[U-Boot] [PATCH v2] ppc/85xx: PIO Support for FSL eSDHC Controller Driver

On some Freescale SoC Internal DMA of eSDHC controller has bug.
So PIO Mode has introduced to do data transfer using CPU. In PIO mode data transfer performance will be degraded by a large extent.
Note: In PIO mode multiple block read/write requires delay to complete the transfer.
Signed-off-by: Dipen Dudhat dipen.dudhat@freescale.com --- drivers/mmc/fsl_esdhc.c | 87 +++++++++++++++++++++++++++++++++++++++++++++- include/fsl_esdhc.h | 2 + 2 files changed, 87 insertions(+), 2 deletions(-)
diff --git a/drivers/mmc/fsl_esdhc.c b/drivers/mmc/fsl_esdhc.c index 27e4c48..7db8c0c 100644 --- a/drivers/mmc/fsl_esdhc.c +++ b/drivers/mmc/fsl_esdhc.c @@ -73,8 +73,10 @@ uint esdhc_xfertyp(struct mmc_cmd *cmd, struct mmc_data *data) uint xfertyp = 0;
if (data) { - xfertyp |= XFERTYP_DPSEL | XFERTYP_DMAEN; - + xfertyp |= XFERTYP_DPSEL; +#ifndef CONFIG_SYS_FSL_ESDHC_USE_PIO + xfertyp |= XFERTYP_DMAEN; +#endif if (data->blocks > 1) { xfertyp |= XFERTYP_MSBSEL; xfertyp |= XFERTYP_BCEN; @@ -98,12 +100,88 @@ uint esdhc_xfertyp(struct mmc_cmd *cmd, struct mmc_data *data) return XFERTYP_CMD(cmd->cmdidx) | xfertyp; }
+#ifdef CONFIG_SYS_FSL_ESDHC_USE_PIO +/* + * PIO Read/Write Mode reduce the performace as DMA is not used in this mode. + */ +static int +esdhc_pio_read_write(struct mmc *mmc, struct mmc_data *data) +{ + struct fsl_esdhc *regs = mmc->priv; + uint blocks; + char *buffer; + uint databuf; + uint size; + uint irqstat; + uint timeout; + + if (data->flags & MMC_DATA_READ) { + blocks = data->blocks; + buffer = data->dest; + while (blocks) { + timeout = MAX_TIMEOUT; + size = data->blocksize; + irqstat = in_be32(®s->irqstat); + while (!(in_be32(®s->prsstat) & PRSSTAT_BREN) + && --timeout); + if (timeout <= 0) { + printf("\nData Read Failed in PIO Mode."); + return TIMEOUT; + } + while (size && (!(irqstat & IRQSTAT_TC))) { + udelay(100); + irqstat = in_be32(®s->irqstat); + databuf = in_le32(®s->datport); + *((uint *)buffer) = databuf; + buffer += 4; + size -= 4; + } + blocks--; + } + } else { + blocks = data->blocks; + buffer = data->src; + while (blocks) { + timeout = MAX_TIMEOUT; + size = data->blocksize; + irqstat = in_be32(®s->irqstat); + while (!(in_be32(®s->prsstat) & PRSSTAT_BWEN) + && --timeout); + if (timeout <= 0) { + printf("\nData Write Failed in PIO Mode."); + return TIMEOUT; + } + while (size && (!(irqstat & IRQSTAT_TC))) { + udelay(100); + databuf = *((uint *)buffer); + buffer += 4; + size -= 4; + irqstat = in_be32(®s->irqstat); + out_le32(®s->datport, databuf); + } + blocks--; + } + } +} +#endif + static int esdhc_setup_data(struct mmc *mmc, struct mmc_data *data) { uint wml_value; int timeout; struct fsl_esdhc *regs = mmc->priv;
+#ifdef CONFIG_SYS_FSL_ESDHC_USE_PIO + if (!(data->flags & MMC_DATA_READ)) { + if ((in_be32(®s->prsstat) & PRSSTAT_WPSPL) == 0) { + printf("\nThe SD card is locked. " + "Can not write to a locked card.\n\n"); + return TIMEOUT; + } + out_be32(®s->dsaddr, (u32)data->src); + } else + out_be32(®s->dsaddr, (u32)data->dest); +#else wml_value = data->blocksize/4;
if (data->flags & MMC_DATA_READ) { @@ -125,6 +203,7 @@ static int esdhc_setup_data(struct mmc *mmc, struct mmc_data *data) }
out_be32(®s->wml, wml_value); +#endif
out_be32(®s->blkattr, data->blocks << 16 | data->blocksize);
@@ -217,6 +296,9 @@ esdhc_send_cmd(struct mmc *mmc, struct mmc_cmd *cmd, struct mmc_data *data)
/* Wait until all of the blocks are transferred */ if (data) { +#ifdef CONFIG_SYS_FSL_ESDHC_USE_PIO + esdhc_pio_read_write(mmc, data); +#else do { irqstat = in_be32(®s->irqstat);
@@ -227,6 +309,7 @@ esdhc_send_cmd(struct mmc *mmc, struct mmc_cmd *cmd, struct mmc_data *data) return TIMEOUT; } while (!(irqstat & IRQSTAT_TC) && (in_be32(®s->prsstat) & PRSSTAT_DLA)); +#endif }
out_be32(®s->irqstat, -1); diff --git a/include/fsl_esdhc.h b/include/fsl_esdhc.h index 89b8304..5948d37 100644 --- a/include/fsl_esdhc.h +++ b/include/fsl_esdhc.h @@ -86,6 +86,7 @@ #define PRSSTAT_CDPL (0x00040000) #define PRSSTAT_CINS (0x00010000) #define PRSSTAT_BREN (0x00000800) +#define PRSSTAT_BWEN (0x00000400) #define PRSSTAT_DLA (0x00000004) #define PRSSTAT_CICHB (0x00000002) #define PRSSTAT_CIDHB (0x00000001) @@ -117,6 +118,7 @@ #define XFERTYP_DMAEN 0x00000001
#define CINS_TIMEOUT 1000 +#define MAX_TIMEOUT 100000
#define DSADDR 0x2e004

