powerpc: Update find_linux_pte_or_hugepte to handle transparent hugepages
authorAneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Thu, 20 Jun 2013 09:00:17 +0000 (14:30 +0530)
committerBenjamin Herrenschmidt <benh@kernel.crashing.org>
Fri, 21 Jun 2013 06:01:54 +0000 (16:01 +1000)
Reviewed-by: David Gibson <dwg@au1.ibm.com>
Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
arch/powerpc/mm/hugetlbpage.c

index 2865077e01598f14a0b651790ecf7dc45cb12d0a..49282045ee968fb50d3beaec4a3bba13c1c2cadf 100644 (file)
@@ -936,30 +936,50 @@ pte_t *find_linux_pte_or_hugepte(pgd_t *pgdir, unsigned long ea, unsigned *shift
 
        pg = pgdir + pgd_index(ea);
 
-       if (pgd_huge(*pg)) {
+       /*
+        * we should first check for none. That takes care of a
+        * a parallel hugetlb or THP pagefault moving none entries
+        * to respective types.
+        */
+       if (pgd_none(*pg))
+               return NULL;
+       else if (pgd_huge(*pg)) {
                ret_pte = (pte_t *) pg;
                goto out;
        } else if (is_hugepd(pg))
                hpdp = (hugepd_t *)pg;
-       else if (!pgd_none(*pg)) {
+       else {
                pdshift = PUD_SHIFT;
                pu = pud_offset(pg, ea);
 
-               if (pud_huge(*pu)) {
+               if (pud_none(*pu))
+                       return NULL;
+               else if (pud_huge(*pu)) {
                        ret_pte = (pte_t *) pu;
                        goto out;
                } else if (is_hugepd(pu))
                        hpdp = (hugepd_t *)pu;
-               else if (!pud_none(*pu)) {
+               else {
                        pdshift = PMD_SHIFT;
                        pm = pmd_offset(pu, ea);
+                       /*
+                        * A hugepage collapse is captured by pmd_none, because
+                        * it mark the pmd none and do a hpte invalidate.
+                        *
+                        * A hugepage split is captured by pmd_trans_splitting
+                        * because we mark the pmd trans splitting and do a
+                        * hpte invalidate
+                        *
+                        */
+                       if (pmd_none(*pm) || pmd_trans_splitting(*pm))
+                               return NULL;
 
-                       if (pmd_huge(*pm)) {
+                       if (pmd_huge(*pm) || pmd_large(*pm)) {
                                ret_pte = (pte_t *) pm;
                                goto out;
                        } else if (is_hugepd(pm))
                                hpdp = (hugepd_t *)pm;
-                       else if (!pmd_none(*pm))
+                       else
                                return pte_offset_kernel(pm, ea);
                }
        }