]> err.no Git - linux-2.6/commitdiff
x86: CPA: avoid split of alias mappings
authorThomas Gleixner <tglx@linutronix.de>
Sat, 23 Feb 2008 19:43:21 +0000 (20:43 +0100)
committerIngo Molnar <mingo@elte.hu>
Fri, 29 Feb 2008 17:55:42 +0000 (18:55 +0100)
avoid over-eager large page splitup.

When the target area needs to be split or is split already (ioremap)
then the current code enforces the split of large mappings in the alias
regions even if we could avoid it.

Use a separate variable processed in the cpa_data structure to carry
the number of pages which have been processed instead of reusing the
numpages variable. This keeps numpages intact and gives the alias code
a chance to keep large mappings intact.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
arch/x86/mm/pageattr.c

index 14e48b5a94ba193c1cb4acc667482e05905d373e..7049294fb469b0a66e3a21ecc13d5f4c56b01524 100644 (file)
@@ -26,6 +26,7 @@ struct cpa_data {
        pgprot_t        mask_set;
        pgprot_t        mask_clr;
        int             numpages;
+       int             processed;
        int             flushtlb;
        unsigned long   pfn;
 };
@@ -290,8 +291,8 @@ try_preserve_large_page(pte_t *kpte, unsigned long address,
         */
        nextpage_addr = (address + psize) & pmask;
        numpages = (nextpage_addr - address) >> PAGE_SHIFT;
-       if (numpages < cpa->numpages)
-               cpa->numpages = numpages;
+       if (numpages < cpa->processed)
+               cpa->processed = numpages;
 
        /*
         * We are safe now. Check whether the new pgprot is the same:
@@ -318,7 +319,7 @@ try_preserve_large_page(pte_t *kpte, unsigned long address,
         */
        addr = address + PAGE_SIZE;
        pfn++;
-       for (i = 1; i < cpa->numpages; i++, addr += PAGE_SIZE, pfn++) {
+       for (i = 1; i < cpa->processed; i++, addr += PAGE_SIZE, pfn++) {
                pgprot_t chk_prot = static_protections(new_prot, addr, pfn);
 
                if (pgprot_val(chk_prot) != pgprot_val(new_prot))
@@ -342,7 +343,7 @@ try_preserve_large_page(pte_t *kpte, unsigned long address,
         * that we limited the number of possible pages already to
         * the number of pages in the large page.
         */
-       if (address == (nextpage_addr - psize) && cpa->numpages == numpages) {
+       if (address == (nextpage_addr - psize) && cpa->processed == numpages) {
                /*
                 * The address is aligned and the number of pages
                 * covers the full page.
@@ -572,7 +573,7 @@ repeat:
                        set_pte_atomic(kpte, new_pte);
                        cpa->flushtlb = 1;
                }
-               cpa->numpages = 1;
+               cpa->processed = 1;
                return 0;
        }
 
@@ -583,7 +584,7 @@ repeat:
        do_split = try_preserve_large_page(kpte, address, cpa);
        /*
         * When the range fits into the existing large page,
-        * return. cp->numpages and cpa->tlbflush have been updated in
+        * return. cp->processed and cpa->tlbflush have been updated in
         * try_large_page:
         */
        if (do_split <= 0)
@@ -662,7 +663,7 @@ static int __change_page_attr_set_clr(struct cpa_data *cpa, int checkalias)
                 * Store the remaining nr of pages for the large page
                 * preservation check.
                 */
-               cpa->numpages = numpages;
+               cpa->numpages = cpa->processed = numpages;
 
                ret = __change_page_attr(cpa, checkalias);
                if (ret)
@@ -679,9 +680,9 @@ static int __change_page_attr_set_clr(struct cpa_data *cpa, int checkalias)
                 * CPA operation. Either a large page has been
                 * preserved or a single page update happened.
                 */
-               BUG_ON(cpa->numpages > numpages);
-               numpages -= cpa->numpages;
-               cpa->vaddr += cpa->numpages * PAGE_SIZE;
+               BUG_ON(cpa->processed > numpages);
+               numpages -= cpa->processed;
+               cpa->vaddr += cpa->processed * PAGE_SIZE;
        }
        return 0;
 }