On 11/21/2012 01:28 AM, Linus Walleij wrote:

Oh and there was this comment/TODO:

On Mon, Nov 19, 2012 at 7:31 PM, Stephen Warren [off-list ref] wrote:

quoted

@@ -17,15 +17,6 @@
  *   Initialise the kernels jiffy timer source, claim interrupt
  *   using setup_irq.  This is called early on during initialisation
  *   while interrupts are still disabled on the local CPU.
- * - suspend
- *   Suspend the kernel jiffy timer source, if necessary.  This
- *   is called with interrupts disabled, after all normal devices
- *   have been suspended.  If no action is required, set this to
- *   NULL.
- * - resume
- *   Resume the kernel jiffy timer source, if necessary.  This
- *   is called with interrupts disabled before any normal devices
- *   are resumed.  If no action is required, set this to NULL.
  * - offset
  *   Return the timer offset in microseconds since the last timer
  *   interrupt.  Note: this must take account of any unprocessed

@@ -33,8 +24,6 @@
  */
 struct sys_timer {
        void                    (*init)(void);
-       void                    (*suspend)(void);
-       void                    (*resume)(void);
 };

So from the above it is quite clear that the sys_timer is breaking
the suspend_noirq/resume_noirq naming convention from
runtime PM as IRQs are disabled on these paths.

The same goes for struct clock_event_device ...

So while this looks just as bad after as before the patch,
we should take a mental notice to rename the .suspend
and .resume hooks in the clock_event_device to
.suspend_noirq and .resume_noirq at some point.

I was thinking that if your patch set is introducing a
plethora of new users of these hooks we should maybe
stick a patch at the beginning of the series renaming the
hooks to *_noirq, but if it's a major obstacle it can surely wait.

I think I'd prefer to keep that rename separate and do it later; while
this series does introduce a few new users of the type, there are many
more pre-existing users. Renaming all the users would end up making this
series potentially conflict with addition of new users in tip.git or
elsewhere, whereas any possible current conflicts from this series
should be resolvable in arm-soc pretty easily I hope, so I'd rather
create a separate series that does the rename, and probably apply it to
tip.git, probably for 3.10?