Michael L. Hitch wrote:
> On Tue, 3 Apr 2007, Johnny Billquist wrote:
> 
>>>> Also, probably very unrelated. A kernel as built today is not 
>>>> useable. It goes up to about 100% system time as soon as I copy 
>>>> anything from an NFS mounted disk to a local disk on my VAX.
>>>> This is 4.99.16. I have an older build (4.99.15) where this don't 
>>>> happen, so something changed recently. But like I said, I don't 
>>>> think that is related to GCC.
>>>
>>>   I think this is related to mutex_spin_{enter,exit), which I am 
>>> looking at.  I think it also causes the le0 device to not be detected 
>>> on my 4000/VLC, and probably on the 3100 mentioned in an earlier 
>>> port-vax message.
>>>
>>>   A work-around may be to build a kernel with LOCKDEBUG.
>>
>> Hmm. Might try that later. Anyway. The problem is something that 
>> happened after march 15, since my kernel from march 15 works fine.
> 
>   It appears that the gcc changes that enabled PIC support may be to 
> blame.  I built a kernel using sources from 2007.04.01.06.00.00, but 
> using the tools from the day before.  That kernel seemed to run fine.  
> Then I rebuilt the tools and another kernel.  That kernel now exhibits 
> the problem of spending a lot of time in the kernel.  It seems to be 
> related to writing to the filesystem - that's the only time I've seen 
> the problem.

I believe the gcc codegen problems have been fixed.