Regression in 2.6.18-rc2-mm1: mbind() not binding

Regression in 2.6.18-rc2-mm1: mbind() not binding

[Lee Schermerhorn]

Just a heads up:  it appears that mbind() does not work--e.g. on
anonymous pages--in 2.6.18-rc2-mm1.

Found with my memtoy tool, available at:
	http://free.linux.hp.com/~lts/Tools/memtoy-latest.tar.gz

Requires a NUMA platform or fakenuma kernel to see this.  I'm not sure
yet whether the specified policy is not being installed, or it's just
being ignored at allocation time.  Note that default policy works:  when
I change the cpu/node affinity of the test, allocation tracks to new
node.  This indicates that get_mempolicy(...,  MPOL_F_NODE|MPOL_F_ADDR)
isn't lying to me.

Works in 2.6.18-rc2.  I've just grabbed the broken out series, and will
attempt to isolate the patch.  If anyone else has come across this and
already knows what's causing it--that would save me some effort.

Lee

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: Regression in 2.6.18-rc2-mm1: mbind() not binding

[Lee Schermerhorn]

On Wed, 2006-08-02 at 12:06 -0400, Lee Schermerhorn wrote:
> Just a heads up:  it appears that mbind() does not work--e.g. on
> anonymous pages--in 2.6.18-rc2-mm1.
> 
> Found with my memtoy tool, available at:
> 	http://free.linux.hp.com/~lts/Tools/memtoy-latest.tar.gz
> 
> Requires a NUMA platform or fakenuma kernel to see this.  I'm not sure
> yet whether the specified policy is not being installed, or it's just
> being ignored at allocation time.  Note that default policy works:  when
> I change the cpu/node affinity of the test, allocation tracks to new
> node.  This indicates that get_mempolicy(...,  MPOL_F_NODE|MPOL_F_ADDR)
> isn't lying to me.
> 
> Works in 2.6.18-rc2.  I've just grabbed the broken out series, and will
> attempt to isolate the patch.  If anyone else has come across this and
> already knows what's causing it--that would save me some effort.

Update:  looks like only the MPOL_BIND policy is affected.  Preferred
and Interleaved seem to work.  Maybe something to do with zone lists?
Still investigating.

Lee


--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>