Dear Dipen Dudhat,
In message 1252589856-4970-1-git-send-email-dipen.dudhat@freescale.com you wrote:
On some Freescale SoC Internal DMA of eSDHC controller has bug.
So PIO Mode has introduced to do data transfer using CPU. In PIO mode data transfer performance will be degraded by a large extent.
Note: In PIO mode multiple block read/write requires delay to complete the transfer.
Does that mean you just delay, i. e. you do not check for some ready state?
...
--- a/include/fsl_esdhc.h +++ b/include/fsl_esdhc.h @@ -86,6 +86,7 @@ #define PRSSTAT_CDPL (0x00040000) #define PRSSTAT_CINS (0x00010000) #define PRSSTAT_BREN (0x00000800) +#define PRSSTAT_BWEN (0x00000400) #define PRSSTAT_DLA (0x00000004) #define PRSSTAT_CICHB (0x00000002) #define PRSSTAT_CIDHB (0x00000001) @@ -117,6 +118,7 @@ #define XFERTYP_DMAEN 0x00000001
#define CINS_TIMEOUT 1000 +#define MAX_TIMEOUT 100000
MAX_TIMEOUT is a very generic name, please chose a more specific one so interactions with other code are prevented. (Also, TIMEOUT itself already means a maximum time, so MAX_TIMEOUT is a tautology.)
Best regards,
Wolfgang Denk

Hi Wolfgang,
There are some IP specific requirements to get the transaction complete in PIO mode.
Thanks Dipen
-----Original Message----- From: Wolfgang Denk [mailto:wd@denx.de] Sent: Wednesday, September 23, 2009 12:12 AM To: Dudhat Dipen-B09055 Cc: u-boot@lists.denx.de Subject: Re: [U-Boot] [PATCH v2] ppc/85xx: PIO Support for FSL eSDHC Controller Driver
Dear Dipen Dudhat,
In message 1252589856-4970-1-git-send-email-dipen.dudhat@freescale.com you wrote:
On some Freescale SoC Internal DMA of eSDHC controller has bug.
So PIO Mode has introduced to do data transfer using CPU. In PIO mode data transfer performance will be degraded by a large
extent.
Note: In PIO mode multiple block read/write requires delay to complete the
transfer.
Does that mean you just delay, i. e. you do not check for some ready state?
...
--- a/include/fsl_esdhc.h +++ b/include/fsl_esdhc.h @@ -86,6 +86,7 @@ #define PRSSTAT_CDPL (0x00040000) #define PRSSTAT_CINS (0x00010000) #define PRSSTAT_BREN (0x00000800) +#define PRSSTAT_BWEN (0x00000400) #define PRSSTAT_DLA (0x00000004) #define PRSSTAT_CICHB (0x00000002) #define PRSSTAT_CIDHB (0x00000001) @@ -117,6 +118,7 @@ #define XFERTYP_DMAEN 0x00000001
#define CINS_TIMEOUT 1000 +#define MAX_TIMEOUT 100000
MAX_TIMEOUT is a very generic name, please chose a more specific one so interactions with other code are prevented. (Also, TIMEOUT itself already means a maximum time, so MAX_TIMEOUT is a tautology.)
Dipen, Will update that.
Best regards,
Wolfgang Denk

