Problem of concurrency in arch/ppc/8260_io/uart.c

[Joakim Tjernlund]

> Jean-Denis Boyer wrote:
>
> > I have just tested your patch on 8260 and it fixes the problem when printk is called ...
>
> The use of printk is strictly a debug function.  If it is necessary to see
> it print out pretty, then use a different I/O path for the application or
> dump out the log buffer separately.  Serial ports don't guarantee any
> transactional operation with multiple users, printk isn't any different
> although it does the bad thing of blocking interrupts while it processes
> the message.
hmm, it does not block interrupts as is, does it? Only if you apply the patch?

>
> Adding this patch adds lots of interrupt latency and I would recommend
> against it.  It appears there is a bug that should be fixed to ensure
> the proper synchronization of the BDs, but not with such large scale
> blocking of interrupts.
Agreed, but is there another mechanism but to disable interrupts? I can't think of any.
It is possible to shrink the window with interrupts off dramatically.

   Jocke
>
> Thanks.
>
>
> -- Dan


** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/