Re: [PATCH v2 1/2] block: define new rqf_flags RQF_XFER_SINGLE_BLK

[Jens Axboe <axboe@kernel.dk>]

On 3/24/26 1:11 PM, Bin Liu wrote:
> Hi Jens,
> 
> On Tue, Mar 24, 2026 at 12:48:11PM -0600, Jens Axboe wrote:
>> On 3/24/26 8:34 AM, Bin Liu wrote:
>>> This new flag will be used to handle multi-block transfer failure, in
>>> which case single-block transfer will be used.
>>
>> Maybe I missed this in previous discussions, but why can't this flag go
>> into struct mmc_queue_req?
> 
> You didn't miss anything. We didn't discuss about it until now.
> 
> I added the flag in struct request because it already has rq_flags to
> host all the flag bits. But I don't see struct mmc_queue_req has any
> member for this flag purpose. Do you mean we should add a new flag
> member into struct mmc_queue_req?

It's a bad idea to add driver specific flag to the core flags,
particularly when you already have per-rq data associated with the
driver. Then it should surely go there. The block layer doesn't care
about RQF_XFER_SINGLE_BLK, that's strictly a device property. Hence it
should go in the driver, NOT in the block layer.

So yes, add a bool or flags or whatever you need on the driver side.

-- 
Jens Axboe