[PATCH v2 0/10] mm: SLxB cleaning and trace accuracy improvement

[PATCH v2 0/10] mm: SLxB cleaning and trace accuracy improvement

[Ezequiel Garcia]

Hi everyone,

This is the second spin of my patchset to clean SLxB and improve kmem
trace events accuracy.

For this v2, the most relevant stuff is:

I've dropped two patches that were not very well received:
Namely this two are now gone:
  mm, slob: Use only 'ret' variable for both slob object and returned pointer
  mm, slob: Trace allocation failures consistently
I believe consistency is important but perhaps this is just me being paranoid.

There's a lot of dumb movement and renaming. This might seem stupid
(and maybe it is) but it's necessary to create some common code between SLAB
and SLUB, and then factor it out.

Also, there's a patch to add a new option to disable gcc auto-inlining.
I know we hate to add new options, but this is necessary to get
accurate call site
traces. Plus, the option is in "Kernel Hacking", so it's for kernel
developers only.

This work is part of CELF Workgroup Project:
"Kernel_dynamic_memory_allocation_tracking_and_reduction" [1]

Feedback, comments, suggestions are very welcome.

Ezequiel Garcia (10):
 mm: Factor SLAB and SLUB common code
 mm, slub: Rename slab_alloc() -> slab_alloc_node() to match SLAB
 mm, slab: Rename __cache_alloc() -> slab_alloc()
 mm, slab: Match SLAB and SLUB kmem_cache_alloc_xxx_trace() prototype
 mm, slab: Replace 'caller' type, void* -> unsigned long
 mm, util: Use dup_user to duplicate user memory
 mm, slob: Add support for kmalloc_track_caller()
 mm, slab: Remove silly function slab_buffer_size()
 mm, slob: Use NUMA_NO_NODE instead of -1
 Makefile: Add option CONFIG_DISABLE_GCC_AUTOMATIC_INLINING

 Makefile                 |    4 ++
 include/linux/slab.h     |    6 ++-
 include/linux/slab_def.h |   14 ++---
 include/linux/slob_def.h |    6 ++-
 include/linux/slub_def.h |    2 +
 lib/Kconfig.debug        |   11 ++++
 mm/slab.c                |  122 +++++++++-------------------------------------
 mm/slab_common.c         |   57 +++++++++++++++++++++
 mm/slob.c                |   33 ++++++++++--
 mm/slub.c                |   55 +++------------------
 mm/util.c                |   11 +++-
 11 files changed, 154 insertions(+), 167 deletions(-)

Thanks!
Ezequiel.

Re: [PATCH v2 0/10] mm: SLxB cleaning and trace accuracy improvement

[Christoph Lameter]

On Sat, 8 Sep 2012, Ezequiel Garcia wrote:

> This is the second spin of my patchset to clean SLxB and improve kmem
> trace events accuracy.

Please redo the patches on top of the patchsets that create
mm/slab_common.c. You will be able to extract a lot more common code and
help the goal of having as much common code as possible. PLease move as
much as possible of the common functions into slab_common.c

Re: [PATCH v2 0/10] mm: SLxB cleaning and trace accuracy improvement

[Ezequiel Garcia]

Christoph,

On Sat, Sep 8, 2012 at 7:30 PM, Christoph Lameter <cl@linux.com> wrote:
> On Sat, 8 Sep 2012, Ezequiel Garcia wrote:
>
>> This is the second spin of my patchset to clean SLxB and improve kmem
>> trace events accuracy.
>
> Please redo the patches on top of the patchsets that create
> mm/slab_common.c. You will be able to extract a lot more common code and
> help the goal of having as much common code as possible. PLease move as
> much as possible of the common functions into slab_common.c
>

Ah, I wasn't sure where to base my patches. I can split this patchset in two and
base the SLAB/SLUB commonize part on top of your tree, or perhaps just
based everything on top of your tree.

Is it this one?
http://west.gentwo.org/gitweb/?p=christoph;a=shortlog;h=refs/heads/common

I have to admit I started thinking of this commonization after seeing
your common
code patches.

Thanks!
Ezequiel.

Re: [PATCH v2 0/10] mm: SLxB cleaning and trace accuracy improvement

[Christoph]

Please pull from pekkas tree on github or use next (check if the back and forth with patchset in and out has been resolved first).

Re: [PATCH v2 0/10] mm: SLxB cleaning and trace accuracy improvement

[Ezequiel Garcia]

Hi Pekka,

On Sat, Sep 8, 2012 at 5:49 PM, Ezequiel Garcia <elezegarcia@gmail.com> wrote:
> Hi everyone,
>
> This is the second spin of my patchset to clean SLxB and improve kmem
> trace events accuracy.
>
> For this v2, the most relevant stuff is:
>
> I've dropped two patches that were not very well received:
> Namely this two are now gone:
>   mm, slob: Use only 'ret' variable for both slob object and returned pointer
>   mm, slob: Trace allocation failures consistently
> I believe consistency is important but perhaps this is just me being paranoid.
>
> There's a lot of dumb movement and renaming. This might seem stupid
> (and maybe it is) but it's necessary to create some common code between SLAB
> and SLUB, and then factor it out.
>
> Also, there's a patch to add a new option to disable gcc auto-inlining.
> I know we hate to add new options, but this is necessary to get
> accurate call site
> traces. Plus, the option is in "Kernel Hacking", so it's for kernel
> developers only.
>
> This work is part of CELF Workgroup Project:
> "Kernel_dynamic_memory_allocation_tracking_and_reduction" [1]
>
> Feedback, comments, suggestions are very welcome.
>
> Ezequiel Garcia (10):
>  mm: Factor SLAB and SLUB common code
>  mm, slub: Rename slab_alloc() -> slab_alloc_node() to match SLAB
>  mm, slab: Rename __cache_alloc() -> slab_alloc()
>  mm, slab: Match SLAB and SLUB kmem_cache_alloc_xxx_trace() prototype
>  mm, slab: Replace 'caller' type, void* -> unsigned long
>  mm, util: Use dup_user to duplicate user memory
>  mm, slob: Add support for kmalloc_track_caller()
>  mm, slab: Remove silly function slab_buffer_size()
>  mm, slob: Use NUMA_NO_NODE instead of -1
>  Makefile: Add option CONFIG_DISABLE_GCC_AUTOMATIC_INLINING
>

Can you pick patches 2, 3, 4, and 5?
Namely only those related to SLOB and to simple cleanups.

I'll redo SLAB/SLUB commonization, as Christoph requested.

Thanks,
Ezequiel.