Chris Gianelloni wrote:
> On Fri, 2007-05-11 at 10:25 +1000, Paul de Vrieze wrote:
>   
>> M. Edward (Ed) Borasky wrote:
>>     
>>> I can't tell yet whether it's hardware or software, but I'm guessing the 
>>> kernel at this point rather than hardware -- I have some 
>>> /var/log/messages traces that don't look like hardware. Once I get a 
>>> stable OS, I'll load Gentoo from it and do the real debugging. The 
>>> 2.4.20 kernel in Gentoo stable has to be better that the 2.4.18s the 
>>> other distros seem to be carrying.
>>>       
>> What I find to be a cause of a lot of stability issues. Probably even 
>> more so on multiprocessor/core systems is preemption. Voluntary 
>> preemption is OK, but other preemption still seems to be a bit flaky at 
>> places.
>>     
>
> Umm... What?
>
> I use preemption all the time on all of my machines, which are *all*
> multi-processor or multi-core.  The kernel preemption works just fine on
> all of them.  What tends to be the problem is shoddy APIC or ACPI
> implementations on the cheaper (read, not server/workstation class)
> motherboards.
>
>   
I'm still testing some things, but there are some bugs in the NVidia 
"sata_nv" part of the AMD64 Linux kernel that have showed up in the 
Debian bug archives. Apparently it worked in older kernels and got 
broken "recently". I haven't found any differences with or without 
either pre-emption, SMP vs. UP, ACPI or APIC. The only thing I've found 
so far is that 32-bit kernels seem to work and 64-bit ones don't. At the 
moment I've only got a gigabyte of RAM in the machine so I'm not losing 
a heck of a lot by running a 32-bit kernel. :)
-- 
gentoo-releng@g.o mailing list