Re: ia64 allmodconfig on current mainline

Re: ia64 allmodconfig on current mainline

[Rafael J. Wysocki]

On Wednesday, 13 of August 2008, Andrew Morton wrote:
> /opt/crosstool/gcc-3.4.5-glibc-2.3.6/ia64-unknown-linux-gnu/bin/ia64-unknown-linux-gnu-ld: section .data [0000000004ea0000 -> 0000000004f656e7] overlaps section .data.percpu [0000000004e90000 -> 0000000004ea0d17]
> make: *** [.tmp_vmlinux1] Error 1

I gather this is a regression from .26?

Re: ia64 allmodconfig on current mainline

[Andrew Morton]

On Wed, 13 Aug 2008 00:22:50 +0200
"Rafael J. Wysocki" <rjw@sisk.pl> wrote:

> On Wednesday, 13 of August 2008, Andrew Morton wrote:
> > /opt/crosstool/gcc-3.4.5-glibc-2.3.6/ia64-unknown-linux-gnu/bin/ia64-unknown-linux-gnu-ld: section .data [0000000004ea0000 -> 0000000004f656e7] overlaps section .data.percpu [0000000004e90000 -> 0000000004ea0d17]
> > make: *** [.tmp_vmlinux1] Error 1
> 
> I gather this is a regression from .26?

yes, more recent than that, actually.  2.6.27-rc1-mm1 was OK.

RE: ia64 allmodconfig on current mainline

[Luck, Tony]

> /opt/crosstool/gcc-3.4.5-glibc-2.3.6/ia64-unknown-linux-gnu/bin/ia64-unknown-linux-gnu-ld: section .data [0000000004ea0000 -> 0000000004f656e7] overlaps section .data.percpu [0000000004e90000 -> 0000000004ea0d17]

This means there is too much DEFINE_PER_CPU() going on.  Current
ia64 limit for the sum of all per cpu objects is 64K.  For a
build using defconfig we seem to be close (56K/64K ... look for
the address of __per_cpu_end and see how close it is to the top
of the address space: 0xffffffffffffffff).

-Tony

Re: ia64 allmodconfig on current mainline

[Robin Holt]

On Tue, Aug 12, 2008 at 05:21:09PM -0700, Luck, Tony wrote:
> > /opt/crosstool/gcc-3.4.5-glibc-2.3.6/ia64-unknown-linux-gnu/bin/ia64-unknown-linux-gnu-ld: section .data [0000000004ea0000 -> 0000000004f656e7] overlaps section .data.percpu [0000000004e90000 -> 0000000004ea0d17]
> 
> This means there is too much DEFINE_PER_CPU() going on.  Current
> ia64 limit for the sum of all per cpu objects is 64K.  For a
> build using defconfig we seem to be close (56K/64K ... look for
> the address of __per_cpu_end and see how close it is to the top
> of the address space: 0xffffffffffffffff).

Not sure if anybody is looking into this.  Just tried a compile of generic
and allmod.  The differences are not that significant with the exception
of per_cpu__kmem_cache_cpu at 0x28a0.  A couple other low-hanging fruit
here are per_cpu__shadow_flush_counts and per_cpu__mmu_gathers each of
which takes approx 16k.

Thanks,
Robin

RE: ia64 allmodconfig on current mainline

[Luck, Tony]

> of per_cpu__kmem_cache_cpu at 0x28a0.  A couple other low-hanging fruit
> here are per_cpu__shadow_flush_counts and per_cpu__mmu_gathers each of
> which takes approx 16k.

Ah.  That perhaps explains the recent jump in usage that has put
allmodconfig over the limit.  mmu_gathers is an int[NR_CPUS], and
NR_CPUS was just raised from 1024 to 4096

-Tony