BUG_ON(PageHighMem(page));
-       address = (unsigned long)page_address(page);
-
-       kpte = lookup_address(address);
-       if (!kpte)
-               return -EINVAL;
-       kpte_page = virt_to_page(kpte);
-       BUG_ON(PageLRU(kpte_page));
-       BUG_ON(PageCompound(kpte_page));
-
-       if (pgprot_val(prot) != pgprot_val(PAGE_KERNEL)) { 
-               if (!pte_huge(*kpte)) {
-                       set_pte_atomic(kpte, mk_pte(page, prot)); 
-               } else {
-                       pgprot_t ref_prot;
-                       struct page *split;
-
-                       ref_prot =
-                       ((address & LARGE_PAGE_MASK) < (unsigned long)&_etext)
-                               ? PAGE_KERNEL_EXEC : PAGE_KERNEL;
-                       split = split_large_page(address, prot, ref_prot);
-                       if (!split)
-                               return -ENOMEM;
-                       set_pmd_pte(kpte,address,mk_pte(split, ref_prot));
-                       kpte_page = split;
-               }
-               page_private(kpte_page)++;
-       } else if (!pte_huge(*kpte)) {
-               set_pte_atomic(kpte, mk_pte(page, PAGE_KERNEL));
-               BUG_ON(page_private(kpte_page) == 0);
-               page_private(kpte_page)--;
-       } else
-               BUG();
-
-       /*
-        * If the pte was reserved, it means it was created at boot
-        * time (not via split_large_page) and in turn we must not
-        * replace it with a largepage.
-        */
-
-       save_page(kpte_page);
-       if (!PageReserved(kpte_page)) {
-               if (cpu_has_pse && (page_private(kpte_page) == 0)) {
-                       paravirt_release_pt(page_to_pfn(kpte_page));
-                       revert_page(kpte_page, address);
-               }
-       }
-       return 0;
-} 
-
-static inline void flush_map(struct list_head *l)
-{
-       on_each_cpu(flush_kernel_map, l, 1, 1);
-}
-
-/*
- * Change the page attributes of an page in the linear mapping.
- *
- * This should be used when a page is mapped with a different caching policy
- * than write-back somewhere - some CPUs do not like it when mappings with
- * different caching policies exist. This changes the page attributes of the
- * in kernel linear mapping too.
- * 
- * The caller needs to ensure that there are no conflicting mappings elsewhere.
- * This function only deals with the kernel linear map.
- * 
- * Caller must call global_flush_tlb() after this.
- */
-int change_page_attr(struct page *page, int numpages, pgprot_t prot)
-{
-       int err = 0; 
-       int i; 
-       unsigned long flags;
-
-       spin_lock_irqsave(&cpa_lock, flags);
-       for (i = 0; i < numpages; i++, page++) { 
-               err = __change_page_attr(page, prot);
-               if (err) 
-                       break; 
-       }       
-       spin_unlock_irqrestore(&cpa_lock, flags);
-       return err;
-}
-
-void global_flush_tlb(void)
-{
-       struct list_head l;
-       struct page *pg, *next;
-
-       BUG_ON(irqs_disabled());
-
-       spin_lock_irq(&cpa_lock);
-       list_replace_init(&df_list, &l);
-       spin_unlock_irq(&cpa_lock);
-       flush_map(&l);
-       list_for_each_entry_safe(pg, next, &l, lru) {
-               list_del(&pg->lru);
-               clear_bit(PG_arch_1, &pg->flags);
-               if (PageReserved(pg) || !cpu_has_pse || page_private(pg) != 0)
-                       continue;
-               ClearPagePrivate(pg);
-               __free_page(pg);
-       }
-}
-
-#ifdef CONFIG_DEBUG_PAGEALLOC
-void kernel_map_pages(struct page *page, int numpages, int enable)
-{
-       if (PageHighMem(page))
-               return;
-       if (!enable)
-               debug_check_no_locks_freed(page_address(page),
-                                          numpages * PAGE_SIZE);
-
-       /* the return value is ignored - the calls cannot fail,
-        * large pages are disabled at boot time.
-        */
-       change_page_attr(page, numpages, enable ? PAGE_KERNEL : __pgprot(0));
-       /* we should perform an IPI and flush all tlbs,
-        * but that can deadlock->flush only current cpu.
-        */
-       __flush_tlb_all();
-}
-#endif
-
-EXPORT_SYMBOL(change_page_attr);
-EXPORT_SYMBOL(global_flush_tlb);
diff --git a/arch/i386/mm/pageattr_32.c b/arch/i386/mm/pageattr_32.c
new file mode 100644 (file)
index 0000000..4241a74
--- /dev/null
@@ -0,0 +1,278 @@
+/* 
+ * Copyright 2002 Andi Kleen, SuSE Labs. 
+ * Thanks to Ben LaHaise for precious feedback.
