On Di, 2018-06-12 at 21:57 +0200, Sebastian Andrzej Siewior wrote:
> > Yes, the atomic case should be rare. It will only happen on errors, and
> > IIUC that's only to work around issues caused by reporting errors back
> > to userspace without actually wanting to err out anyway.
>
> Yup. The missing part is if this was done to workaround a specific
> userland application or most/all of them.
Yes, If possible we should not regress in that regard.
> > I believe it would be better to decide one an error policy and stick to
> > that. Then you could just simplify away that whole mess, by either
> > ignoring the error and continue or bailing out and die.
>
> "Bailing out and die" would be a revert of commit c1da59dad0eb
> ("cdc-wdm: Clear read pipeline in case of error")?
> And ignoring the error would be "not updating rerr" in
> wdm_in_callback().
> I don't care either way. I can do whatever works for you/users best.
It seems to me that the core of the problem is handling an error
in irq context potentially. How about shifting it to a work queue?
Regards
Oliver
--
To unsubscribe from this list: send the line "unsubscribe linux-usb" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
