cc'ng Russell/LAKML/Fengguang..
On 9 January 2014 14:08, Olof Johansson olof@lixom.net wrote:
This patch breaks a bunch of ARM boards. In particular, the following defconfigs no longer build:
That's really bad, Rafael will scold me again :)
assabet_defconfig badge4_defconfig cerfcube_defconfig collie_defconfig h3600_defconfig hackkit_defconfig jornada720_defconfig lart_defconfig neponset_defconfig pleb_defconfig shannon_defconfig simpad_defconfigError is:
drivers/built-in.o: In function `cpufreq_generic_get': drivers/cpufreq/cpufreq.c:189: undefined reference to `clk_get_rate'
Seems like this needs to be guarded by HAVE_CLK?
Naah.. After some investigation I found this:
- We already have dummy implementations of clk routines in case CONFIG_HAVE_CLK is not defined (I added them long back).
- There is one thing common among all above defconfigs, all belong to SA1100 family :)
- And the problem is: SA1100 wanted to define its own clk routines and selects CLKDEV_LOOKUP (which enables HAVE_CLK), but it doesn't implement all clk routines. Which is *wrong*.
So, actually this patch brought an _existing_ bug in limelight. And this should be fixed by adding dummy or meaningful implementation of missing clk routines.
@Russell: If above looks correct then can you please communicate what should we do here? I don't really know what exactly these routines should have, simply a copy of dummy routines from clk.h or some meaningful stuff. So, maybe you can write a patch, otherwise let me know what to write and I will give it a try.
@Rafael: Please *don't* revert this patch, its not my fault this time :)
@Fengguang: Would it make sense to add build tests for all ARM defconfigs in your build system? I thought its already there :) That way I can always be sure that my stuff (would be helpful for others as well though) isn't breaking build (Atleast) for any platform.
Thanks.
-- viresh