+ */ 
+
+#include <linux/mm.h>
+#include <linux/sched.h>
+#include <linux/highmem.h>
+#include <linux/module.h>
+#include <linux/slab.h>
+#include <asm/uaccess.h>
+#include <asm/processor.h>
+#include <asm/tlbflush.h>
+#include <asm/pgalloc.h>
+#include <asm/sections.h>
+
+static DEFINE_SPINLOCK(cpa_lock);
+static struct list_head df_list = LIST_HEAD_INIT(df_list);
+
+
+pte_t *lookup_address(unsigned long address) 
+{ 
+       pgd_t *pgd = pgd_offset_k(address);
+       pud_t *pud;
+       pmd_t *pmd;
+       if (pgd_none(*pgd))
+               return NULL;
+       pud = pud_offset(pgd, address);
+       if (pud_none(*pud))
+               return NULL;
+       pmd = pmd_offset(pud, address);
+       if (pmd_none(*pmd))
+               return NULL;
+       if (pmd_large(*pmd))
+               return (pte_t *)pmd;
+        return pte_offset_kernel(pmd, address);
+} 
+
+static struct page *split_large_page(unsigned long address, pgprot_t prot,
+                                       pgprot_t ref_prot)
+{ 
+       int i; 
+       unsigned long addr;
+       struct page *base;
+       pte_t *pbase;
+
+       spin_unlock_irq(&cpa_lock);
+       base = alloc_pages(GFP_KERNEL, 0);
+       spin_lock_irq(&cpa_lock);
+       if (!base) 
+               return NULL;
+
+       /*
+        * page_private is used to track the number of entries in
+        * the page table page that have non standard attributes.
+        */
+       SetPagePrivate(base);
+       page_private(base) = 0;
+
+       address = __pa(address);
+       addr = address & LARGE_PAGE_MASK; 
+       pbase = (pte_t *)page_address(base);
+       paravirt_alloc_pt(&init_mm, page_to_pfn(base));
+       for (i = 0; i < PTRS_PER_PTE; i++, addr += PAGE_SIZE) {
+               set_pte(&pbase[i], pfn_pte(addr >> PAGE_SHIFT,
+                                          addr == address ? prot : ref_prot));
+       }
+       return base;
+} 
+
+static void cache_flush_page(struct page *p)
+{ 
+       unsigned long adr = (unsigned long)page_address(p);
+       int i;
+       for (i = 0; i < PAGE_SIZE; i += boot_cpu_data.x86_clflush_size)
+               asm volatile("clflush (%0)" :: "r" (adr + i));
+}
+
+static void flush_kernel_map(void *arg)
+{
+       struct list_head *lh = (struct list_head *)arg;
+       struct page *p;
+
+       /* High level code is not ready for clflush yet */
+       if (0 && cpu_has_clflush) {
+               list_for_each_entry (p, lh, lru)
+                       cache_flush_page(p);
+       } else if (boot_cpu_data.x86_model >= 4)
+               wbinvd();
+
+       /* Flush all to work around Errata in early athlons regarding 
+        * large page flushing. 
+        */
+       __flush_tlb_all();      
+}
+
+static void set_pmd_pte(pte_t *kpte, unsigned long address, pte_t pte) 
+{ 
+       struct page *page;
+       unsigned long flags;
+
+       set_pte_atomic(kpte, pte);      /* change init_mm */
+       if (SHARED_KERNEL_PMD)
+               return;
+
+       spin_lock_irqsave(&pgd_lock, flags);
+       for (page = pgd_list; page; page = (struct page *)page->index) {
+               pgd_t *pgd;
+               pud_t *pud;
+               pmd_t *pmd;
+               pgd = (pgd_t *)page_address(page) + pgd_index(address);
+               pud = pud_offset(pgd, address);
+               pmd = pmd_offset(pud, address);
+               set_pte_atomic((pte_t *)pmd, pte);
+       }
+       spin_unlock_irqrestore(&pgd_lock, flags);
+}
+
+/* 
+ * No more special protections in this 2/4MB area - revert to a
+ * large page again. 
