OK, this time I think it's *really* fixed.  The funny thing is that I
always test in timing mode, since that's the much harder case to get
right, and it wasn't until yesterday I realized you were running in
atomic mode.  Turns out there was a rare but simple case where the
common code assumed it was in timing mode so it stuck a writeback in
the writeback buffer to be processed later, except in atomic mode you
never look in the writeback buffer so that writeback got lost.

Anyway, let me know if this doesn't work, but it works for me (even
with the -t -s options to Radix).

Thanks for your patience...

Steve

Attachment: cache.patch
Description: Binary data

_______________________________________________
m5-users mailing list
[email protected]
http://m5sim.org/cgi-bin/mailman/listinfo/m5-users

Reply via email to