[PATCH] OMAP: HSMMC: Implement scatter-gather emulation

From: Juha Yrjola
Date: Fri Nov 14 2008 - 08:22:00 EST


Instead of using the bounce buffer, using scatter-gather emulation
(as in the OMAP1/2 MMC driver) removes the need of one extra memory
copy and improves performance.

Signed-off-by: Juha Yrjola <juha.yrjola@xxxxxxxxxxxxx>
---
drivers/mmc/host/omap_hsmmc.c | 172 ++++++++++++++++++++++------------------
1 files changed, 95 insertions(+), 77 deletions(-)

diff --git a/drivers/mmc/host/omap_hsmmc.c b/drivers/mmc/host/omap_hsmmc.c
index 1f84bd4..483e591 100644
--- a/drivers/mmc/host/omap_hsmmc.c
+++ b/drivers/mmc/host/omap_hsmmc.c
@@ -100,9 +100,6 @@
#define OMAP_MMC1_DEVID 0
#define OMAP_MMC2_DEVID 1

-#define OMAP_MMC_DATADIR_NONE 0
-#define OMAP_MMC_DATADIR_READ 1
-#define OMAP_MMC_DATADIR_WRITE 2
#define MMC_TIMEOUT_MS 20
#define OMAP_MMC_MASTER_CLOCK 96000000
#define DRIVER_NAME "mmci-omap-hs"
@@ -138,16 +135,14 @@ struct mmc_omap_host {
resource_size_t mapbase;
unsigned int id;
unsigned int dma_len;
- unsigned int dma_dir;
+ unsigned int dma_sg_idx;
unsigned char bus_mode;
- unsigned char datadir;
u32 *buffer;
u32 bytesleft;
int suspended;
int irq;
int carddetect;
int use_dma, dma_ch;
- int initstr;
int slot_id;
int dbclk_enabled;
int response_busy;
@@ -281,6 +276,15 @@ mmc_omap_start_command(struct mmc_omap_host *host, struct mmc_command *cmd,
OMAP_HSMMC_WRITE(host->base, CMD, cmdreg);
}

+static int
+mmc_omap_get_dma_dir(struct mmc_omap_host *host, struct mmc_data *data)
+{
+ if (data->flags & MMC_DATA_WRITE)
+ return DMA_TO_DEVICE;
+ else
+ return DMA_FROM_DEVICE;
+}
+
/*
* Notify the transfer complete to MMC core
*/
@@ -300,9 +304,7 @@ mmc_omap_xfer_done(struct mmc_omap_host *host, struct mmc_data *data)

if (host->use_dma && host->dma_ch != -1)
dma_unmap_sg(mmc_dev(host->mmc), data->sg, host->dma_len,
- host->dma_dir);
-
- host->datadir = OMAP_MMC_DATADIR_NONE;
+ mmc_omap_get_dma_dir(host, data));

if (!data->error)
data->bytes_xfered += data->blocks * (data->blksz);
@@ -352,13 +354,12 @@ static void mmc_dma_cleanup(struct mmc_omap_host *host, int errno)

if (host->use_dma && host->dma_ch != -1) {
dma_unmap_sg(mmc_dev(host->mmc), host->data->sg, host->dma_len,
- host->dma_dir);
+ mmc_omap_get_dma_dir(host, host->data));
omap_free_dma(host->dma_ch);
host->dma_ch = -1;
up(&host->sem);
}
host->data = NULL;
- host->datadir = OMAP_MMC_DATADIR_NONE;
}

/*
@@ -592,6 +593,55 @@ static irqreturn_t omap_mmc_cd_handler(int irq, void *dev_id)
return IRQ_HANDLED;
}

