[226] in linux-scsi channel archive
Re: ELF kernel support causes SCSI timeout (FIXED)
daemon@ATHENA.MIT.EDU (Don Parsons)
Sun Jun 4 15:04:19 1995
Date: Sun, 4 Jun 1995 14:36:46 -0400 (EDT)
From: Don Parsons <dparsons@synapse.kent.edu>
Reply-To: Don Parsons <dparsons@synapse.kent.edu>
To: Eric Youngdale <eric@aib.com>
cc: linuxgcc <linux-gcc@vger.rutgers.edu>,
linuxscsi <linux-scsi@vger.rutgers.edu>,
"UltraStor Corp." <ustor@PrimeNet.Com>
In-Reply-To: <9505301350.ZM22744@aib.com>
On Tue, 30 May 1995, Eric Youngdale wrote:
> On May 26, 12:59am, Don Parsons wrote:
> > Subject: ELF kernel support causes SCSI timeout
> >
> > I've been reading about SCSI timeout problems recently and I thought I was
> > lucky because I didn't have any. Well today it happened to me every
> > time(2) I executed:
> >
> > tar tvzf gcc-2.6.3.bin.tar.gz | more (actually most)
> >
> > where the .tar.gz file size is 3.6MB. I had already done this with
[omit more details]
>
> Since you are not using any ELF binaries (the fact that you can use
> a non-ELF kernel suggests this much), the ELF loader itself cannot be
> to blame for this. My guess is that the addresses of some things got
> shifted for some reason, which resulted in some other subtle difference.
>
> It is possible that there is some uninitialized memory being used
> somewhere, and the change in memory layout is responsible for the problem.
> This is just a wild guess, however.
>
> > Info: Using H. Lu's gcc-2.6.2 for compiling both kernels.
> > Additional opts: -O6 -fexpensive-optimizations -fstrength-reduce
> > 386DX-33 + 387 8MB RAM
> > 1.23GB Toshiba SCSI UltraStor 14F Old firmware 2.00
> > UltraStor driver version1.12. Using 16 SG lists.
> >
> > I copied the following off the console screen (note PID should only be in
> > range 100 to 200, not 16546 !? ):
> The pid is not the process id, but the id of the scsi request.
> The scsi requests are each given a unique number, and this helps to isolate
> problems when looking at printouts.
>
> > Also the red SCSI drive light was on continuously.
>
> Sounds almost like a bus hang.
> Unfortunately it makes no sense to me whatsoever. I can keep it in
> mind in case any ideas come up.
>
> -Eric
>
I have now run successfully the above 'bad' tar command which caused the
bus hang, both with and without X running. Since I couldn't use gcc for
the past week, there is only one change in my system: I had been getting
ready to upgrade my UltraStor 14F ROMs, BIOS and Firmware, so I took this
opportunity to do the ROM upgrades:
14F Firmware -27C256, 100ns from 28004.004 to 28004.006
14F BIOS -27C128, 125ns from 38004.004 to 38004.005 v2.01 now
(code from: ftp.primenet.com/users/u/ustor/14f-hex.zip
warning: I have been getting connection refused the last few weeks!?)
This has changed my 2 out of 2 tar failures to 2 out of 2 successes.
Since nobody in the past 3 years reported 14F ROM upgrades helping SCSI
bus hangs or SCSI timeouts, I'm reporting it!
It still seems strange that answering Yes to ELF in the kernel make config
caused such problems.
Bye for now,
Don Parsons
