Subject: Re: My experiences with gcc 2.7.0
To: None <bsdnat@zumdick.rhein-main.de>
From: Leo Weppelman <leo@wau.mis.ah.nl>
List: port-atari
Date: 11/03/1995 20:58:21
> 
> As a 'beta tester' I'd like to relay a few more of my experiences.
I _love_ to hear these kind of experiences as it's the only way
to get the bugs out ;-)

> 
> With the compilation of gcc-2.7.0 I for the first time witnessed
> instabilities in the NetBSD-port. I would not be surprised if the
> majority -if not all - of the problems are due to faulty disk I/O.
> 
> Currently I am busy building the third stage of the gcc compiler.
> I used -g -O2 -fomit-frame-pointer as compiler options.
> So far none of the makes made it through, there always was some
> crash. But after the crash the same code compiled flawlessly, 
> which means that its a kernel problem (IMO).
It sounds as either a SCSI or RAM problem. Both can cause this
kind of trouble; ie. corrupted memory pages.

> 
> Crashes witnessed:
> 
>    two or three times, machine freezes harddisk light stays ON
When you get this, could you please try to get into the kernel
debugger - <alt><lshift><f9> - and try the command 'call scsi_show'.
If this works, please send me the output.
>    
>    address kernel panic (UNLK A6), probably some stack messup
>    
>    cc1 exited because of signal 6 (Abort I think)
>    
> Compilation is __slow__. Although I do have Blowup installed, 
> I am not really using it, I usually run with a conservative 16/32 
> (normal) sometimes 18/36 setting. Once a compile file took 16 Minutes!!
> Raytracing is faster than gcc :)
It has to be said, that the SCSI driver on the Falcon currently operates
in polled-DMA mode; it uses DMA but busy-waits for it's end. Currently
Thomas Gerner and I are trying to solve this.
> 
> I think the Linux port was also plagued with Falcon SCSI problems, 
> and they finally fixed their driver. Maybe one could take a few hints
> from their source ?
Just reading the source of the SCSI driver does not help I'm afraid.
Usually the the problems and there solutions are rather subtle and
hard to extract from 45kb or more of source code - I took the size
of the NetBSD SCSI driver ;-) If you have however a recollection of
what their problems - and maybe solutions - where, this _sure_ would
make things easier.

Leo.