Frank van der Linden wrote:

>On Thu, Oct 31, 2002 at 09:33:24AM -0800, Alex Barclay wrote:
>  
>
>>I've started my search. I'm pulling syssrc from cvs going back a month 
>>at a time. I'll let you know when the problem started. Hopefully I'll be 
>>able to run single use from the kernels so I won't have to keep getting 
>>an older userland - I haven't deleted any old libraries for ages so I 
>>should be OK.
>>    
>>
>
>Well, kernels should always be able to run an older userland (for older
>versions you'll need the appropriate COMPAT_* options, of course).
>
>Anyway, in the meantime, I reshuffled the interrupt priorities a bit,
>favouring serial interrupts more. Maybe you can try an updated
>-current kernel and see if things are better now.
>
>For the MP case, the messy issue of having an interrupt that is
>  
>
>>the level blocking scheduler structies needs to be solved more
>>    
>>
>cleanly. Currently it's worked around by installing a handler
>that doesn't take the kernel lock, assuming that it won't
>mess with anything. But I guess I'll have to tackle that
>in my changes in the intr code.
>
>- Frank
>
>  
>

Thanks for getting back to me - it seems that this was a _possibly_ 
false alarm. I've now tried about 10 different kernels dating back to a 
1.5D. The problem manifested itself back to a 1.5ZA then when I went 
back to the 1.5D everything was fine. Now I find that my 1.6H is also 
fine. Something odd has occurred here! Another observation is that the 
speed on the initial boot was being reported as 400MHz and is now back 
to the 200MHz which it should be - it appears that this is correlated 
but I have no idea why this would affect things or how to reproduce the 
defect. I'll keep and eye on things and see if it reoccurs. The only 
other thing I did was to remove a Adaptec 1520 ISA SCSI card but that 
has now been replaced (this was because the IRQ clashed with 11 which 
all of my PCI devices are on)

Alex.