Re: [PATCH 0/4] numa, iommu/smmu: IOMMU/SMMU driver optimization for NUMA systems

From: John Garry
Date: Wed Aug 22 2018 - 12:08:40 EST

On 22/08/2018 15:56, Robin Murphy wrote:
Hi John,

On 22/08/18 14:44, John Garry wrote:
On 21/09/2017 09:59, Ganapatrao Kulkarni wrote:
Adding numa aware memory allocations used for iommu dma allocation and
memory allocated for SMMU stream tables, page walk tables and command

With this patch, iperf testing on ThunderX2, with 40G NIC card on
NODE 1 PCI shown same performance(around 30% improvement) as NODE 0.

Ganapatrao Kulkarni (4):
mm: move function alloc_pages_exact_nid out of __meminit
numa, iommu/io-pgtable-arm: Use NUMA aware memory allocation for smmu
translation tables
iommu/arm-smmu-v3: Use NUMA memory allocations for stream tables and
comamnd queues
iommu/dma, numa: Use NUMA aware memory allocations in

drivers/iommu/arm-smmu-v3.c | 57
drivers/iommu/dma-iommu.c | 17 +++++++------
drivers/iommu/io-pgtable-arm.c | 4 ++-
include/linux/gfp.h | 2 +-
mm/page_alloc.c | 3 ++-
5 files changed, 67 insertions(+), 16 deletions(-)

Hi Ganapatrao,

Have you any plans for further work on this patchset? I have not seen
anything since this v1 was posted+discussed.

Hi Robin,

Thanks for the info. I thought I remembered 4b12 but couldn't put my finger on it.

Looks like I ended up doing the version of the io-pgtable change that I
suggested here, which was merged recently (4b123757eeaa). Patch #3
should also be effectively obsolete now since the SWIOTLB/dma-direct
rework (21f237e4d085). Apparently I also started reworking patch #4 in
my tree at some point but sidelined it - I think that was at least
partly due to another thread[1] which made it seem less clear-cut
whether this is always the right thing to do.

Right, so #4 seems less straightforward and not directly related to IOMMU driver anyway.