On Thu, Oct 01, 2009 at 07:52:21PM -0700, Eric W. Biederman wrote:
> Alan Cox <alan@lxorguk.ukuu.org.uk> writes:
> 
> >> As it stands today ftdi_sio does indeed call tty_flip_buffer_push from
> >> interrupt context with low_latency set and that is obviously incorrect,
> >> right?
> >
> > It seems to do it from a work queue - or did I miss a case ?
> 
> ftdi_sio crash quite regularly for me with 2.6.31.
> 
> With a bunch of nasties like:
> BUG: scheduling while atomic: swapper/0/0x00010000
> bad: scheduling from the idle thread!

It's the same problem.

Greg, can't we apply the patch for stable at least? Then we can massage
ftdi_sio into actually using the work queue for doing _all_ processing
in the meantime if deemed necessary.

Alan, did you have time to look at it? Are there any reasons for wanting
to keep low_latency in ftdi_sio when it was removed from all other
drivers processing in interrupt context (without doing work queue
re-implementations)?

Thanks,
Johan