Hi Per,
We noticed on some of our systems if we ADMA or SDMA and a bounce buffer it is significantly faster then SDMA.
I believe ADMA will do large transfers. Another data point.
Philip
On May 7, 2011, at 12:14 PM, Per Forlin wrote:
How significant is the cache maintenance over head? It depends, the eMMC are much faster now compared to a few years ago and cache maintenance cost more due to multiple cache levels and speculative cache pre-fetch. In relation the cost for handling the caches have increased and is now a bottle neck dealing with fast eMMC together with DMA.
The intention for introducing none blocking mmc requests is to minimize the time between a mmc request ends and another mmc request starts. In the current implementation the MMC controller is idle when dma_map_sg and dma_unmap_sg is processing. Introducing none blocking mmc request makes it possible to prepare the caches for next job parallel with an active mmc request.
This is done by making the issue_rw_rq() none blocking. The increase in throughput is proportional to the time it takes to prepare (major part of preparations is dma_map_sg and dma_unmap_sg) a request and how fast the memory is. The faster the MMC/SD is the more significant the prepare request time becomes. Measurements on U5500 and Panda on eMMC and SD shows significant performance gain for for large reads when running DMA mode. In the PIO case the performance is unchanged.
There are two optional hooks pre_req() and post_req() that the host driver may implement in order to move work to before and after the actual mmc_request function is called. In the DMA case pre_req() may do dma_map_sg() and prepare the dma descriptor and post_req runs the dma_unmap_sg.
Details on measurements from IOZone and mmc_test: https://wiki.linaro.org/WorkingGroups/Kernel/Specs/StoragePerfMMC-async-req
Under consideration:
- Make pre_req and post_req private to core.c.
- Generalize implementation and make it available for SDIO.
Changes since v2:
- Fix compile warnings in core.c and block.c
- Simplify max transfer size in mmc_test
- set TASK_RUNNING in queue.c before issue_req()
Per Forlin (12): mmc: add none blocking mmc request function mmc: mmc_test: add debugfs file to list all tests mmc: mmc_test: add test for none blocking transfers mmc: add member in mmc queue struct to hold request data mmc: add a block request prepare function mmc: move error code in mmc_block_issue_rw_rq to a separate function. mmc: add a second mmc queue request member mmc: add handling for two parallel block requests in issue_rw_rq mmc: test: add random fault injection in core.c omap_hsmmc: use original sg_len for dma_unmap_sg omap_hsmmc: add support for pre_req and post_req mmci: implement pre_req() and post_req()
drivers/mmc/card/block.c | 493 +++++++++++++++++++++++++++-------------- drivers/mmc/card/mmc_test.c | 340 +++++++++++++++++++++++++++- drivers/mmc/card/queue.c | 180 ++++++++++------ drivers/mmc/card/queue.h | 31 ++- drivers/mmc/core/core.c | 132 ++++++++++- drivers/mmc/core/debugfs.c | 5 + drivers/mmc/host/mmci.c | 146 +++++++++++- drivers/mmc/host/mmci.h | 8 + drivers/mmc/host/omap_hsmmc.c | 90 +++++++- include/linux/mmc/core.h | 9 +- include/linux/mmc/host.h | 13 +- lib/Kconfig.debug | 11 + 12 files changed, 1174 insertions(+), 284 deletions(-)
-- 1.7.4.1
-- To unsubscribe from this list: send the line "unsubscribe linux-mmc" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html