Re: [PATCH v2 5/5] vfio: ccw: add traceponits for interesting error paths

[Halil Pasic]

On 04/30/2018 05:03 PM, Cornelia Huck wrote:
> > I think the naming of this fctl thing is a bit cryptic,
> > but if we don't see this as ABI I'm fine with it -- can be improved.
> > What would be a better name? I was thinking along the lines accept_request.
> > (Bad error code would mean that the request did not get accepted. Good
> > code does not mean the requested function was performed successfully.)
> I think fctl is fine (if you don't understand what 'fctl' is, you're
> unlikely to understand it even if it were named differently.)

AFAIU this fctl is a bit more complicated than the normal fctl. But
better let sleeping dogs lie.

> > Also I think vfio_ccw_io_fctl with no zero error code would make sense
> > as dev_warn. If I were an admin looking into a problem I would very much
> > appreciate seeing something in the messages log (and not having to enable
> > tracing first). This point seems to be a good one for high level 'request gone
> > bad' kind of report. Opinions?
> I'd also exclude -EOPNOTSUPP (as this also might happen with e.g. a halt/clear enabled user space, which probes availability of halt/clear support by giving it a try once (yes, I really want to post my patches this week.))

I'm looking forward to the clear/halt. It hope it will help me understand
the big vfio-ccw picture better. There are still dark spots, but I don't
feel like doing something against this, as there is quite some activity
going on here -- and I don't want to hamper the efforts by binding resources.

Regards,
Halil