bounce limit in mmc?

bounce limit in mmc?

[Christoph Hellwig]

Hi all, any idea why the mmc driver defaults to bouncing highmem
pages if the device doesn't have a DMA mask?  highmem is defined
different on different architectures or even based on config options,
so it seems like a very odd choice.

Re: bounce limit in mmc?

[Shawn Lin]

Hi Christoph,

On 2017/8/29 1:48, Christoph Hellwig wrote:
> Hi all, any idea why the mmc driver defaults to bouncing highmem
> pages if the device doesn't have a DMA mask?  highmem is defined
> different on different architectures or even based on config options,
> so it seems like a very odd choice.

Not sure why mmc stack defaults to use BLK_BOUNCE_HIGH as it seems that
it has been there when mmc was added into linux.

I *guess* this is a copy-and-paste from other block drivers, especially
see drivers/memstick/core/ms_block.c which does actual the same thing.

And that raises another question that if mmc stack doesn't defaults to
boucing highmem pages but host drivers don't provide DMA mask either, so
what should we use for saftly limiting bouncing for DMA capable?

> --
> To unsubscribe from this list: send the line "unsubscribe linux-mmc" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> 
> 
> 

Re: bounce limit in mmc?

[Pierre Ossman]

On 30/08/17 15:09, Avri Altman wrote:
> better ask the author -  (mmc: bounce requests for simple hosts) and
> (mmc_block: bounce buffer highmem support).
> 

Sorry for the late reply. This got lost in my inbox.

> On Wed, Aug 30, 2017 at 9:14 AM, Shawn Lin <shawn.lin@rock-chips.com> wrote:
> 
>> Hi Christoph,
>>
>> On 2017/8/29 1:48, Christoph Hellwig wrote:
>>
>>> Hi all, any idea why the mmc driver defaults to bouncing highmem
>>> pages if the device doesn't have a DMA mask?  highmem is defined
>>> different on different architectures or even based on config options,
>>> so it seems like a very odd choice.
>>>
>>
>> Not sure why mmc stack defaults to use BLK_BOUNCE_HIGH as it seems that
>> it has been there when mmc was added into linux.
>>
>> I *guess* this is a copy-and-paste from other block drivers, especially
>> see drivers/memstick/core/ms_block.c which does actual the same thing.
>>
>> And that raises another question that if mmc stack doesn't defaults to
>> boucing highmem pages but host drivers don't provide DMA mask either, so
>> what should we use for saftly limiting bouncing for DMA capable?
>>

I'm afraid it's way too long ago, so I don't really remember. git blame 
seems to suggest it has been there since before I took over though. RMK 
was the one in charge before me.

The only thing I do remember is that the original bounce stuff was done 
for the wbsd driver, which is a ISA bus device so it had to work around 
the limitations of that system.

Regards
-- 
Pierre Ossman