]> asedeno.scripts.mit.edu Git - linux.git/commit
staging: vc04_services: Replace dmac_map_area with dmac_map_sg
authorMichael Zoran <mzoran@crowfest.net>
Wed, 26 Oct 2016 02:23:27 +0000 (19:23 -0700)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Thu, 27 Oct 2016 13:19:37 +0000 (15:19 +0200)
commitcf9caf1929882b66922aee698e99e6c8f357bee5
tree268c72ee77bbf5c6264a847caab2f5dc2c0f78b6
parentc4b370efc7fba6a489eda2cafd41109acbce81fb
staging: vc04_services: Replace dmac_map_area with dmac_map_sg

The original arm implementation uses dmac_map_area which is not
portable.  Replace it with an architecture neutral version
which uses dma_map_sg.

As you can see that for larger page sizes, the dma_map_sg
implementation is faster then the original unportable dma_map_area
implementation.

Test                       dmac_map_area   dma_map_page dma_map_sg
vchiq_test -b 4 10000      51us/iter       76us/iter    76us
vchiq_test -b 8 10000      70us/iter       82us/iter    91us
vchiq_test -b 16 10000     94us/iter       118us/iter   121us
vchiq_test -b 32 10000     146us/iter      173us/iter   187us
vchiq_test -b 64 10000     263us/iter      328us/iter   299us
vchiq_test -b 128 10000    529us/iter      631us/iter   595us
vchiq_test -b 256 10000    2285us/iter     2275us/iter  2001us
vchiq_test -b 512 10000    4372us/iter     4616us/iter  4123us

For message sizes >= 64KB, dma_map_sg is faster then dma_map_page.

For message size >= 256KB, the dma_map_sg is the fastest
implementation.

"Normal" messages sizes should be about 1MB which is beyond
the length that this change shows a speed increase.

This is v2 of the patch which includes extra WARN_ONs and
incorporates feedback from Eric Anholt <eric@anholt.net>.

Signed-off-by: Michael Zoran <mzoran@crowfest.net>
Reviewed-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drivers/staging/vc04_services/interface/vchiq_arm/vchiq_2835_arm.c