+static int mmc_omap_get_dma_sync_dev(struct mmc_omap_host *host,
+ struct mmc_data *data)
+{
+ int sync_dev;
+
+ if (data->flags & MMC_DATA_WRITE) {
+ if (host->id == OMAP_MMC1_DEVID)
+ sync_dev = OMAP24XX_DMA_MMC1_TX;
+ else
+ sync_dev = OMAP24XX_DMA_MMC2_TX;
+ } else {
+ if (host->id == OMAP_MMC1_DEVID)
+ sync_dev = OMAP24XX_DMA_MMC1_RX;
+ else
+ sync_dev = OMAP24XX_DMA_MMC2_RX;
+ }
+ return sync_dev;
+}
+
+static void mmc_omap_config_dma_params(struct mmc_omap_host *host,
+ struct mmc_data *data,
+ struct scatterlist *sgl)
+{
+ int blksz, nblk, dma_ch;
+
+ dma_ch = host->dma_ch;
+ if (data->flags & MMC_DATA_WRITE) {
+ omap_set_dma_dest_params(dma_ch, 0, OMAP_DMA_AMODE_CONSTANT,
+ (host->mapbase + OMAP_HSMMC_DATA), 0, 0);
+ omap_set_dma_src_params(dma_ch, 0, OMAP_DMA_AMODE_POST_INC,
+ sg_dma_address(sgl), 0, 0);
+ } else {
+ omap_set_dma_src_params(dma_ch, 0, OMAP_DMA_AMODE_CONSTANT,
+ (host->mapbase + OMAP_HSMMC_DATA), 0, 0);
+ omap_set_dma_dest_params(dma_ch, 0, OMAP_DMA_AMODE_POST_INC,
+ sg_dma_address(sgl), 0, 0);
+ }
+
+ blksz = host->data->blksz;
+ nblk = sg_dma_len(sgl) / blksz;
+
+ omap_set_dma_transfer_params(dma_ch, OMAP_DMA_DATA_TYPE_S32,
+ blksz / 4, nblk, OMAP_DMA_SYNC_FRAME,
+ mmc_omap_get_dma_sync_dev(host, data),
+ !(data->flags & MMC_DATA_WRITE));
+
+ omap_start_dma(dma_ch);
+}
+
/*
* DMA call back function
*/
@@ -605,6 +655,14 @@ static void mmc_omap_dma_cb(int lch, u16 ch_status, void *data)
if (host->dma_ch < 0)
return;

+ host->dma_sg_idx++;
+ if (host->dma_sg_idx < host->dma_len) {
+ /* Fire up the next transfer. */
+ mmc_omap_config_dma_params(host, host->data,
+ host->data->sg + host->dma_sg_idx);
+ return;
+ }
+
omap_free_dma(host->dma_ch);
host->dma_ch = -1;
/*
@@ -615,38 +673,28 @@ static void mmc_omap_dma_cb(int lch, u16 ch_status, void *data)
}

/*
- * Configure dma src and destination parameters
- */
-static int mmc_omap_config_dma_param(int sync_dir, struct mmc_omap_host *host,
- struct mmc_data *data)
-{
- if (sync_dir == 0) {
- omap_set_dma_dest_params(host->dma_ch, 0,
- OMAP_DMA_AMODE_CONSTANT,
- (host->mapbase + OMAP_HSMMC_DATA), 0, 0);
- omap_set_dma_src_params(host->dma_ch, 0,
- OMAP_DMA_AMODE_POST_INC,
- sg_dma_address(&data->sg[0]), 0, 0);
- } else {
- omap_set_dma_src_params(host->dma_ch, 0,
- OMAP_DMA_AMODE_CONSTANT,
- (host->mapbase + OMAP_HSMMC_DATA), 0, 0);
- omap_set_dma_dest_params(host->dma_ch, 0,
- OMAP_DMA_AMODE_POST_INC,
- sg_dma_address(&data->sg[0]), 0, 0);
- }
- return 0;
-}
-/*
* Routine to configure and start DMA for the MMC card
*/
static int
mmc_omap_start_dma_transfer(struct mmc_omap_host *host, struct mmc_request *req)
{
- int sync_dev, sync_dir = 0;
- int dma_ch = 0, ret = 0, err = 1;
+ int dma_ch = 0, ret = 0, err = 1, i;
struct mmc_data *data = req->data;

+ /* Sanity check: all the SG entries must be aligned by block size. */
+ for (i = 0; i < host->dma_len; i++) {
+ struct scatterlist *sgl;
+
+ sgl = data->sg + i;
+ if (sgl->length % data->blksz)
+ return -EINVAL;
+ }
+ if ((data->blksz % 4) != 0)
+ /* REVISIT: The MMC buffer increments only when MSB is written.
+ * Return error for blksz which is non multiple of four.
+ */
+ return -EINVAL;
+
/*
* If for some reason the DMA transfer is still active,
* we wait for timeout period and free the dma
@@ -665,49 +713,22 @@ mmc_omap_start_dma_transfer(struct mmc_omap_host *host, struct mmc_request *req)
return err;
}

