On 09/19/2014 06:58 AM, Sebastian Andrzej Siewior wrote:

On 09/19/2014 12:22 PM, Heikki Krogerus wrote:

quoted

Couldn't you just replace the handle_irq with a custom irq routine in
the omap driver like you did with set_termios? Where you would do
those tricks and/or call serial8250_handle_irq()?

Tricks within serial8250_handle_irq(), see [0]. It is not a lot but
still. I could provide my own handle irq, just asking what you would
prefer.

[0]
http://git.breakpoint.cc/cgit/bigeasy/linux.git/commit/?h=uart_v10_pre2&id=f26f161d998ee4a84a0aa6ddff69a435c25f204d

quoted

The 8250_core changes in that patch #10 only modify
serial8250_handle_irg right?

Correct. However there is another change due to the RX_DMA_BUG. A while
ago you said that this RX_DMA_BUG thing might be something that other
SoC could use, too.
If you ask me now for my own irq routine it would make sense to move RX
bug handling into omap specific code as well.

I'm not excited at the prospect of an omap-specific irq handler, especially
if this is just a silicon bug. I think it will create and encourage
unnecessary code variation in the 8250 driver. The inertia of an
omap-specific irq handler will mean it will probable stay forever. Suppose
this tx dma bug is later discovered to be fixable inline (rather than by
state), then we'll be stuck with an irq handler that no one will want to
integrate.

Regards,
Peter Hurley