Re: [PATCH] mmc: dw_mmc: return -EILSEQ for EBE and SBE error

From: Jaehoon Chung
Date: Sun Aug 21 2016 - 22:52:05 EST


Hi Shawn,

On 08/20/2016 12:38 PM, Shawn Lin wrote:
> The following log we found indicate the fact that dw_mmc
> didn't treat EBE or SBE as a similar problem as CRC error.
> -EIO is quite not informative as it may indicate that the device
> is broken rather than that of tuning stuff.
>
> ...
> [ 89.057226] bcmsdh_sdmmc: Failed to Read byte F1:@0x1001f=ff, Err: -5
> [ 89.058811] bcmsdh_sdmmc: Failed to Read byte F1:@0x1001f=ff, Err: -5
> [ 89.059415] bcmsdh_sdmmc: Failed to Read byte F1:@0x1000e=ff, Err: -84
> [ 89.254248] dwmmc_rockchip fe310000.dwmmc: Successfully tuned phase to 199
> [ 89.273912] dhd_set_suspend: Remove extra suspend setting
> [ 89.274478] dhd_enable_packet_filter: enter, value = 0
> 64 bytes from 112.90.83.112: icmp_seq=24 ttl=53 time=1321 ms
> 64 bytes from 112.90.83.112: icmp_seq=25 ttl=53 time=319 ms
> 64 bytes from 112.90.83.112: icmp_seq=26 ttl=53 time=69.8 ms
> 64 bytes from 112.90.83.112: icmp_seq=27 ttl=53 time=37.5 ms
> ...
>
> For the host, when failing to sample cmd's response due to
> tuning stuff, we still return -EIO as it's quite vague to figure
> out whether it related to signal or just the broken devices, especially
> for the card type detection when booting kernel as all things go well
> but the cmd set used.
>
> But for the data phase, if receiving the cmd's response which
> carrys data transfer, we should have more confidence that it
> is very probably related to the tuning stuff.

s/carrys/carries

>
> Just as the log shown above, we sometimes suffer too much
> this kind of pain as the dw_mmc return -EIO for the case, so
> mmc-core will not do retune and caller drivers like bcm's wifi
> driver, still retry the failure more and more untile dw_mmc
> finally generate CRC.

s/untile/until ?

>
> Adrian suggested that drivers who care the specific cases should
> call mmc_retune_needed rather than doing it in mmc core. It makes
> sense but I'm considering that -EILSEQ actually means illegal sequence
> , so we use it for CRC cases. Meanwhile, SBE/EBE indicate the illegal
> sequence of start bit or end bit for data0~7. So I realize that we should
> use -EILSEQ for them both as well CRC cases.

Make sense. SBE/EBE are illegal sequence error.
Thanks!

Best Regards,
Jaehoon Chung

>
> Suggested-by: Adrian Hunter <adrian.hunter@xxxxxxxxx>
> Signed-off-by: Shawn Lin <shawn.lin@xxxxxxxxxxxxxx>
> ---
>
> drivers/mmc/host/dw_mmc.c | 4 ++--
> 1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/mmc/host/dw_mmc.c b/drivers/mmc/host/dw_mmc.c
> index 32380d5..b36a4f5 100644
> --- a/drivers/mmc/host/dw_mmc.c
> +++ b/drivers/mmc/host/dw_mmc.c
> @@ -1691,11 +1691,11 @@ static int dw_mci_data_complete(struct dw_mci *host, struct mmc_data *data)
> data->error = -ETIMEDOUT;
> } else if (host->dir_status ==
> DW_MCI_RECV_STATUS) {
> - data->error = -EIO;
> + data->error = -EILSEQ;
> }
> } else {
> /* SDMMC_INT_SBE is included */
> - data->error = -EIO;
> + data->error = -EILSEQ;
> }
>
> dev_dbg(host->dev, "data error, status 0x%08x\n", status);
>