On Tue, Mar 30, 2010 at 04:32:25PM +1100, Benjamin Herrenschmidt wrote:

On Tue, 2010-03-30 at 16:24 +1100, Paul Mackerras wrote:

quoted

On Tue, Mar 23, 2010 at 07:37:02PM +0530, K.Prasad wrote:

quoted

Index: linux-2.6.ppc64_test/arch/powerpc/kernel/exceptions-64s.S
===================================================================

--- linux-2.6.ppc64_test.orig/arch/powerpc/kernel/exceptions-64s.S
+++ linux-2.6.ppc64_test/arch/powerpc/kernel/exceptions-64s.S

@@ -735,6 +735,9 @@ _STATIC(do_hash_page)
 	std	r3,_DAR(r1)
 	std	r4,_DSISR(r1)
 
+	andis.  r0,r4,0x0040		/* Data Address Breakpoint match? */

Minor comment: why not DSISR_DABRMATCH@h instead of 0x0040?

quoted

+	bne-    handle_dabr_fault
+
 	andis.	r0,r4,0xa450		/* weird error? */
 	bne-	handle_page_fault	/* if not, try to insert a HPTE */
 BEGIN_FTR_SECTION

@@ -823,6 +826,15 @@ END_FW_FTR_SECTION_IFCLR(FW_FEATURE_ISER
 	bl	.raw_local_irq_restore
 	b	11f

I would move your new test to the "weird error" case (ie, after the bne-
handle_page_fault) to avoid hitting the fast path.

Done that...so basically the branch to handle_page_fault will happen
only if 0xa410 matches.

The changes can be seen here: linuxppc-dev: message-id:
20100330095925.GB14403@in.ibm.com.

Thanks,
K.Prasad