>>>>> On Wed, 28 Sep 2005 20:34:29 +0900 (JST), Atsushi Nemoto <anemo@xxxxxxxxxxxxx> said:
anemo> If I used indexed-flush for executable page in
anemo> flush_cache_page(), the problem disappear. Is this a right
anemo> fix?
Sorry, this would corrupt cpu_has_ic_fills_f_dc case. Revised.
diff -u linux-mips/arch/mips/mm/c-r4k.c linux/arch/mips/mm/c-r4k.c
--- linux-mips/arch/mips/mm/c-r4k.c 2005-09-22 10:38:23.000000000 +0900
+++ linux/arch/mips/mm/c-r4k.c 2005-09-28 20:55:55.000000000 +0900
@@ -409,8 +409,9 @@
* for every cache flush operation. So we do indexed flushes
* in that case, which doesn't overly flush the cache too much.
*/
- if ((mm == current->active_mm) && (pte_val(*ptep) & _PAGE_VALID)) {
- if (cpu_has_dc_aliases || (exec && !cpu_has_ic_fills_f_dc)) {
+ if ((mm == current->active_mm) && (pte_val(*ptep) & _PAGE_VALID) &&
+ !(exec && !cpu_has_ic_fills_f_dc)) {
+ if (cpu_has_dc_aliases) {
r4k_blast_dcache_page(page);
if (exec && !cpu_icache_snoops_remote_store)
r4k_blast_scache_page(page);
diff -u linux-mips/arch/mips/mm/c-tx39.c linux/arch/mips/mm/c-tx39.c
--- linux-mips/arch/mips/mm/c-tx39.c 2005-09-05 10:16:59.000000000 +0900
+++ linux/arch/mips/mm/c-tx39.c 2005-09-28 18:51:43.000000000 +0900
@@ -213,12 +213,10 @@
* for every cache flush operation. So we do indexed flushes
* in that case, which doesn't overly flush the cache too much.
*/
- if ((mm == current->active_mm) && (pte_val(*ptep) & _PAGE_VALID)) {
- if (cpu_has_dc_aliases || exec)
+ if ((mm == current->active_mm) && (pte_val(*ptep) & _PAGE_VALID) &&
+ !exec) {
+ if (cpu_has_dc_aliases)
tx39_blast_dcache_page(page);
- if (exec)
- tx39_blast_icache_page(page);
-
return;
}
---
Atsushi Nemoto
