per cpu data section contains two types of data.  One set which is
exclusively accessed by the local cpu and the other set which is per cpu,
but also shared by remote cpus.  In the current kernel, these two sets are
not clearely separated out.  This can potentially cause the same data
cacheline shared between the two sets of data, which will result in
unnecessary bouncing of the cacheline between cpus.

One way to fix the problem is to cacheline align the remotely accessed per
cpu data, both at the beginning and at the end.  Because of the padding at
both ends, this will likely cause some memory wastage and also the
interface to achieve this is not clean.

This patch:

Moves the remotely accessed per cpu data (which is currently marked
as ____cacheline_aligned_in_smp) into a different section, where all the data
elements are cacheline aligned. And as such, this differentiates the local
only data and remotely accessed data cleanly.
Signed-off-by: NFenghua Yu <fenghua.yu@intel.com>
Acked-by: NSuresh Siddha <suresh.b.siddha@intel.com>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Christoph Lameter <clameter@sgi.com>
Cc: <linux-arch@vger.kernel.org>
Cc: "Luck, Tony" <tony.luck@intel.com>
Cc: Andi Kleen <ak@suse.de>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Add the per_cpu_offset() generic method. (used by the lock validator)
Signed-off-by: NIngo Molnar <mingo@elte.hu>
Signed-off-by: NArjan van de Ven <arjan@linux.intel.com>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

There are several instances of per_cpu(foo, raw_smp_processor_id()), which
is semantically equivalent to __get_cpu_var(foo) but without the warning
that smp_processor_id() can give if CONFIG_DEBUG_PREEMPT is enabled.  For
those architectures with optimized per-cpu implementations, namely ia64,
powerpc, s390, sparc64 and x86_64, per_cpu() turns into more and slower
code than __get_cpu_var(), so it would be preferable to use __get_cpu_var
on those platforms.

This defines a __raw_get_cpu_var(x) macro which turns into per_cpu(x,
raw_smp_processor_id()) on architectures that use the generic per-cpu
implementation, and turns into __get_cpu_var(x) on the architectures that
have an optimized per-cpu implementation.
Signed-off-by: NPaul Mackerras <paulus@samba.org>
Acked-by: NDavid S. Miller <davem@davemloft.net>
Acked-by: NIngo Molnar <mingo@elte.hu>
Acked-by: NMartin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

for_each_cpu() actually iterates across all possible CPUs.  We've had mistakes
in the past where people were using for_each_cpu() where they should have been
iterating across only online or present CPUs.  This is inefficient and
possibly buggy.

We're renaming for_each_cpu() to for_each_possible_cpu() to avoid this in the
future.

This patch replaces for_each_cpu with for_each_possible_cpu.
Signed-off-by: NKAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

When we stop allocating percpu memory for not-possible CPUs we must not touch
the percpu data for not-possible CPUs at all.  The correct way of doing this
is to test cpu_possible() or to use for_each_cpu().

This patch is a kernel-wide sweep of all instances of NR_CPUS.  I found very
few instances of this bug, if any.  But the patch converts lots of open-coded
test to use the preferred helper macros.

Cc: Mikael Starvik <starvik@axis.com>
Cc: David Howells <dhowells@redhat.com>
Acked-by: NKyle McMartin <kyle@parisc-linux.org>
Cc: Anton Blanchard <anton@samba.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Cc: Paul Mundt <lethal@linux-sh.org>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: William Lee Irwin III <wli@holomorphy.com>
Cc: Andi Kleen <ak@muc.de>
Cc: Christian Zankel <chris@zankel.net>
Cc: Philippe Elie <phil.el@wanadoo.fr>
Cc: Nathan Scott <nathans@sgi.com>
Cc: Jens Axboe <axboe@suse.de>
Cc: Eric Dumazet <dada1@cosmosbay.com>
Signed-off-by: NAndrew Morton <akpm@osdl.org>
Signed-off-by: NLinus Torvalds <torvalds@osdl.org>

The current ppc64 per cpu data implementation is quite slow. eg:

        lhz 11,18(13)           /* smp_processor_id() */
        ld 9,.LC63-.LCTOC1(30)  /* per_cpu__variable_name */
        ld 8,.LC61-.LCTOC1(30)  /* __per_cpu_offset */
        sldi 11,11,3            /* form index into __per_cpu_offset */
        mr 10,9
        ldx 9,11,8              /* __per_cpu_offset[smp_processor_id()] */
        ldx 0,10,9              /* load per cpu data */

5 loads for something that is supposed to be fast, pretty awful. One
reason for the large number of loads is that we have to synthesize 2
64bit constants (per_cpu__variable_name and __per_cpu_offset).

By putting __per_cpu_offset into the paca we can avoid the 2 loads
associated with it:

        ld 11,56(13)            /* paca->data_offset */
        ld 9,.LC59-.LCTOC1(30)  /* per_cpu__variable_name */
        ldx 0,9,11              /* load per cpu data

Longer term we can should be able to do even better than 3 loads.
If per_cpu__variable_name wasnt a 64bit constant and paca->data_offset
was in a register we could cut it down to one load. A suggestion from
Rusty is to use gcc's __thread extension here. In order to do this we
would need to free up r13 (the __thread register and where the paca
currently is). So far Ive had a few unsuccessful attempts at doing that :)

The patch also allocates per cpu memory node local on NUMA machines.
This patch from Rusty has been sitting in my queue _forever_ but stalled
when I hit the compiler bug. Sorry about that.

Finally I also only allocate per cpu data for possible cpus, which comes
straight out of the x86-64 port. On a pseries kernel (with NR_CPUS == 128)
and 4 possible cpus we see some nice gains:

             total       used       free     shared    buffers cached
Mem:       4012228     212860    3799368          0          0 162424

             total       used       free     shared    buffers cached
Mem:       4016200     212984    3803216          0          0 162424

A saving of 3.75MB. Quite nice for smaller machines. Note: we now have
to be careful of per cpu users that touch data for !possible cpus.

At this stage it might be worth making the NUMA and possible cpu
optimisations generic, but per cpu init is done so early we have to be
careful that all architectures have their possible map setup correctly.
Signed-off-by: NAnton Blanchard <anton@samba.org>
Signed-off-by: NPaul Mackerras <paulus@samba.org>

They differed in either simple comments or in the protecting ifdefs.
Signed-off-by: NStephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: NPaul Mackerras <paulus@samba.org>