Re: [PATCH v6 01/26] fuse: Fix dax truncate/punch_hole fault path

On Thu, Feb 06, 2025 at 03:59:03PM +0100, Albert Esteve wrote:

Hi!

On Thu, Feb 6, 2025 at 3:30 PM Stefan Hajnoczi [off-list ref] wrote:

quoted

On Thu, Feb 06, 2025 at 08:37:07AM -0500, Vivek Goyal wrote:

quoted

And then there are challenges at QEMU level. virtiofsd needs additional
vhost-user commands to implement DAX and these never went upstream in
QEMU. I hope these challenges are sorted at some point of time.

Albert Esteve has been working on QEMU support:
https://lore.kernel.org/qemu-devel/20240912145335.129447-1-aesteve@redhat.com/ (local)

He has a viable solution. I think the remaining issue is how to best
structure the memory regions. The reason for slow progress is not
because it can't be done, it's probably just because this is a
background task.

It is partially that, indeed. But what has me blocked for now on posting the
next version is that I was reworking a bit the MMAP strategy.
Following David comments, I am relying more on RAMBlocks and
subregions for mmaps. But this turned out more difficult than anticipated.

I hope I can make it work this month and then post the next version.
If there are no major blockers/reworks, further iterations on the
patch shall go smoother.

I have a separate patch for the vhost-user spec which could
iterate faster, if that'd help.

Let's see if anyone needs the vhost-user spec extension now. Otherwise
it seems fine to merge it together with the implementation of that spec.

Stefan