x86: cpa: simplify locking
Ingo Molnar [Wed, 30 Jan 2008 12:33:57 +0000 (13:33 +0100)]
further simplify cpa locking: since the largepage-split is a
slowpath, use the pgd_lock for the whole operation, intead
of the mmap_sem.

This also makes it suitable for DEBUG_PAGEALLOC purposes again.

Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

arch/x86/mm/pageattr_32.c

index 0966023..9cf2fea 100644 (file)
@@ -37,9 +37,8 @@ pte_t *lookup_address(unsigned long address, int *level)
        return pte_offset_kernel(pmd, address);
 }
 
-static void set_pmd_pte(pte_t *kpte, unsigned long address, pte_t pte)
+static void __set_pmd_pte(pte_t *kpte, unsigned long address, pte_t pte)
 {
-       unsigned long flags;
        struct page *page;
 
        /* change init_mm */
@@ -47,7 +46,6 @@ static void set_pmd_pte(pte_t *kpte, unsigned long address, pte_t pte)
        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;
@@ -58,12 +56,12 @@ static void set_pmd_pte(pte_t *kpte, unsigned long address, pte_t pte)
                pmd = pmd_offset(pud, address);
                set_pte_atomic((pte_t *)pmd, pte);
        }
-       spin_unlock_irqrestore(&pgd_lock, flags);
 }
 
 static int split_large_page(pte_t *kpte, unsigned long address)
 {
        pgprot_t ref_prot = pte_pgprot(pte_clrhuge(*kpte));
+       unsigned long flags;
        unsigned long addr;
        pte_t *pbase, *tmp;
        struct page *base;
@@ -73,7 +71,7 @@ static int split_large_page(pte_t *kpte, unsigned long address)
        if (!base)
                return -ENOMEM;
 
-       down_write(&init_mm.mmap_sem);
+       spin_lock_irqsave(&pgd_lock, flags);
        /*
         * Check for races, another CPU might have split this page
         * up for us already:
@@ -95,11 +93,11 @@ static int split_large_page(pte_t *kpte, unsigned long address)
        /*
         * Install the new, split up pagetable:
         */
-       set_pmd_pte(kpte, address, mk_pte(base, ref_prot));
+       __set_pmd_pte(kpte, address, mk_pte(base, ref_prot));
        base = NULL;
 
 out_unlock:
-       up_write(&init_mm.mmap_sem);
+       spin_unlock_irqrestore(&pgd_lock, flags);
 
        if (base)
                __free_pages(base, 0);