proposed sysfs semantic change.

proposed sysfs semantic change.

[Dave Jones]

Users with dual/quad core systems can currently set individual cores to
have different governors (and drivers!), which doesn't really make a lot
of sense.

I'm thinking of changing things so scaling_governor file for the 2nd/3rd/4th etc
cores is a symlink to the same file in cpu0/
So a write to one file will automatically set things on all related cores within
the same package.

For completeness we could also symlink related_cpus and driver.

thoughts ?

	Dave

Re: proposed sysfs semantic change.

[Dominik Brodowski]

Hey,

On Wed, Aug 11, 2010 at 05:54:43PM -0400, Dave Jones wrote:
> Users with dual/quad core systems can currently set individual cores to
> have different governors (and drivers!), which doesn't really make a lot
> of sense.
> 
> I'm thinking of changing things so scaling_governor file for the 2nd/3rd/4th etc
> cores is a symlink to the same file in cpu0/
> So a write to one file will automatically set things on all related cores within
> the same package.
> 
> For completeness we could also symlink related_cpus and driver.
> 
> thoughts ?

good idea.

Best,
	Doinik

Re: proposed sysfs semantic change.

[Andi Kleen]

Dave Jones <davej@redhat.com> writes:

> Users with dual/quad core systems can currently set individual cores to
> have different governors (and drivers!), which doesn't really make a lot
> of sense.

Why not? Imagine a system that is partitioned using cpusets or cgroups
for different sockets and the partitions runs different workloads.

I could well imagine using different governours (= policies) 
for those different sockets.

Different drivers probably doesn't make much sense though.

-Andi

-- 
ak@linux.intel.com -- Speaking for myself only.

Re: proposed sysfs semantic change.

[Dave Jones]

On Thu, Aug 12, 2010 at 06:35:54PM +0200, Andi Kleen wrote:
 > Dave Jones <davej@redhat.com> writes:
 > 
 > > Users with dual/quad core systems can currently set individual cores to
 > > have different governors (and drivers!), which doesn't really make a lot
 > > of sense.
 > 
 > Why not? Imagine a system that is partitioned using cpusets or cgroups
 > for different sockets and the partitions runs different workloads.
 > 
 > I could well imagine using different governours (= policies) 
 > for those different sockets.

Different sockets is fine. I'm talking about different cores within the same socket.

	Dave

Re: proposed sysfs semantic change.

[Andi Kleen]

On Thu, Aug 12, 2010 at 12:41:54PM -0400, Dave Jones wrote:
> On Thu, Aug 12, 2010 at 06:35:54PM +0200, Andi Kleen wrote:
>  > Dave Jones <davej@redhat.com> writes:
>  > 
>  > > Users with dual/quad core systems can currently set individual cores to
>  > > have different governors (and drivers!), which doesn't really make a lot
>  > > of sense.
>  > 
>  > Why not? Imagine a system that is partitioned using cpusets or cgroups
>  > for different sockets and the partitions runs different workloads.
>  > 
>  > I could well imagine using different governours (= policies) 
>  > for those different sockets.
> 
> Different sockets is fine. I'm talking about different cores within the same socket.

If you have per core p-states that makes sense too.
Even without it could in principle make sense.

-Andi
-- 
ak@linux.intel.com -- Speaking for myself only.