[PATCH 6/6] null_blk: Add support for power_of_2 emulation to the null blk device

[Pankaj Raghav]

On 2022-03-09 05:09, Damien Le Moal wrote:
> On 3/9/22 01:53, Pankaj Raghav wrote:
>> power_of_2(PO2) emulation support is added to the null blk device to
>> measure performance.
>>
>> Callbacks are added in the hotpaths that require PO2 emulation specific
>> behaviour to reduce the impact on exisiting path.
>>
>> The power_of_2 emulation support is wired up for both the request and
>> the bio queue mode and it is automatically enabled when the given zone
>> size is non-power_of_2.
> 
> This does not make any sense. Why would you want to add power of 2 zone
> size emulation to nullblk ? Just set the zone size to be a power of 2...
> 
> If this is for test purpose, then use QEMU. These changes make no sense
> to me here.
> 
I see your point but this was mainly added to measure the performance impact.

I ran the conformance test with different configurations in QEMU but I don't
think QEMU would be a preferred option to measure performance, especially, if we
want to account for changes we did to the hot path with a indirection. 

As ZNS drives are not available in retail stores, this patch also provides a way
for the community to reproduce the performance analysis that we did without needing
a real device.

> A change that would make sense in the context of this series is to allow
> for setting a zoned null_blk device zone size to a non power of 2 size.
This is not possible with the block layer expecting the zone sizes to be po2.
A null blk device with non po2 zone size will only work with the emulation that is
added as a part of this patch.

As I said before, once we relax the block layer requirement, then we could allow
non po2 zone sizes without a lot of changes to the null blk device.
> But this series does not actually deal with that. So do not touch this
> driver please.
> 
If you really think it doesn't belong here, then I can take it out in the next
revision.
-- 
Regards,
Pankaj