[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]
kern/42376: ahd-driver bug in cmd-queuing - duplicate TAG send to device
>Synopsis: ahd-driver bug in cmd-queuing - duplicate TAG send to device
>Arrival-Date: Wed Nov 25 12:45:00 +0000 2009
>Originator: W. Stukenbrock
>Release: NetBSD 4.0
Dr. Nagler & Company GmbH
System: NetBSD test-s0 4.0 NetBSD 4.0 (NSW-WS) #2: Fri Nov 6 11:14:49 CET 2009
The ADH driver manages 512 SBC entries, but there are only 256 different
cmd-TAGs possbile on the SCSI-bus.
So there is a SCB-manaagment implemented, that avoid the use of a
conflicting SCB-tag on a device.
Accedently the check, if a conflict may happen or not is broken.
Command queuing is enabled on the scspi-layer, if xs_tag_type is
set to the correct SCSI-message value.
Now in aic79xx_osm.c this value is checked for non-zero to be
non-cnflicting in any case. It should be checked for zero instead.
This is not easy ...
I've the problem, that on some disks on heavy load some commands are
rejected by some oth the disk drives with ASC=4d (Deivce reports conflicting
After that a device timeout happens with the SCB with the same TAG
pattern in the lower 8 bits as reported .
I've found no way to reproduce it on demand up to now.
I've analysed the code to find the reason for it - see patch and
The following patch to /usr/src/sys/ic/aic79xx_osm.c will fix the
--- aic79xx_osm.c 2008/09/26 11:20:25 1.2
+++ aic79xx_osm.c 2009/11/25 10:12:40
@@ -312,7 +312,7 @@
tinfo = ahd_fetch_transinfo(ahd, channel, our_id,
- if (xs->xs_tag_type != 0 ||
+ if (xs->xs_tag_type == 0 ||
(tinfo->curr.ppr_options & MSG_EXT_PPR_IU_REQ) != 0)
col_idx = AHD_NEVER_COL_IDX;
Main Index |
Thread Index |