Re: linux-next: build failure after merge of the akpm-current tree

[Waiman Long]

On 5/12/21 1:28 AM, Stephen Rothwell wrote:
> Hi all,
>
> As Randy also reported ...
>
> After merging the akpm-current tree, today's linux-next build (powerpc
> ppc64_defconfig) failed like this:
>
> mm/slab_common.c:764:8: error: array index in initializer exceeds array bounds
>    764 |  .name[KMALLOC_RECLAIM] = "kmalloc-rcl-" #__short_size, \
>        |        ^~~~~~~~~~~~~~~
> mm/slab_common.c:776:2: note: in expansion of macro 'INIT_KMALLOC_INFO'
>    776 |  INIT_KMALLOC_INFO(0, 0),
>        |  ^~~~~~~~~~~~~~~~~
> mm/slab_common.c:756:39: error: array index in initializer exceeds array bounds
>    756 | #define KMALLOC_CGROUP_NAME(sz) .name[KMALLOC_CGROUP] = "kmalloc-cg-" #sz,
>        |                                       ^~~~~~~~~~~~~~
> mm/slab_common.c:765:2: note: in expansion of macro 'KMALLOC_CGROUP_NAME'
>    765 |  KMALLOC_CGROUP_NAME(__short_size)   \
>        |  ^~~~~~~~~~~~~~~~~~~
>
> and more
>
> Caused by commits
>
>    4d57437a3d3c ("mm: memcg/slab: create a new set of kmalloc-cg-<n> caches")
>    0727bf117622 ("mm: memcg/slab: don't create kmalloc-cg caches with cgroup.memory=nokmem")
>    d795c307bd3a ("mm: memcg/slab: disable cache merging for KMALLOC_NORMAL caches")
>
> At least they are the commits I have reverted to fix this.
Sorry about that. It turns out that construct like that below doesn't work.

enum kmalloc_cache_type {
        KMALLOC_NORMAL = 0,
#ifdef CONFIG_MEMCG_KMEM
        KMALLOC_CGROUP,
#else
        KMALLOC_CGROUP = KMALLOC_NORMAL,
#endif
        KMALLOC_RECLAIM,
#ifdef CONFIG_ZONE_DMA
        KMALLOC_DMA,
#else
        KMALLOC_DMA = KMALLOC_NORMAL,
#endif
        NR_KMALLOC_TYPES
};

The first ifdef is fine, but the second one will reset the enumeration 
count back to 0 if CONFIG_ZONE_DMA is not defined causing 
NR_KMALLOC_TYPES set to 1. Will send a new version to fix that.

Cheers,
Longman