On 9/8/20 8:47 AM, Valentin Schneider wrote:
> 
> On 08/09/20 16:03, Vincent Guittot wrote:
>> Adding Valentin as this seems related to the patch "sched/topology:
>> Move sd_flag_debug out of linux/sched/topology.h"
>>
> 
> Thanks!
> 
>>
>> On Tue, 8 Sep 2020 at 16:48, Randy Dunlap <rdunlap@infradead.org> wrote:
>>>
>>> On 9/8/20 3:56 AM, Stephen Rothwell wrote:
>>>> Hi all,
>>>>
>>>> Changes since 20200903:
>>>>
>>>
>>> on i386:
>>>
>>> ld: kernel/sched/topology.o: in function `cpu_attach_domain':
>>> topology.c:(.text+0xf03): undefined reference to `sd_flag_debug'
>>> ld: topology.c:(.text+0xf4f): undefined reference to `sd_flag_debug'
>>> ld: topology.c:(.text+0xfc2): undefined reference to `sd_flag_debug'
>>>
> 
> So that should be in sched_domain_debug_one() which only gets defined for
> CONFIG_SCHED_DEBUG. Now, sd_flag_debug is:
> - declared in include/linux/sched/topology.h if CONFIG_SCHED_DEBUG
> - defined in kernel/sched/debug.c (only built if CONFIG_SCHED_DEBUG)
> 
> I've compile-tested this for arm64 & x86 (default configs + toggling
> CONFIG_SCHED_DEBUG). UP can't be the cause because topology.c wouldn't even
> be compiled then.
> 
> IIUC this is a build with CONFIG_SCHED_DEBUG but for some reason it doesn't
> link with kernel/sched/debug.o?
> 
>>>
>>> Full randconfig file is attached.
>>>
> 
> I wanted to peek at that config, but can't find that email. The thread view
> on lore [1] also shows it as AWOL; any chance you (or someone else) could
> re-send it?
> 
> [1]: https://lore.kernel.org/lkml/20200908205659.361b0a1b@canb.auug.org.au/
> 
>>>
>>> --
>>> ~Randy
>>> Reported-by: Randy Dunlap <rdunlap@infradead.org>

Sure, here it is again.  And
CONFIG_SMP=y
CONFIG_SCHED_DEBUG=y

thanks.

-- 
~Randy