[src/trunk]: src/sys/arch/sparc64/sparc64 Band aid - the infamous dcache_flus...

[martin <martin%NetBSD.org@localhost>]

details:   https://anonhg.NetBSD.org/src/rev/f20ddd25a6b8
branches:  trunk
changeset: 544109:f20ddd25a6b8
user:      martin <martin%NetBSD.org@localhost>
date:      Tue Mar 11 21:28:23 2003 +0000

description:
Band aid - the infamous dcache_flush_page change independently suggested
by Andrey Petrov and Jason Thorpje (AFAIK).

This is a hack that covers some symptoms while we have no idea where
the real problem is. Anyway, since this avoids random data corruption
we better be safe and have this in-tree until the problem is solved the
right way.

diffstat:

 sys/arch/sparc64/sparc64/pmap.c |  12 +++++++++++-
 1 files changed, 11 insertions(+), 1 deletions(-)

diffs (33 lines):

diff -r 873a29503066 -r f20ddd25a6b8 sys/arch/sparc64/sparc64/pmap.c
--- a/sys/arch/sparc64/sparc64/pmap.c   Tue Mar 11 19:43:04 2003 +0000
+++ b/sys/arch/sparc64/sparc64/pmap.c   Tue Mar 11 21:28:23 2003 +0000
@@ -1,4 +1,4 @@
-/*     $NetBSD: pmap.c,v 1.135 2003/01/31 19:05:57 martin Exp $        */
+/*     $NetBSD: pmap.c,v 1.136 2003/03/11 21:28:23 martin Exp $        */
 #undef NO_VCACHE /* Don't forget the locked TLB in dostart */
 #define        HWREF
 /*
@@ -2540,6 +2540,7 @@
 pmap_clear_modify(pg)
        struct vm_page *pg;
 {
+       paddr_t pa = VM_PAGE_TO_PHYS(pg);
        pv_entry_t pv;
        int i, changed = 0;
 #ifdef DEBUG
@@ -2601,6 +2602,15 @@
                        pv->pv_va &= ~(PV_MOD);
                        simple_unlock(&pmap->pm_lock);
                }
+
+       /*
+        * XXX
+        * This should not be necessary - but empirically it is.
+        * We need to find the reason this makes a difference and fix
+        * the root of the problem - then remove this band aid.
+        */
+       dcache_flush_page(pa);
+
        pv_check();
 #ifdef DEBUG
        if (pmap_is_modified(pg)) {