On Thu, Sep 10, 2009 at 9:37 PM, Dipen Dudhat dipen.dudhat@freescale.com wrote:
while (size && (!(irqstat & IRQSTAT_TC))) {
udelay(100);
irqstat = in_be32(®s->irqstat);
databuf = in_le32(®s->datport);
*((uint *)buffer) = databuf;
buffer += 4;
size -= 4;
}
Is the udelay(100) the recommended delay time by the chip manual or empirical value? Can you add a comment before the udelay?
Regards, Bin Meng

Hi Bin,
We can know the block transfer complete using IRQSTAT(Transfer Complete). But reading & writing in PIO mode takes time for byte by byte transfers and there is no way to poll that transfer, that's why I have added delay there.
I can add comment like, * Wait before last byte transfer complete *\ Is that ok ??
Regards, Dipen
-----Original Message----- From: Bin Meng [mailto:bmeng.cn@gmail.com] Sent: Monday, September 28, 2009 4:41 AM To: Dudhat Dipen-B09055 Cc: u-boot@lists.denx.de Subject: Re: [U-Boot] [PATCH v2] ppc/85xx: PIO Support for FSL eSDHC Controller Driver
On Thu, Sep 10, 2009 at 9:37 PM, Dipen Dudhat dipen.dudhat@freescale.com wrote:
while (size && (!(irqstat & IRQSTAT_TC))) {
udelay(100);
irqstat = in_be32(®s->irqstat);
databuf = in_le32(®s->datport);
*((uint *)buffer) = databuf;
buffer += 4;
size -= 4;
}
Is the udelay(100) the recommended delay time by the chip manual or empirical value? Can you add a comment before the udelay?
Regards, Bin Meng

Hi Dipen,
On Thu, Oct 1, 2009 at 12:48 PM, Dudhat Dipen-B09055 Dipen.Dudhat@freescale.com wrote:
Hi Bin,
We can know the block transfer complete using IRQSTAT(Transfer Complete). But reading & writing in PIO mode takes time for byte by byte transfers and there is no way to poll that transfer, that's why I have added delay there.
Yes, I understand there is no such a register in Freescale eSDHC that provides status check for PIO mode. I am just wondering where the delay value 100 us comes from. Is it the required minimum value?
I can add comment like, * Wait before last byte transfer complete *\ Is that ok ??
Sure, adding a comment line would help a lot.
Regards, Bin Meng

-----Original Message----- From: Bin Meng [mailto:bmeng.cn@gmail.com] Sent: Monday, October 05, 2009 1:59 PM To: Dudhat Dipen-B09055 Cc: u-boot@lists.denx.de Subject: Re: [U-Boot] [PATCH v2] ppc/85xx: PIO Support for FSL eSDHC Controller Driver
Hi Dipen,
On Thu, Oct 1, 2009 at 12:48 PM, Dudhat Dipen-B09055 Dipen.Dudhat@freescale.com wrote:
Hi Bin,
We can know the block transfer complete using IRQSTAT(Transfer Complete). But reading & writing in PIO mode takes time for byte by byte transfers and there is no way to poll that transfer, that's why I have added delay there.
Yes, I understand there is no such a register in Freescale eSDHC that provides status check for PIO mode. I am just wondering where the delay value 100 us comes from. Is it the required minimum value?
Dipen, Yes it's a minimum value. Less than that it hangs and read/write fails in PIO Mode.
I can add comment like, * Wait before last byte transfer complete *\ Is that ok ??
Sure, adding a comment line would help a lot.
Regards, Bin Meng
Thanks Dipen
participants (4)
-
Bin Meng
-
Dipen Dudhat
-
Dudhat Dipen-B09055
-
Wolfgang Denk