- if (!(data->flags & MMC_DATA_WRITE)) {
- host->dma_dir = DMA_FROM_DEVICE;
- if (host->id == OMAP_MMC1_DEVID)
- sync_dev = OMAP24XX_DMA_MMC1_RX;
- else
- sync_dev = OMAP24XX_DMA_MMC2_RX;
- } else {
- host->dma_dir = DMA_TO_DEVICE;
- if (host->id == OMAP_MMC1_DEVID)
- sync_dev = OMAP24XX_DMA_MMC1_TX;
- else
- sync_dev = OMAP24XX_DMA_MMC2_TX;
- }
-
- ret = omap_request_dma(sync_dev, "MMC/SD", mmc_omap_dma_cb,
- host, &dma_ch);
+ ret = omap_request_dma(mmc_omap_get_dma_sync_dev(host, data), "MMC/SD",
+ mmc_omap_dma_cb,host, &dma_ch);
if (ret != 0) {
- dev_dbg(mmc_dev(host->mmc),
+ dev_err(mmc_dev(host->mmc),
"%s: omap_request_dma() failed with %d\n",
mmc_hostname(host->mmc), ret);
return ret;
}

host->dma_len = dma_map_sg(mmc_dev(host->mmc), data->sg,
- data->sg_len, host->dma_dir);
+ data->sg_len, mmc_omap_get_dma_dir(host, data));
host->dma_ch = dma_ch;
+ host->dma_sg_idx = 0;

- if (!(data->flags & MMC_DATA_WRITE))
- mmc_omap_config_dma_param(1, host, data);
- else
- mmc_omap_config_dma_param(0, host, data);
-
- if ((data->blksz % 4) == 0)
- omap_set_dma_transfer_params(dma_ch, OMAP_DMA_DATA_TYPE_S32,
- (data->blksz / 4), data->blocks, OMAP_DMA_SYNC_FRAME,
- sync_dev, sync_dir);
- else
- /* REVISIT: The MMC buffer increments only when MSB is written.
- * Return error for blksz which is non multiple of four.
- */
- return -EINVAL;
+ mmc_omap_config_dma_params(host, data, data->sg);

- omap_start_dma(dma_ch);
return 0;
}

@@ -757,7 +778,6 @@ mmc_omap_prepare_data(struct mmc_omap_host *host, struct mmc_request *req)
host->data = req->data;

if (req->data == NULL) {
- host->datadir = OMAP_MMC_DATADIR_NONE;
OMAP_HSMMC_WRITE(host->base, BLK, 0);
return 0;
}
@@ -766,9 +786,6 @@ mmc_omap_prepare_data(struct mmc_omap_host *host, struct mmc_request *req)
| (req->data->blocks << 16));
set_data_timeout(host, req);

- host->datadir = (req->data->flags & MMC_DATA_WRITE) ?
- OMAP_MMC_DATADIR_WRITE : OMAP_MMC_DATADIR_READ;
-
if (host->use_dma) {
ret = mmc_omap_start_dma_transfer(host, req);
if (ret != 0) {
@@ -1027,10 +1044,11 @@ static int __init omap_mmc_probe(struct platform_device *pdev)
else
host->dbclk_enabled = 1;

-#ifdef CONFIG_MMC_BLOCK_BOUNCE
- mmc->max_phys_segs = 1;
- mmc->max_hw_segs = 1;
-#endif
+ /* Since we do only SG emulation, we can have as many segs
+ * as we want. */
+ mmc->max_phys_segs = 1024;
+ mmc->max_hw_segs = 1024;
+
mmc->max_blk_size = 512; /* Block Length at max can be 1024 */
mmc->max_blk_count = 0xFFFF; /* No. of Blocks is 16 bits */
mmc->max_req_size = mmc->max_blk_size * mmc->max_blk_count;
--
1.5.6.3

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/