diff options
author | Russell King <rmk+kernel@arm.linux.org.uk> | 2011-01-11 22:35:53 +0300 |
---|---|---|
committer | Russell King <rmk+kernel@arm.linux.org.uk> | 2011-02-04 16:25:49 +0300 |
commit | c8ebae37034c0ead62eb4df8ef88e999ddb8d5cf (patch) | |
tree | c9925f03a9c627d7408ef483d7920e25a927e633 /drivers/mmc/host/mmci.h | |
parent | 51d4375dd72f352594f1a4f1d7598bf9a75b8dfe (diff) | |
download | linux-c8ebae37034c0ead62eb4df8ef88e999ddb8d5cf.tar.xz |
ARM: mmci: add dmaengine-based DMA support
Based on a patch from Linus Walleij.
Add dmaengine based support for DMA to the MMCI driver, using the
Primecell DMA engine interface. The changes over Linus' driver are:
- rename txsize_threshold to dmasize_threshold, as this reflects the
purpose more.
- use 'mmci_dma_' as the function prefix rather than 'dma_mmci_'.
- clean up requesting of dma channels.
- don't release a single channel twice when it's shared between tx and rx.
- get rid of 'dma_enable' bool - instead check whether the channel is NULL.
- detect incomplete DMA at the end of a transfer. Some DMA controllers
(eg, PL08x) are unable to be configured for scatter DMA and also listen
to all four DMA request signals [BREQ,SREQ,LBREQ,LSREQ] from the MMCI.
They can do one or other but not both. As MMCI uses LBREQ/LSREQ for the
final burst/words, PL08x does not transfer the last few words.
- map and unmap DMA buffers using the DMA engine struct device, not the
MMCI struct device - the DMA engine is doing the DMA transfer, not us.
- avoid double-unmapping of the DMA buffers on MMCI data errors.
- don't check for negative values from the dmaengine tx submission
function - Dan says this must never fail.
- use new dmaengine helper functions rather than using the ugly function
pointers directly.
- allow DMA code to be fully optimized away using dma_inprogress() which
is defined to constant 0 if DMA engine support is disabled.
- request maximum segment size from the DMA engine struct device and
set this appropriately.
- removed checking of buffer alignment - the DMA engine should deal with
its own restrictions on buffer alignment, not the individual DMA engine
users.
- removed setting DMAREQCTL - this confuses some DMA controllers as it
causes LBREQ to be asserted for the last seven transfers, rather than
six SREQ and one LSREQ.
- removed burst setting - the DMA controller should not burst past the
transfer size required to complete the DMA operation.
Tested-by: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Diffstat (limited to 'drivers/mmc/host/mmci.h')
-rw-r--r-- | drivers/mmc/host/mmci.h | 13 |
1 files changed, 13 insertions, 0 deletions
diff --git a/drivers/mmc/host/mmci.h b/drivers/mmc/host/mmci.h index 164ce060fc1f..ec9a7bc6d0df 100644 --- a/drivers/mmc/host/mmci.h +++ b/drivers/mmc/host/mmci.h @@ -148,8 +148,10 @@ struct clk; struct variant_data; +struct dma_chan; struct mmci_host { + phys_addr_t phybase; void __iomem *base; struct mmc_request *mrq; struct mmc_command *cmd; @@ -179,5 +181,16 @@ struct mmci_host { struct sg_mapping_iter sg_miter; unsigned int size; struct regulator *vcc; + +#ifdef CONFIG_DMA_ENGINE + /* DMA stuff */ + struct dma_chan *dma_current; + struct dma_chan *dma_rx_channel; + struct dma_chan *dma_tx_channel; + +#define dma_inprogress(host) ((host)->dma_current) +#else +#define dma_inprogress(host) (0) +#endif }; |