Re: [PATCH] mmc: core: need retune if error value is -EIO

From: Adrian Hunter
Date: Thu Aug 18 2016 - 04:47:22 EST


On 18/08/16 11:10, Shawn Lin wrote:
> On 2016/8/18 15:19, Adrian Hunter wrote:
>> On 04/08/16 11:30, Shawn Lin wrote:
>>> We need to do retune if receiving -EIO, otherwise we
>>> could see debug dump like:
>>>
>>> [ 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
>>> ...
>>>
>>> In this case we see dw_mmc finally enter retune process, but
>>> if this patch is applied, we could save more time to make it
>>> work. Also many host drivers will generate -EIO, so this patch
>>> can also prevent them from failing to enter retune process.
>>
>> The current logic is re-tune on CRC errors. -EIO isn't informative
>> and drivers can use it for cases that clearly are not related to tuning.
>>
>
> It actually relates to tuning. If failing to sample data or cmd-resp,
> the controller generate timeout interrupt in principle rather than
> explicit CRC ones.

So that is driver-specific. So that driver needs to use mmc_retune_needed()
in that case.

> explicit CRC ones. So it makes sense for them to return -EIO instead of
> -EILSEQ as it's hard for the driver to understand what was happening,
> crc? device is broken? ...
>
>> A driver can call mmc_retune_needed() itself in other cases.
>
> It's no so clear to this retune design as if the driver already knows
> it's a CRC, it will generate -EILSEQ and let core do tuning again, so
> it means they don't need to call mmc_return_needed in their drivers.
> Unless let drivers return CRC in any cases of CRC or timeout, but that
> may make core do tuning more frequently even if not relating to tuning
> which seems a little painful.
>
> So it looks quite vague to me.:)
>
>>
>>>
>>> Signed-off-by: Shawn Lin <shawn.lin@xxxxxxxxxxxxxx>
>>> ---
>>>
>>> drivers/mmc/core/core.c | 3 ++-
>>> 1 file changed, 2 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/drivers/mmc/core/core.c b/drivers/mmc/core/core.c
>>> index e55cde6..18d0af5 100644
>>> --- a/drivers/mmc/core/core.c
>>> +++ b/drivers/mmc/core/core.c
>>> @@ -133,7 +133,8 @@ void mmc_request_done(struct mmc_host *host, struct
>>> mmc_request *mrq)
>>> /* Flag re-tuning needed on CRC errors */
>>> if ((cmd->opcode != MMC_SEND_TUNING_BLOCK &&
>>> cmd->opcode != MMC_SEND_TUNING_BLOCK_HS200) &&
>>> - (err == -EILSEQ || (mrq->sbc && mrq->sbc->error == -EILSEQ) ||
>>> + (err == -EILSEQ || err == -EIO ||
>>> + (mrq->sbc && mrq->sbc->error == -EILSEQ) ||
>>> (mrq->data && mrq->data->error == -EILSEQ) ||
>>> (mrq->stop && mrq->stop->error == -EILSEQ)))
>>> mmc_retune_needed(host);
>>>
>>
>>
>>
>>
>
>