Skip to content

Further optimizing SLAB/SLUB bulking

Jesper Dangaard Brouer linux-mm netdev 2 versions – In review

Get series with b4
b4 am 20150929154605.14465.98995.stgit@canyon

Versions newest first · linked by author and title

  1. v4 Latest6 patches 23 repliesIn review
  2. v17 patches 11 repliesSuperseded

Patches in v4 No applies recorded yet

#Subject and filesSizeRepliesApplied
1/6slub: create new ___slab_alloc function that can be called with irqs disabledmm/slub.c+29 -150 replies
2/6slub: Avoid irqoff/on in bulk allocationmm/slub.c+11 -130 replies
3/6slub: mark the dangling ifdef #else of CONFIG_SLUB_DEBUGmm/slub.c+1 -10 replies
4/6slab: implement bulking for SLAB allocatormm/slab.c+62 -250 replies
5/6slub: support for bulk free with SLUB freelistsmm/slub.c+69 -184 replies
5/6slub: support for bulk free with SLUB freelistsmm/slub.c+67 -1816 replies
6/6slub: optimize bulk slowpath free by detached freelistmm/slub.c+78 -303 replies

Reviews trailers per patch

5 of 7 patches have a Reviewed-by or Acked-by.

PatchChristoph Lameter
1/6
2/6
3/6A-b
4/6A-b
5/6A-b
5/6A-b
6/6A-b

Trailers given in replies are filled; those carried in the patches from earlier versions are outlined with dashes.

Read the v4 thread

Cover letter [MM PATCH V4 0/6] Further optimizing SLAB/SLUB bulking

Most important part of this patchset is the introducing of what I call detached freelist, for improving SLUB performance of object freeing in the "slowpath" of kmem_cache_free_bulk.

Tagging patchset with "V4" to avoid confusion with "V2":
(V2) http://thread.gmane.org/gmane.linux.kernel.mm/137469

Addressing comments from:
("V3") http://thread.gmane.org/gmane.linux.kernel.mm/139268

I've added Christoph Lameter's ACKs from prev review.
* Only patch 5 is changed significantly and needs review.
* Benchmarked, performance is the same

Notes for patches:
* First two patches (from Christoph) are already in AKPM MMOTS.
* Patch 3 is trivial
* Patch 4 is a repost, implements bulking for SLAB.
- http://thread.gmane.org/gmane.linux.kernel.mm/138220
* Patch 5 and 6 are the important patches
- Patch 5 handle "freelists" in slab_free() and __slab_free().
- Patch 6 intro detached freelists, and significant performance improvement

Patches should be ready for the MM-tree, as I'm now handling kmem debug support.

Based on top of commit 519f526d39 in net-next, but I've tested it applies on top of mmotm-2015-09-18-16-08.

The benchmarking tools are avail here:
https://github.com/netoptimizer/prototype-kernel/tree/master/kernel/mm
See: slab_bulk_test0{1,2,3}.c

This was joint work with Alexander Duyck while still at Red Hat.

This patchset is part of my network stack use-case. I'll post the network side of the patchset as soon as I've cleaned it up, rebased it on net-next and re-run all the benchmarks.

---

Christoph Lameter (2):
slub: create new ___slab_alloc function that can be called with irqs disabled
slub: Avoid irqoff/on in bulk allocation

Jesper Dangaard Brouer (4):
slub: mark the dangling ifdef #else of CONFIG_SLUB_DEBUG
slab: implement bulking for SLAB allocator
slub: support for bulk free with SLUB freelists
slub: optimize bulk slowpath free by detached freelist

 mm/slab.c |   87 ++++++++++++++------
 mm/slub.c |  263 +++++++++++++++++++++++++++++++++++++++++++------------------
 2 files changed, 249 insertions(+), 101 deletions(-)

--
Best regards,
Jesper Dangaard Brouer
MSc.CS, Principal Kernel Engineer at Red Hat
Author of http://www.iptv-analyzer.org
LinkedIn: http://www.linkedin.com/in/brouer

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