[ On Thursday, February 28, 2002 at 13:46:11 (-0500), Colin Surprenant wrote: ]
> Subject: AIC-7899 & RAIDFrame SCSI problems in 1.5.2
>
> I have successfully installed NetBSD 1.5.2 on a Dell PowerEdge 1500SC using
> the AIC-7899 patch on pchb.c backported by Johnny Lam from current (see
> archives).
> 
>[[....]]
> 
> Here's the trace of what happened once I pulled the drive:
> 
>     ahc1: AHC_INTR - referenced SCB not valid during SELTO SCP(14,26)
>     ... (a bunch of these)
> 
>     sd0 (ahc1:0:0) SCB 1c : timed out while idle, SEQ ADDR == 0x43
>     SCSI RATE == 0x0
>     sd0 (ahc1:0:0) SCB 1c : internal reset. Flags = 0x4040
>     sd0 (ahc1:0:0) SCB 1c : no longer in timeout, status = 0
>     panic: biodone already

Hmmm.... IIRC that's similar to what happens in -current on AIC-7899
when under load with multiple U-160 targets on the bus.  ("load" is
relative here -- I was sometimes unable to get the machine even so far
as to multi-user mode with even just two targets mounted and active use
of the filesystems only on one of them)

I had understood the 1.5.2 ahc driver would work well enough with the
AIC-7899, only with degraded performance (at least with U-160 drives).
Your report suggests this is not correct and that even in 1.5.2 the
AIC-7899 support is not good enough for production use with multiple
U-160 targets even in degraded mode.

(I have a machine with an Intel STL-2 motherboard, with AIC-7899, and a
single U-160 drive that's been running 1.5W happily for many months, and
it does quite a bit of disk work, and has rebuilt NetBSD and various
kernels and many packages too.  It's in production now as a database
server actually.  I.e. the -current driver is OK with a single U-160
target on an AIC-7899.)

I do know that the ahc driver in FreeBSD-4.4 is very stable and gets
enormous and wonderful performance from the AIC-7899 with multiple
Seagate Cheetah U-160 drives attached.  I've been running two machines
identical to the above STL-2 but with four drives each (and more RAM) as
squid servers under reasonably heavy daily load (up to 120 requests per
second each) -- with squid running since the first week of Dec. and only
restarted to upgrade it the other day.  I've been hoping this newer
version of the driver would be re-ported to NetBSD soon, but I haven't
heard anything about it in many months now.

-- 
								Greg A. Woods

+1 416 218-0098;  <gwoods@acm.org>;  <g.a.woods@ieee.org>;  <woods@robohack.ca>
Planix, Inc. <woods@planix.com>; VE3TCP; Secrets of the Weird <woods@weird.com>