[PATCH] mmc: meson-gx: set max block count to 256

Kevin Hilman khilman at baylibre.com
Mon Jan 30 08:53:56 PST 2017


Heiner Kallweit <hkallweit1 at gmail.com> writes:

> Am 27.01.2017 um 19:28 schrieb Kevin Hilman:
>> Heiner Kallweit <hkallweit1 at gmail.com> writes:
>> 
>>> So far max_blk_count isn't set what results in a default of value 8
>>> to be used (PAGE_SIZE / block size).
>>>
>>> Set max_blk_count to 256 as used in the uboot driver.
>>>
>>> Signed-off-by: Heiner Kallweit <hkallweit1 at gmail.com>
>> 
>> Thanks for the patch. A minor comment below:
>> 
>>> ---
>>>  drivers/mmc/host/meson-gx-mmc.c | 1 +
>>>  1 file changed, 1 insertion(+)
>>>
>>> diff --git a/drivers/mmc/host/meson-gx-mmc.c b/drivers/mmc/host/meson-gx-mmc.c
>>> index 3308b5b6..030425be 100644
>>> --- a/drivers/mmc/host/meson-gx-mmc.c
>>> +++ b/drivers/mmc/host/meson-gx-mmc.c
>>> @@ -799,6 +799,7 @@ static int meson_mmc_probe(struct platform_device *pdev)
>>>  		goto free_host;
>>>  	}
>>>  
>>> +	mmc->max_blk_count = 256;
>> 
>> A step in the right direction, but the # blocks is actually a 9-bit
>> field (c.f. length field of cmd_cfg part of descriptor (bits 8:0).
>> 
>> So this should be more like:
>> 
>> 	/* blk count is 'length' field of cmd_cfg part of descriptor */
>> 	mmc->max_blk_count = CMD_CFG_LENGTH_MASK + 1;
>> 
>> I tested that change locally, so after changing that, please resend and
>> feel free to add:
>> 
> max_blk_count = 0 means unlimited number of blocks. Therefore I think
> we have to go with: mmc->max_blk_count = CMD_CFG_LENGTH_MASK

Ah, yes.  Good catch.

Kevin

> And I figured out later that we also have to set mmc->max_req_size.
>
> In my tests the system never read / wrote more than 8 blocks in a row,
> most likely because the upper block layers use 4k blocks.
>
> Therefore setting max_blk_count doesn't really provide a benefit but
> I think it's cleaner to expose the actual chip limit.
>
>
>> Reviewed-by: Kevin Hilman <khilman at baylibre.com>
>> Tested-by: Kevin Hilman <khilman at baylibre.com>
>> 
>> Kevin
>> 



More information about the linux-amlogic mailing list