Re: 2.6.12-rc2-mm1

[Nick Piggin <nickpiggin@yahoo.com.au>]

Siddha, Suresh B wrote:
> On Tue, Apr 05, 2005 at 05:33:49PM +1000, Nick Piggin wrote:

>>Suresh's underlying problem with the unnecessary sched domains
>>is a failing of sched-balance-exec and sched-balance-fork, which
> 
> 
> That wasn't the only motivation. For example, on non-HT cpu's we shouldn't
> be setting up SMT sched-domain, same with NUMA domains on non-NUMA systems.
> 

Yep, sure. It is a good, if slight, optimisation. And I've also just
slightly extended your patch, so we don't have any domains if booting
with maxcpus=1

> 
>>I am working on now.
>>
>>Removing unnecessary domains is a nice optimisation, but just
>>needs to account for a few more flags before declaring that a
> 
> 
> Can you elaborate when we require a domain with special flags but has
> no or only one group in it.
> 

The SD_WAKE_* flags do not use groups, so it would be legitimate to
have a domain that has one of these set, with no groups.

> 
>>domain is unnecessary (not to mention this probably breaks if
>>isolcpus= is used). I have made some modifications to the patch
> 
> 
> I have tested my patch with "ioslcpus=" and it works just fine.
> 

OK, my apologies ;)

> 
>>to fix these problems.
>>
>>Lastly, I'd like to be a bit less intrusive with pinned task
>>handling improvements. I think we can do this while still being
>>effective in preventing livelocks.
> 
> 
> We want to see this fixed. Please post your patch and I can let you know
> the test results.
> 

I will try to get it working and tested tonight for you.

> 
>>I will keep you posted with regards to the various scheduler
>>patches.
> 
> 
> Nick, Can you post the patches you sent me earlier to this list?
> 

Yep, I'll post them.


-- 
SUSE Labs, Novell Inc.