]> nv-tegra.nvidia Code Review - linux-2.6.git/commit
mxs-dma : rewrite the last parameter of mxs_dma_prep_slave_sg()
authorHuang Shijie <b32955@freescale.com>
Thu, 16 Feb 2012 06:17:33 +0000 (14:17 +0800)
committerDavid Woodhouse <David.Woodhouse@intel.com>
Mon, 26 Mar 2012 23:37:28 +0000 (00:37 +0100)
commit921de864b7c6413f15224d8f5e677541e8e1ac6d
tree325815e4a65a26b961796314fdb0b2cd6e0b9975
parent3946860409130038ef6e0e5c50f2203053eae2b7
mxs-dma : rewrite the last parameter of mxs_dma_prep_slave_sg()

[1] Background :
    The GPMI does ECC read page operation with a DMA chain consist of three DMA
    Command Structures. The middle one of the chain is used to enable the BCH,
    and read out the NAND page.

    The WAIT4END(wait for command end) is a comunication signal between
    the GPMI and MXS-DMA.

[2] The current DMA code sets the WAIT4END bit at the last one, such as:

    +-----+               +-----+                      +-----+
    | cmd | ------------> | cmd | ------------------>  | cmd |
    +-----+               +-----+                      +-----+
                                                          ^
                                                          |
                                                          |
                                                     set WAIT4END here

    This chain works fine in the mx23/mx28.

[3] But in the new GPMI version (used in MX50/MX60), the WAIT4END bit should
    be set not only at the last DMA Command Structure,
    but also at the middle one, such as:

    +-----+               +-----+                      +-----+
    | cmd | ------------> | cmd | ------------------>  | cmd |
    +-----+               +-----+                      +-----+
                             ^                            ^
                             |                            |
                             |                            |
                        set WAIT4END here too        set WAIT4END here

    If we do not set WAIT4END, the BCH maybe stalls in "ECC reading page" state.
    In the next ECC write page operation, a DMA-timeout occurs.
    This has been catched in the MX6Q board.

[4] In order to fix the bug, rewrite the last parameter of mxs_dma_prep_slave_sg(),
    and use the dma_ctrl_flags:
    ---------------------------------------------------------
      DMA_PREP_INTERRUPT : append a new DMA Command Structrue.
      DMA_CTRL_ACK       : set the WAIT4END bit for this DMA Command Structure.
    ---------------------------------------------------------

[5] changes to the relative drivers:
    <1> For mxs-mmc driver, just use the new flags, do not change any logic.
    <2> For gpmi-nand driver, and use the new flags to set the DMA
        chain, especially for ecc read page.

Acked-by: Shawn Guo <shawn.guo@linaro.org>
Signed-off-by: Huang Shijie <b32955@freescale.com>
Acked-by: Vinod Koul <vinod.koul@linux.intel.com>
Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>
drivers/dma/mxs-dma.c
drivers/mmc/host/mxs-mmc.c
drivers/mtd/nand/gpmi-nand/gpmi-lib.c