| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Message-Id: <1405917465-18185-1-git-send-email-apelete@seketeli.net>
Date: Mon, 21 Jul 2014 06:37:43 +0200
From: Apelete Seketeli <apelete@...eteli.net>
To: Chris Ball <chris@...ntf.net>,
Ulf Hansson <ulf.hansson@...aro.org>,
H Hartley Sweeten <hsweeten@...ionengravers.com>,
Lars-Peter Clausen <lars@...afoo.de>,
Wei Yongjun <yongjun_wei@...ndmicro.com.cn>,
Alex Smith <alex.smith@...tec.com>,
Laurent Pinchart <laurent.pinchart+renesas@...asonboard.com>,
linux-mmc@...r.kernel.org
Cc: linux-kernel@...r.kernel.org
Subject: [PATCH v6 0/2] Use DMA for data transfers in JZ4740 MMC driver
Hello,
MMC driver for JZ4740 SoC is currently relying on PIO mode only for
data transfers.
The patches that come as a follow-up of this message allow the use of
DMA for data transfers.
Changes since v5:
- added a new patch to this series, on top of the previous one, to
prepare next dma transfer in parallel with current transfer,
- updated the benchmarks in this cover letter for the aforementioned
patch.
Changes since v4:
- prefix functions with jz4740_mmc_ instead of jz4740_ to keep
consistency with the rest of the code,
- test (host->sg_len == 0) instead of (host->sg_len < 0) after
dma_map_sg() in jz4740_mmc_start_dma_transfer() according to
dmaengine.txt instructions.
Changes since v3:
- use DMA engine device instead of MMC device in dma_map_sg() and
dma_unmap_sg() as the memory transfers will be performed by the
DMA engine,
- remove unnecessary mem_res == NULL check in jz4740_mmc_probe()
since devm_ioremap_resource() handles the mem_res == NULL case and
print a message,
- added a check to host->use_dma in jz4740_mmc_remove() to avoid
calling jz4740_release_dma_channels() unnecessarily.
Changes since v2:
- declare sg_len member of struct jz4740_mmc_host as int instead of
unsigned int.
Changes since v1:
- remove blank line added by mistake in jz_mmc_prepare_data_transfer().
According to the following PIO vs DMA benchmarks there seems to be a
slight improvement in transfer speed with the Ben NanoNote booting
from SD card, while load average seems to be roughly on par.
No noticeable improvement in the DMA vs DMA + MMC asynchronous
requests case. The latter change should help reduce the impact of DMA
preparation overhead on fast SD cards performance though:
* With DMA + MMC asynchronous requests:
Test cases | root@...NanoNote:/# uptime | root@...NanoNote:/# time zcat root/fedora-16.iso.gz > /dev/null && uptime
-----------|----------------------------------------------------------------------------------------------------------------------------------
Test run 1 | 00:01:35 up 1 min, load average: 1.31, 0.44, 0.15 | 00:06:46 up 6 min, load average: 2.53, 1.75, 0.81
Test run 2 | 00:10:09 up 1 min, load average: 1.26, 0.42, 0.14 | 00:15:20 up 6 min, load average: 2.46, 1.73, 0.80
Test run 3 | 00:31:22 up 1 min, load average: 1.30, 0.44, 0.15 | 00:36:33 up 6 min, load average: 2.45, 1.73, 0.80
-----------|----------------------------------------------------------------------------------------------------------------------------------
Average | 1 min, load average: 1.29, 0.43, 0.14 | 6 min, load average: 2.48, 1.73, 0.80
* With DMA:
Test cases | root@...NanoNote:/# uptime | root@...NanoNote:/# time zcat root/fedora-16.iso.gz > /dev/null && uptime
-----------|----------------------------------------------------------------------------------------------------------------------------------
Test run 1 | 00:20:55 up 1 min, load average: 1.26, 0.42, 0.14 | 00:26:10 up 6 min, load average: 2.89, 1.94, 0.89
Test run 2 | 00:30:22 up 1 min, load average: 1.16, 0.38, 0.13 | 00:35:34 up 6 min, load average: 2.68, 1.86, 0.85
Test run 3 | 00:39:56 up 1 min, load average: 1.16, 0.38, 0.13 | 00:45:06 up 6 min, load average: 2.57, 1.76, 0.81
-----------|----------------------------------------------------------------------------------------------------------------------------------
Average | 1 min, load average: 1.19, 0.39, 0.13 | 6 min, load average: 2.71, 1.85, 0.85
* With PIO:
Test cases | root@...NanoNote:/# uptime | root@...NanoNote:/# time zcat root/fedora-16.iso.gz > /dev/null && uptime
----------------------------------------------------------------------------------------------------------------------------------------------
Test run 1 | 00:50:47 up 1 min, load average: 1.42, 0.49, 0.17 | 00:56:52 up 7 min, load average: 2.47, 2.00, 0.98
Test run 2 | 01:00:19 up 1 min, load average: 1.21, 0.39, 0.14 | 01:06:29 up 7 min, load average: 2.45, 1.96, 0.96
Test run 3 | 01:11:27 up 1 min, load average: 1.15, 0.36, 0.12 | 01:17:33 up 7 min, load average: 2.63, 2.01, 0.97
-----------|----------------------------------------------------------------------------------------------------------------------------------
Average | 1 min, load average: 1.26, 0.41, 0.14 | 7 min, load average: 2.52, 1.99, 0.97
Changes were rebased on top of linux master branch, built and tested
successfully.
The following changes since commit 9a3c414:
Linux 3.16-rc6
are available in the git repository at:
git://git.seketeli.net/~apelete/linux.git jz4740-mmc-dma
Apelete Seketeli (2):
mmc: jz4740: add dma infrastructure for data transfers
mmc: jz4740: prepare next dma transfer in parallel with current
transfer
drivers/mmc/host/jz4740_mmc.c | 268 +++++++++++++++++++++++++++++++++++++++--
1 file changed, 260 insertions(+), 8 deletions(-)
--
1.7.10.4
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Powered by blists - more mailing lists