Thread (9 messages) 9 messages, 5 authors, 2017-03-31
STALE3347d

[PATCH 0/3] IOVA allocation improvements for iommu-dma

From: Nate Watterson <hidden>
Date: 2017-03-22 17:43:16
Also in: linux-iommu

On 2017-03-15 09:33, Robin Murphy wrote:
Hi all,
Hi Robin,
Here's the first bit of lock contention removal to chew on - feedback
welcome! Note that for the current users of the io-pgtable framework,
this is most likely to simply push more contention onto the io-pgtable
lock, so may not show a great improvement alone. Will and I both have
rough proof-of-concept implementations of lock-free io-pgtable code
which we need to sit down and agree on at some point, hopefullt fairly
soon.

I've taken the opportunity to do a bit of cleanup and refactoring
within the series to make the final state of the code nicer, but the
diffstat still turns out surprisingly reasonable in the end - it would
actually be negative but for the new comments!

Magnus, Shimoda-san, the first two patches should be of interest as 
they
constitute the allocation rework I mentioned a while back[1] - if you
still need to implement that scary workaround, this should make it
simple to hook IPMMU-specific calls into the alloc and free paths, and
let the driver take care of the details internally.
I've tested your patches on a QDF2400 platform and generally
see modest improvements in iperf/fio performance. As you
suspected would happen, contention has indeed moved to the
io-pgtable lock. I am looking forward to testing with the
lock-free io-pgtable implementation, however I suspect that
there will still be contention issues acquiring the (SMMUv3)
cmdq lock on the unmap path.

Reviewed/Tested-by: Nate Watterson [off-list ref]
Robin.

[1]:https://lists.linuxfoundation.org/pipermail/iommu/2017-January/020189.html

Robin Murphy (3):
  iommu/dma: Convert to address-based allocation
  iommu/dma: Clean up MSI IOVA allocation
  iommu/dma: Plumb in the per-CPU IOVA caches

 drivers/iommu/dma-iommu.c | 176 
++++++++++++++++++++++++----------------------
 1 file changed, 90 insertions(+), 86 deletions(-)
-- 
Qualcomm Datacenter Technologies, Inc. on behalf of Qualcomm 
Technologies, Inc.
Qualcomm Technologies, Inc. is a member of the Code Aurora Forum, a 
Linux
Foundation Collaborative Project.
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help