This patch can get the benefit of performance.
Looks good to me.
Tested-by: Jaehoon Chung <jh80.ch...@samsung.com>
On 01/26/2012 04:32 PM, Seungwon Jeon wrote:
> This patch implements pre_req and post_req in dw_mmc
> to support asynchronous mmc request.
>
> Signed-off-by: Seungwon Jeon <tgih....@samsung.com>
> ---
> Changes in v2:
> Consider system DMA case as well as IDMAC.
>
> NOTE:
> Performance gains the following.
> Sequential read and write improve 23% and 5% respectively.
>
> drivers/mmc/host/dw_mmc.c | 129 +++++++++++++++++++++++++++++++++++---------
> 1 files changed, 102 insertions(+), 27 deletions(-)
>
> diff --git a/drivers/mmc/host/dw_mmc.c b/drivers/mmc/host/dw_mmc.c
> index 0e34279..5c91acb 100644
> --- a/drivers/mmc/host/dw_mmc.c
> +++ b/drivers/mmc/host/dw_mmc.c
> @@ -297,14 +297,24 @@ static void dw_mci_stop_dma(struct dw_mci *host)
> }
>
> #ifdef CONFIG_MMC_DW_IDMAC
> +static int dw_mci_get_dma_dir(struct mmc_data *data)
> +{
> + if (data->flags & MMC_DATA_WRITE)
> + return DMA_TO_DEVICE;
> + else
> + return DMA_FROM_DEVICE;
> +}
> +
> static void dw_mci_dma_cleanup(struct dw_mci *host)
> {
> struct mmc_data *data = host->data;
>
> if (data)
> - dma_unmap_sg(&host->pdev->dev, data->sg, data->sg_len,
> - ((data->flags & MMC_DATA_WRITE)
> - ? DMA_TO_DEVICE : DMA_FROM_DEVICE));
> + if (!data->host_cookie)
> + dma_unmap_sg(&host->pdev->dev,
> + data->sg,
> + data->sg_len,
> + dw_mci_get_dma_dir(data));
> }
>
> static void dw_mci_idmac_stop_dma(struct dw_mci *host)
> @@ -420,26 +430,15 @@ static int dw_mci_idmac_init(struct dw_mci *host)
> return 0;
> }
>
> -static struct dw_mci_dma_ops dw_mci_idmac_ops = {
> - .init = dw_mci_idmac_init,
> - .start = dw_mci_idmac_start_dma,
> - .stop = dw_mci_idmac_stop_dma,
> - .complete = dw_mci_idmac_complete_dma,
> - .cleanup = dw_mci_dma_cleanup,
> -};
> -#endif /* CONFIG_MMC_DW_IDMAC */
> -
> -static int dw_mci_submit_data_dma(struct dw_mci *host, struct mmc_data *data)
> +static int dw_mci_pre_dma_transfer(struct dw_mci *host,
> + struct mmc_data *data,
> + bool next)
> {
> struct scatterlist *sg;
> - unsigned int i, direction, sg_len;
> - u32 temp;
> -
> - host->using_dma = 0;
> + unsigned int i, sg_len;
>
> - /* If we don't have a channel, we can't do DMA */
> - if (!host->use_dma)
> - return -ENODEV;
> + if (!next && data->host_cookie)
> + return data->host_cookie;
>
> /*
> * We don't do DMA on "complex" transfers, i.e. with
> @@ -448,6 +447,7 @@ static int dw_mci_submit_data_dma(struct dw_mci *host,
> struct mmc_data *data)
> */
> if (data->blocks * data->blksz < DW_MCI_DMA_THRESHOLD)
> return -EINVAL;
> +
> if (data->blksz & 3)
> return -EINVAL;
>
> @@ -456,15 +456,88 @@ static int dw_mci_submit_data_dma(struct dw_mci *host,
> struct mmc_data *data)
> return -EINVAL;
> }
>
> - host->using_dma = 1;
> + sg_len = dma_map_sg(&host->pdev->dev,
> + data->sg,
> + data->sg_len,
> + dw_mci_get_dma_dir(data));
> + if (sg_len == 0)
> + return -EINVAL;
>
> - if (data->flags & MMC_DATA_READ)
> - direction = DMA_FROM_DEVICE;
> - else
> - direction = DMA_TO_DEVICE;
> + if (next)
> + data->host_cookie = sg_len;
> +
> + return sg_len;
> +}
> +
> +static struct dw_mci_dma_ops dw_mci_idmac_ops = {
> + .init = dw_mci_idmac_init,
> + .start = dw_mci_idmac_start_dma,
> + .stop = dw_mci_idmac_stop_dma,
> + .complete = dw_mci_idmac_complete_dma,
> + .cleanup = dw_mci_dma_cleanup,
> +};
> +#else
> +static int dw_mci_pre_dma_transfer(struct dw_mci *host,
> + struct mmc_data *data,
> + bool next)
> +{
> + return -ENOSYS;
> +}
> +#endif /* CONFIG_MMC_DW_IDMAC */
> +
> +static void dw_mci_pre_req(struct mmc_host *mmc,
> + struct mmc_request *mrq,
> + bool is_first_req)
> +{
> + struct dw_mci_slot *slot = mmc_priv(mmc);
> + struct mmc_data *data = mrq->data;
> +
> + if (!slot->host->use_dma || !data)
> + return;
> +
> + if (data->host_cookie) {
> + data->host_cookie = 0;
> + return;
> + }
> +
> + if (dw_mci_pre_dma_transfer(slot->host, mrq->data, 1) < 0)
> + data->host_cookie = 0;
> +}
> +
> +static void dw_mci_post_req(struct mmc_host *mmc,
> + struct mmc_request *mrq,
> + int err)
> +{
> + struct dw_mci_slot *slot = mmc_priv(mmc);
> + struct mmc_data *data = mrq->data;
> +
> + if (!slot->host->use_dma || !data)
> + return;
> +
> + if (data->host_cookie)
> + dma_unmap_sg(&slot->host->pdev->dev,
> + data->sg,
> + data->sg_len,
> + dw_mci_get_dma_dir(data));
> + data->host_cookie = 0;
> +}
> +
> +static int dw_mci_submit_data_dma(struct dw_mci *host, struct mmc_data *data)
> +{
> + int sg_len;
> + u32 temp;
>
> - sg_len = dma_map_sg(&host->pdev->dev, data->sg, data->sg_len,
> - direction);
> + host->using_dma = 0;
> +
> + /* If we don't have a channel, we can't do DMA */
> + if (!host->use_dma)
> + return -ENODEV;
> +
> + sg_len = dw_mci_pre_dma_transfer(host, data, 0);
> + if (sg_len < 0)
> + return sg_len;
> +
> + host->using_dma = 1;
>
> dev_vdbg(&host->pdev->dev,
> "sd sg_cpu: %#lx sg_dma: %#lx sg_len: %d\n",
> @@ -795,6 +868,8 @@ static void dw_mci_enable_sdio_irq(struct mmc_host *mmc,
> int enb)
>
> static const struct mmc_host_ops dw_mci_ops = {
> .request = dw_mci_request,
> + .pre_req = dw_mci_pre_req,
> + .post_req = dw_mci_post_req,
> .set_ios = dw_mci_set_ios,
> .get_ro = dw_mci_get_ro,
> .get_cd = dw_mci_get_cd,
--
To unsubscribe from this list: send the line "unsubscribe linux-mmc" in
the body of a message to majord...@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
