Re: [gentoo-user] m.2 nvme stick not what I was expecting at all.

[Michael <confabulate@kintzios.com>]

[-- Attachment #1: Type: text/plain, Size: 1654 bytes --]

On Tuesday, 25 February 2025 19:18:57 Greenwich Mean Time Rich Freeman wrote:
> On Tue, Feb 25, 2025 at 2:00 PM Michael <confabulate@kintzios.com> wrote:
> > Unless I'm wrong there is/was a speed penalty when accessing a fs over
> > FUSE. Anyway, I was configuring kernel 6.12.16-gentoo today and came
> > across this:
> > 
> > CONFIG_FUSE_PASSTHROUGH
> > 
> > More details here:
> > 
> > https://lwn.net/Articles/832430/
> > 
> > It looks quite promising.
> 
> It isn't 100% clear when this will work.  This seems to be about
> skipping the FUSE userspace driver to directly connect an application
> to the ultimate backing store, but this assumes the kernel even
> implements the backing store.  I get that this might often be the
> case, but I can imagine that in a lot of FUSE applications there is no
> linux-native filesystem involved.
> 
> I'm not surprised to hear that FUSE performance isn't great - it just
> isn't seen as a mainstream way to mount things.  On a microkernel
> there is no such thing as a kernel-native filesystem implementation,
> so the kernel maintainers obviously need to optimize for this use
> case.  I imagine that they will still have many context switches to
> deal with.

From what I read in this paper, the FUSE driver gets the OK from the FUSE 
daemon for accessing the fs directly, instead of having to route each read/
write via the FUSE daemon:

https://source.android.com/docs/core/storage/fuse-passthrough

Perhaps this is some Android specific setup to allow userspace management of 
attached devices and their fs, instead of the wider Linux desktop environment.  
:-/


[-- Attachment #2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 833 bytes --]