On Friday 26 October 2012 12:45 AM, Tony Lindgren wrote:

* Santosh Shilimkar [off-list ref] [121024 23:34]:

quoted

On Thursday 25 October 2012 06:12 AM, Tony Lindgren wrote:

quoted

* Tony Lindgren [off-list ref] [121024 17:36]:

quoted

* Santosh Shilimkar [off-list ref] [121017 06:35]:

quoted

(Looping Arnd and Olof)

On Wednesday 17 October 2012 06:58 PM, Lokesh Vutla wrote:

quoted

When building omap_l3_noc/smx drivers as modules, the following
warning appears:

CC [M]  drivers/bus/omap_l3_smx.o
drivers/bus/omap_l3_smx.c:291: warning: data definition has no type or storage class
drivers/bus/omap_l3_smx.c:291: warning: type defaults to 'int' in declaration of 'postcore_initcall_sync'
drivers/bus/omap_l3_smx.c:291: warning: parameter names (without types) in function declaration
drivers/bus/omap_l3_smx.c:287: warning: 'omap3_l3_init' defined but not used
CC [M]  drivers/bus/omap_l3_noc.o
drivers/bus/omap_l3_noc.c:260: warning: data definition has no type or storage class
drivers/bus/omap_l3_noc.c:260: warning: type defaults to 'int' in declaration of 'arch_initcall_sync'
drivers/bus/omap_l3_noc.c:260: warning: parameter names (without types) in function declaration
drivers/bus/omap_l3_noc.c:256: warning: 'omap4_l3_init' defined but not used

Adding module_init() and macros in omap_l3_noc/smx drivers when building
as modules to remove the above warning.

Reported-by: Tony Lindgren <tony@atomide.com>
Signed-off-by: Lokesh Vutla <redacted>
---

Thanks for the fix Lokesh. Looks fine to me.

Acked-by: Santosh Shilimkar <redacted>

Looks like nobody else has picked this up so I'll queue this along
with few other omap warnings and regressions.

Hmm actually this might require some more discussion. If we make
it use regular initcalls, then the ugly ifdefs can be left
out. Is there a reason to init this early, can't we just use regular
initcalls?

I thought about it. The whole reason we want interconnect errors
enabled early in the boot to avoid bad accesses issued on
interconnect
in early boot by various init codes. We managed to discovered many
init sequence issues where the a driver is trying to access registers
when clocks are not active, or drivers are using bad mapping. At times
these errors gets un-noticed because of the behavior of interconnect
and later causes serious issues. Leaving the driver init late in the
boot means we can't catch any of the issues happen till the L3 driver
init happens.

OK yeah that makes sense. How about let's just make it
just postcore_initcall instead of postcore_initcall_sync?

_sync was added by purpose since the driver has depedency on
the hwmod initialisation which is postcore_initcall. Without
the sync, we open the race condition and the l3 driver
registration will fail.

In include/linux/module.h we have:

...
#else /* MODULE */

/* Don't use these in loadable modules, but some people do... */
#define early_initcall(fn)              module_init(fn)
#define core_initcall(fn)               module_init(fn)
#define postcore_initcall(fn)           module_init(fn)
...

While the postcore_initcall_sync does not have those.

No idea what the current plan is, but I sort of remember reading
that the _sync versions are going away at some point anyways?

I have no idea either. As mentioned sync was added to avoid the
race. If and when _sync is removed, something should come as an
alternative to avoid initcall completion dependencies for the one
which falls in same group.

Regards,
Santosh