Re: [PATCH RFC 3/4] dm-stripe: limit chunk_sectors to the stripe size

[John Garry <john.g.garry@oracle.com>]

On 09/06/2025 16:19, Mikulas Patocka wrote:
>> diff --git a/drivers/md/dm-table.c b/drivers/md/dm-table.c
>> index 24a857ff6d0b..4f1f7173740c 100644
>> --- a/drivers/md/dm-table.c
>> +++ b/drivers/md/dm-table.c
>> @@ -430,6 +430,10 @@ static int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
>>   		return 0;
>>   	}
>>   
>> +	/* For striped types, limit the chunk_sectors to the chunk size */
>> +	if (dm_target_supports_striped(ti->type))
>> +		limits->chunk_sectors = len >> SECTOR_SHIFT;
>> +
> len is already in sectors, so why do we shift it right?

Actually what I am passing is not the proper value at all. len holds the 
sc->stripe_width, and that seems to be md dev size / # stripes. I think 
that we want chunk_size, right?

> 
> Could this logic be moved to the function stripe_io_hints, so that we
> don't have to add a new flag for that and that we don't have to modify the
> generic dm code?
> 

That would be better. I am going to have to change 
blk_stack_atomic_writes_limits() to work for that, but I think that code 
needs to change anyway if the bottom device has its own chunk_sectors 
(as Nilay mentioned about 4/4).

Thanks,
John