Re: [PATCH 3/9] iio: backend: add debugFs interface

["Nuno Sá" <noname.nuno@gmail.com>]

On Sat, 2024-07-20 at 10:43 +0100, Jonathan Cameron wrote:
> On Thu, 18 Jul 2024 16:32:33 +0200
> Nuno Sá <noname.nuno@gmail.com> wrote:
> 
> > On Tue, 2024-07-16 at 19:14 +0100, Jonathan Cameron wrote:
> > > On Tue, 9 Jul 2024 13:14:30 +0200
> > > Nuno Sa <nuno.sa@analog.com> wrote:
> > >   
> > > > This adds a basic debugfs interface for backends. Two new ops are being
> > > > added:
> > > > 
> > > >  * debugfs_reg_access: Analogous to the core IIO one but for backend
> > > >    devices.
> > > >  * debugfs_print_chan_status: One useful usecase for this one is for
> > > >    testing test tones in a digital interface and "ask" the backend to
> > > >    dump more details on why a test tone might have errors.
> > > > 
> > > > Signed-off-by: Nuno Sa <nuno.sa@analog.com>  
> > > Debugfs deserved docs as well as sysfs.
> > > Same place in Documentation/ABI/
> > > 
> > > Obviously we've neglected this in the past, but nice to do it right
> > > nor new stuff.
> > >   
> > 
> > I see. So you mean adding debugfs-iio?
> 
> Probably debugfs-iio-backend for this stuff, though we should have
> a more general doc as well.
> 
> > 
> > There's one thing I'm not sure though... I'm contemplating the case where
> > one device
> > may have multiple backends in which case I'm doing:
> > 
> > back->name = name;
> > 
> > where name comes from FW (DT usually). That obviously means the interface
> > won't be
> > always consistent which I guess it's not a real problem for debugfs?
> > 
> > How would the interface look in the file? Something like?
> > 
> > /sys/kernel/debug/iio/iio:deviceX/<backend_name>_direct_reg_access
> 
> That's fine - fairly common sort of thing to see in debugfs.
> 
> > 
> > Or should we think in a more reliable naming? One option that came to mind
> > is
> > 
> > /sys/kernel/debug/iio/iio:deviceX/backendY_direct_reg_access
> If you were doing this it might be better as a directory.
> e.g. backendY/direct_reg_access
> > 
> > where Y would be the corresponding index in io-backend-names.
> > 
> > One thing not optimal with the above would be identifying the actual backend
> > device.
> > It would then maybe make sense having a 'backend_name' interface which I
> > think is
> > likely too much just for this?
> It kind of depends on your expected usecase.  These are in debugfs so there
> is an assumption they aren't a 'normal operation' thing.  So if they
> are going to typically be poked by a user, then complex file names are fine.
> If it's going to be scripted, then stable names something like
> backendY/name
> backendY/direct_reg_access etc
> would be easier to use.
> 

Yeah, I think the main usage (the one I do at least) is for the user to directly
play and poke with this. However I don't think that the scripting usecase  to be
that crazy (or unlikely) and I do like stable things :). Also liked your
suggestion about grouping the interfaces in a backendY directory. We'll likely
need an iio_backend_info struct interface for backends to pass in when
registering. Maybe too much for a debug interface but this kind of 'info'
structure may very well be something we'll need in the future. Anyways, I think
I'll give this a try for v2 so we can see how it looks like in practise.

> I'm not bothered as much about consistency of this debug interface as I would
> be about sysfs, so up to you (or other reviewers) for which you prefer.
> 

Yeah, I was kind of expecting that (no one should blindly rely on debugFS) :)

- Nuno Sá