Re: [PATCH] block: make sure big bio is splitted into at most 256 bvecs

[Ming Lei]

On Wed, Apr 6, 2016 at 8:59 AM, Shaohua Li <shli@xxxxxx> wrote:
> On Tue, Apr 05, 2016 at 04:45:55PM -0800, Kent Overstreet wrote:
>> On Tue, Apr 05, 2016 at 05:41:47PM -0700, Shaohua Li wrote:
>> > On Tue, Apr 05, 2016 at 04:36:04PM -0800, Kent Overstreet wrote:
>> > > On Tue, Apr 05, 2016 at 05:30:07PM -0700, Shaohua Li wrote:
>> > > > this one:
>> > > > http://marc.info/?l=linux-kernel&m=145926976808760&w=2
>> > >
>> > > Ah. that patch won't actually fix the bug, since md isn't using
>> > > blk_default_limits, it's using blk_set_stacking_limits().
>> >
>> > Not really, the limit is set by under layer disk not md, otherwise it
>> > should be BLK_SAFE_MAX_SECTORS, but the reported bio has 2560 sectors.
>> > blk_set_stacking_limits() will use it.
>>
>> What? Well, that could should just be deleted, there's no reason anymore for md
>> to care about the queue limits of the devices underneath it.
>>
>> Regardless, using BLK_DEF_MAX_SECTORS to limit # of pages in the biovec is
>> _crazy_. Why would you even do that? We have a separate field in queue limits
>> for # max segments, use it.
>
> We don't limit the max segments in blk_queue_max_segments(), but we can
> add. On the other hand, limit max segments to 256 could be a problem,
> because bvec page isn't always 4k, this might make some bio smaller.

It is nothing with max segments limit, it is about max sectors limit:
think about one big bio which includes 2Mbytes, then 512 bvecs are required,
but the 2M buffer can be continuous physically.


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