On Wed, Jun 23, 2010 at 12:10:08PM +0200, John Ogness wrote:

On 2010-06-23, Ivo Clarysse [off-list ref] wrote:

quoted

But is it OK to use a regular (non-volatile) variable to communicate
between interrupt context and the non-interrupt context ?

In this case, yes.

quoted

My original patch for i.MX21 used completions instead:

http://lists.infradead.org/pipermail/linux-arm-kernel/2010-April/012694.html

Ah. It seems you've been through all this before. I wish I had noticed
that thread before. I will need to check more carefully in the future.

Yes, your original patch achieves the exact same thing. Whether we use
wait_event() with a flag or wait_completion() really is the same
thing. So I guess Sascha can decide what we should do there.

What I like about your original patch is that only the i.MX21 has the
cost of constantly enabling/disabling the irq line. It adds 5
cpu_is_mx21() blocks to the code, but will lead to less work for the CPU
on non-i.MX21 boards.

Ok, if it's the only way out to have 5 cpu_is_* blocks, then lets go for
it.

BTW I observed that at least on i.MX27 the latencies introduced by
waiting for an interrupt cause a significant performance drop. The
driver gets much faster when we just poll all the time. I don't know how
this affects system performance otherwise, but it may be a possibility
to drop interrupt support at least for i.MX21. I have no idea how long
the longest possible time we'd have to poll is though.

Sascha

-- 
Pengutronix e.K.                           |                             |
Industrial Linux Solutions                 | http://www.pengutronix.de/  |
Peiner Str. 6-8, 31137 Hildesheim, Germany | Phone: +49-5121-206917-0    |
Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-5555 |