SH: error: implicit declaration of function 'init_cache_node_node'

SH: error: implicit declaration of function 'init_cache_node_node'

[Randy Dunlap]

on arch/sh/,
CONFIG_SMP is not defined,
CONFIG_NUMA=y,
CONFIG_MEMORY_HOTPLUG is not defined:

../mm/slab.c: In function 'slab_memory_callback':
../mm/slab.c:1143:23: error: implicit declaration of function 'init_cache_node_node'; did you mean 'drain_cache_node_node'? [-Werror=implicit-function-declaration]
  1143 |                 ret = init_cache_node_node(nid);


commit 76af6a054da4
Author: Dave Hansen <dave.hansen@linux.intel.com>
Date:   Mon Oct 18 15:15:32 2021 -0700

     mm/migrate: add CPU hotplug to demotion #ifdef


How should we handle this config?

-- 
~Randy

Re: SH: error: implicit declaration of function 'init_cache_node_node'

[Arnd Bergmann]

On Sun, Nov 14, 2021 at 2:17 AM Randy Dunlap <rdunlap@infradead.org> wrote:
>
> on arch/sh/,
> CONFIG_SMP is not defined,
> CONFIG_NUMA=y,
> CONFIG_MEMORY_HOTPLUG is not defined:
>
> ../mm/slab.c: In function 'slab_memory_callback':
> ../mm/slab.c:1143:23: error: implicit declaration of function 'init_cache_node_node'; did you mean 'drain_cache_node_node'? [-Werror=implicit-function-declaration]
>   1143 |                 ret = init_cache_node_node(nid);
>
>
> commit 76af6a054da4
> Author: Dave Hansen <dave.hansen@linux.intel.com>
> Date:   Mon Oct 18 15:15:32 2021 -0700
>
>      mm/migrate: add CPU hotplug to demotion #ifdef
>
>
> How should we handle this config?

I think we can safely assume that NUMA without SMP is not a useful
configuration on SH and add a dependency in Kconfig for it.

I assume this came from some randconfig build rather than a config
that someone was intentionally using? My guess would be that testing
sh randconfig kernels is not overly productive as you likely need
countless patches before there is a chance of it working most of
the time. I haven't tried this myself, but I spend a lot of time fixing
randconfig failures on arm and x86, and I whenever I try any other
architecture, there is simply too much work needed upfront.

        Arnd

Re: SH: error: implicit declaration of function 'init_cache_node_node'

[Randy Dunlap]

On 11/14/21 2:04 PM, Arnd Bergmann wrote:
> On Sun, Nov 14, 2021 at 2:17 AM Randy Dunlap <rdunlap@infradead.org> wrote:
>>
>> on arch/sh/,
>> CONFIG_SMP is not defined,
>> CONFIG_NUMA=y,
>> CONFIG_MEMORY_HOTPLUG is not defined:
>>
>> ../mm/slab.c: In function 'slab_memory_callback':
>> ../mm/slab.c:1143:23: error: implicit declaration of function 'init_cache_node_node'; did you mean 'drain_cache_node_node'? [-Werror=implicit-function-declaration]
>>    1143 |                 ret = init_cache_node_node(nid);
>>
>>
>> commit 76af6a054da4
>> Author: Dave Hansen <dave.hansen@linux.intel.com>
>> Date:   Mon Oct 18 15:15:32 2021 -0700
>>
>>       mm/migrate: add CPU hotplug to demotion #ifdef
>>
>>
>> How should we handle this config?
> 
> I think we can safely assume that NUMA without SMP is not a useful
> configuration on SH and add a dependency in Kconfig for it.
> 
> I assume this came from some randconfig build rather than a config
> that someone was intentionally using? My guess would be that testing
> sh randconfig kernels is not overly productive as you likely need
> countless patches before there is a chance of it working most of
> the time. I haven't tried this myself, but I spend a lot of time fixing
> randconfig failures on arm and x86, and I whenever I try any other
> architecture, there is simply too much work needed upfront.

I understand why you guess that it's from a randconfig, but it's not.
It's from migor_defconfig.

-- 
~Randy