Will,

On Wed, Oct 04, 2006 at 10:09:21AM -0400, William Cohen wrote:
> 
> I am looking at reorganizing the patches so they are better self-contained 
> as mentioned in:
> http://kernelnewbies.org/UpstreamMerge/MergingStrategy
> 
That is a good idea. 

> I set up a local clone of the bleeding edge kernel using the information 
> from the GIT and cogito information on:
> 
> http://linux.yyz.us/git-howto.html
> http://www.kernel.org/pub/software/scm/cogito/README
> 
> I took a closer look perfmon-new-base-new-lib-060926.diff patch. Moved the 
> related change to the lib/Makefile to the same patch. I stripped out the 
> support for the arch specific support for the time being. How much better 
> in performance is the ia64 assembly version of the routine? Unless it has 

After some benchmarking, I noticed that it was not better. That simplifies
the code and the header files. I will remove it from the next release.

> much better performance than the C routine it might be better to just use 
> the generic version. That would eliminate the need to have an asm/perfmon.h 

Yes, and no asm/carta_random.h Dont know if you have notced but Andrew morton
just added lin/carta_random32.c  to his tree.

> for every architecture even if it doesn't support perfmon (linux/perfmon.h 
> includes asm/perfmon.h). If need be, a later patch can add thet arch 
> support back in. The revised patch made against the leading edge kernel is 
> attached.

Here you are talking about an arch with no perfmon support. Then in that case, 
I would not expect CONFIG_PERFMON to be enable in which case asm/perfmon.h
is never included.

> 
> The modified perfmon-base-mod.diff is also attached. It eliminates the 
> kernel name change and an extraneous white space change to kernel/ptrace.c
> 
Ok. Do you have problm with ehe kernel name change.

> I am wondering how architectures without perfmon support are going to be 
> handled. When linux/perfmon.h is included in various files, an 
> asm/perfmon.h will be needed. The patches don't supply one for all the 
> architectures.
> 
Because if they don't support then CONFIG_PERFMON is not set in which case
asm/perfmon.h is not loaded from linux/perfmon.h Also all kernel hooks
turn into nop.


> It looks like much (but not all) of base.diff is in the newest kernel. 
> Still looking at that.
I will submit the idle notifier changes on Friday when I am nack from the Gelato
meeting in Singapore.

-- 
-Stephane
_______________________________________________
perfmon mailing list
[email protected]
http://www.hpl.hp.com/hosted/linux/mail-archives/perfmon/

Reply via email to