[PATCH 16/16] ARM: omap: disable cpuidle for OMAP3 platforms

On Wed, Feb 08, 2012 at 10:59:06AM -0800, Tony Lindgren wrote:

* Russell King - ARM Linux [off-list ref] [120208 08:10]:

quoted

Commit 2fd149645eb4 (ARM: OMAP2+: UART: Remove omap_uart_can_sleep and add pm_qos)
has caused a regression on OMAP3 platforms.

When the UART is trying to transmit data, if we enter a low power mode,
transmission stops, which makes serial on OMAP3 unusable - a 'dmesg'
takes five minutes to be output at 115200 baud, at a rate of around a
block of 16 characters every couple of seconds.

Unfortunately, the commit above can't be reverted because of many other
changes in this area, so this implements a dirty fix by disabling
CPU idle in the places the original commit does, irrespective of the
UART state.

...

quoted

--- a/arch/arm/mach-omap2/pm34xx.c
+++ b/arch/arm/mach-omap2/pm34xx.c

@@ -421,7 +421,7 @@ static void omap3_pm_idle(void)
 	local_irq_disable();
 	local_fiq_disable();
 
-	if (omap_irq_pending() || need_resched())
+	if (omap_irq_pending() || need_resched() || 1)
 		goto out;
 
 	trace_power_start(POWER_CSTATE, 1, smp_processor_id());

Argh, this is just too ugly. There has got to be a better fix for the
-rc series.

That I'd agree on - because it is ugly and it's the best I could do
to revert the commit mentioned above short of reverting a whole raft
of other useful looking commits.

There's got to be a better solution.

Looks like the patches to fix omap-serial.c are queued for v3.4,
so that won't help.

Well, the fact of the matter is this is a regression, which means it
needs fixing for v3.3.  A simple patch would be preferred over more
complex patches.

So, if there's no other solution but to put the omap-serial patches
in for v3.3, then that's the best solution for v3.3.  First, someone
who understands this code needs to see whether there is a simpler
fix (eg, putting back what was there in a simpler form.)

However, leaving it in its current state for v3.3 is not acceptable.