+ */
+static inline void revert_page(struct page *kpte_page, unsigned long address)
+{
+       pgprot_t ref_prot;
+       pte_t *linear;
+
+       ref_prot =
+       ((address & LARGE_PAGE_MASK) < (unsigned long)&_etext)
+               ? PAGE_KERNEL_LARGE_EXEC : PAGE_KERNEL_LARGE;
+
+       linear = (pte_t *)
+               pmd_offset(pud_offset(pgd_offset_k(address), address), address);
+       set_pmd_pte(linear,  address,
+                   pfn_pte((__pa(address) & LARGE_PAGE_MASK) >> PAGE_SHIFT,
+                           ref_prot));
+}
+
+static inline void save_page(struct page *kpte_page)
+{
+       if (!test_and_set_bit(PG_arch_1, &kpte_page->flags))
+               list_add(&kpte_page->lru, &df_list);
+}
+
+static int
+__change_page_attr(struct page *page, pgprot_t prot)
+{ 
+       pte_t *kpte; 
+       unsigned long address;
+       struct page *kpte_page;
+
+       BUG_ON(PageHighMem(page));
+       address = (unsigned long)page_address(page);
+
+       kpte = lookup_address(address);
+       if (!kpte)
+               return -EINVAL;
+       kpte_page = virt_to_page(kpte);
+       BUG_ON(PageLRU(kpte_page));
+       BUG_ON(PageCompound(kpte_page));
+
+       if (pgprot_val(prot) != pgprot_val(PAGE_KERNEL)) { 
+               if (!pte_huge(*kpte)) {
+                       set_pte_atomic(kpte, mk_pte(page, prot)); 
+               } else {
+                       pgprot_t ref_prot;
+                       struct page *split;
+
+                       ref_prot =
+                       ((address & LARGE_PAGE_MASK) < (unsigned long)&_etext)
+                               ? PAGE_KERNEL_EXEC : PAGE_KERNEL;
+                       split = split_large_page(address, prot, ref_prot);
+                       if (!split)
+                               return -ENOMEM;
+                       set_pmd_pte(kpte,address,mk_pte(split, ref_prot));
+                       kpte_page = split;
+               }
+               page_private(kpte_page)++;
+       } else if (!pte_huge(*kpte)) {
+               set_pte_atomic(kpte, mk_pte(page, PAGE_KERNEL));
+               BUG_ON(page_private(kpte_page) == 0);
+               page_private(kpte_page)--;
+       } else
+               BUG();
+
+       /*
+        * If the pte was reserved, it means it was created at boot
+        * time (not via split_large_page) and in turn we must not
+        * replace it with a largepage.
+        */
+
+       save_page(kpte_page);
+       if (!PageReserved(kpte_page)) {
+               if (cpu_has_pse && (page_private(kpte_page) == 0)) {
+                       paravirt_release_pt(page_to_pfn(kpte_page));
+                       revert_page(kpte_page, address);
+               }
+       }
+       return 0;
+} 
+
+static inline void flush_map(struct list_head *l)
+{
+       on_each_cpu(flush_kernel_map, l, 1, 1);
+}
+
+/*
+ * Change the page attributes of an page in the linear mapping.
+ *
+ * This should be used when a page is mapped with a different caching policy
+ * than write-back somewhere - some CPUs do not like it when mappings with
+ * different caching policies exist. This changes the page attributes of the
+ * in kernel linear mapping too.
+ * 
+ * The caller needs to ensure that there are no conflicting mappings elsewhere.
+ * This function only deals with the kernel linear map.
+ * 
+ * Caller must call global_flush_tlb() after this.
+ */
+int change_page_attr(struct page *page, int numpages, pgprot_t prot)
+{
+       int err = 0; 
+       int i; 
+       unsigned long flags;
+
+       spin_lock_irqsave(&cpa_lock, flags);
+       for (i = 0; i < numpages; i++, page++) { 
+               err = __change_page_attr(page, prot);
+               if (err) 
+                       break; 
+       }       
+       spin_unlock_irqrestore(&cpa_lock, flags);
+       return err;
+}
+
+void global_flush_tlb(void)
+{
+       struct list_head l;
+       struct page *pg, *next;
+
+       BUG_ON(irqs_disabled());
+
+       spin_lock_irq(&cpa_lock);
+       list_replace_init(&df_list, &l);
+       spin_unlock_irq(&cpa_lock);
+       flush_map(&l);
+       list_for_each_entry_safe(pg, next, &l, lru) {
+               list_del(&pg->lru);
+               clear_bit(PG_arch_1, &pg->flags);
+               if (PageReserved(pg) || !cpu_has_pse || page_private(pg) != 0)
+                       continue;
+               ClearPagePrivate(pg);
+               __free_page(pg);
+       }
+}
+
+#ifdef CONFIG_DEBUG_PAGEALLOC
+void kernel_map_pages(struct page *page, int numpages, int enable)
+{
+       if (PageHighMem(page))
+               return;
+       if (!enable)
+               debug_check_no_locks_freed(page_address(page),
+                                          numpages * PAGE_SIZE);
+
+       /* the return value is ignored - the calls cannot fail,
+        * large pages are disabled at boot time.
+        */
+       change_page_attr(page, numpages, enable ? PAGE_KERNEL : __pgprot(0));
+       /* we should perform an IPI and flush all tlbs,
+        * but that can deadlock->flush only current cpu.
+        */
+       __flush_tlb_all();
+}
+#endif
+
+EXPORT_SYMBOL(change_page_attr);
+EXPORT_SYMBOL(global_flush_tlb);