Re: 64k and 4k page size inter-op?

[Daniel Pocock <daniel@pocock.pro>]

On 14/09/2020 09:06, Nikolay Borisov wrote:
> 
> 
> On 14.09.20 г. 1:49 ч., Daniel Pocock wrote:
>>
>> I recently ran into trouble moving btrfs volumes between an amd64 host
>> (4k page size) and powerpc64el (64k page size)
>>
>> As far as I can tell, the only option to ensure compatibility between
>> the two hosts is to
>>
>> a) recompile the powerpc64el kernel for a 4k page size,
>>
>> b) re-format any volumes created with 64k sectorsize so they have 4k instead
>>
>> Is there any other reasonable way to proceed?
>>
>> Are there any downsides to using the 4k page size on powerpc64el?
> 
> There is currently work underway to allow btrfs to work with subpage
> blocksizes i.e being able to mount a 4k filesystem on a 64k page
> filesystem. Currently you cannot do this so as you correctly deduced -
> you'd have to either equalize the page sizes of both systems or wait
> until this work lands in upstream.
> 

Some distributions are currently enabling the 64k page size by default
on their kernels (e.g. Debian, Fedora) for certain architectures like
powerpc64le

People may be surprised by this issue if they create a filesystem on one
of these hosts and try to use it on another host or vice versa.  With
Fedora about to enable btrfs by default, I'll raise it on their mailing
list.  Should they reconsider 64k page size on their kernels for
non-server builds?

Is the 64k page size and the limitation in btrfs going to have an
adverse impact on performance when working with small files, e.g.
compiling source code?

Another thing I noticed is the nouveau driver not working on machines
built like that, it doesn't support 64k either.  Maybe another reason
for kernels to have the 4k default for most users.