Re: [PATCH 1/2] mm: Move definition of 'zone_names' array into mmzone.h

From: Anshuman Khandual
Date: Fri Sep 02 2016 - 00:29:29 EST


On 09/01/2016 02:40 AM, Andrew Morton wrote:
> On Wed, 31 Aug 2016 08:55:49 +0530 Anshuman Khandual <khandual@xxxxxxxxxxxxxxxxxx> wrote:
>
>> zone_names[] is used to identify any zone given it's index which
>> can be used in many other places. So moving the definition into
>> include/linux/mmzone.h for broader access.
>>
>> ...
>>
>> --- a/include/linux/mmzone.h
>> +++ b/include/linux/mmzone.h
>> @@ -341,6 +341,23 @@ enum zone_type {
>>
>> };
>>
>> +static char * const zone_names[__MAX_NR_ZONES] = {
>> +#ifdef CONFIG_ZONE_DMA
>> + "DMA",
>> +#endif
>> +#ifdef CONFIG_ZONE_DMA32
>> + "DMA32",
>> +#endif
>> + "Normal",
>> +#ifdef CONFIG_HIGHMEM
>> + "HighMem",
>> +#endif
>> + "Movable",
>> +#ifdef CONFIG_ZONE_DEVICE
>> + "Device",
>> +#endif
>> +};
>> +
>> #ifndef __GENERATING_BOUNDS_H
>>
>> struct zone {
>> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
>> index 3fbe73a..8e2261c 100644
>> --- a/mm/page_alloc.c
>> +++ b/mm/page_alloc.c
>> @@ -207,23 +207,6 @@ int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
>>
>> EXPORT_SYMBOL(totalram_pages);
>>
>> -static char * const zone_names[MAX_NR_ZONES] = {
>> -#ifdef CONFIG_ZONE_DMA
>> - "DMA",
>> -#endif
>> -#ifdef CONFIG_ZONE_DMA32
>> - "DMA32",
>> -#endif
>> - "Normal",
>> -#ifdef CONFIG_HIGHMEM
>> - "HighMem",
>> -#endif
>> - "Movable",
>> -#ifdef CONFIG_ZONE_DEVICE
>> - "Device",
>> -#endif
>> -};
>> -
>> char * const migratetype_names[MIGRATE_TYPES] = {
>> "Unmovable",
>> "Movable",
>
> This is worrisome. On some (ancient) compilers, this will produce a
> copy of that array into each compilation unit which includes mmzone.h.
>
> On smarter compilers, it will produce a copy of the array in each
> compilation unit which *uses* zone_names[].
>
> On even smarter compilers (and linkers!), only one copy of zone_names[]
> will exist in vmlinux.
>
> I don't know if gcc is an "even smarter compiler" and I didn't check,
> and I didn't check which gcc versions are even smarter. I'd rather not
> have to ;) It is risky.
>
> So, let's just make it non-static and add a declaration into mmzone.h,
> please.
>

I understand your concern, will change it.