On Thu, 9 Apr 2015 10:19:10 +0800 Neil Zhang <neilzhang1123@xxxxxxxxxxx> wrote:
> show detailed free pages per each migrate type in show_free_areas.
>
> After apply this patch, the log printed out will be changed from
>
> [ 558.212844@0] Normal: 218*4kB (UEMC) 207*8kB (UEMC) 126*16kB (UEMC) 21*32kB (UC) 5*64kB (C) 3*128kB (C) 1*256kB (C) 1*512kB (C) 0*1024kB 0*2048kB 1*4096kB (R) = 10784kB
> [ 558.227840@0] HighMem: 3*4kB (UMR) 3*8kB (UMR) 2*16kB (UM) 3*32kB (UMR) 0*64kB 1*128kB (M) 1*256kB (R) 0*512kB 0*1024kB 0*2048kB 0*4096kB = 548kB
>
> to
>
> [ 806.506450@1] Normal: 8969*4kB 4370*8kB 2*16kB 3*32kB 2*64kB 3*128kB 3*256kB 1*512kB 0*1024kB 1*2048kB 0*4096kB = 74804kB
> [ 806.517456@1] orders: 0 1 2 3 4 5 6 7 8 9 10
> [ 806.527077@1] Unmovable: 8287 4370 0 0 0 0 0 0 0 0 0
> [ 806.536699@1] Reclaimable: 681 0 0 0 0 0 0 0 0 0 0
> [ 806.546321@1] Movable: 1 0 0 0 0 0 0 0 0 0 0
> [ 806.555942@1] Reserve: 0 0 2 3 2 3 3 1 0 1 0
> [ 806.565564@1] CMA: 0 0 0 0 0 0 0 0 0 0 0
> [ 806.575187@1] Isolate: 0 0 0 0 0 0 0 0 0 0 0
> [ 806.584810@1] HighMem: 80*4kB 15*8kB 0*16kB 0*32kB 0*64kB 0*128kB 0*256kB 0*512kB 0*1024kB 0*2048kB 0*4096kB = 440kB
> [ 806.595383@1] orders: 0 1 2 3 4 5 6 7 8 9 10
> [ 806.605004@1] Unmovable: 12 0 0 0 0 0 0 0 0 0 0
> [ 806.614626@1] Reclaimable: 0 0 0 0 0 0 0 0 0 0 0
> [ 806.624248@1] Movable: 11 15 0 0 0 0 0 0 0 0 0
> [ 806.633869@1] Reserve: 57 0 0 0 0 0 0 0 0 0 0
> [ 806.643491@1] CMA: 0 0 0 0 0 0 0 0 0 0 0
> [ 806.653113@1] Isolate: 0 0 0 0 0 0 0 0 0 0 0
Thanks. The proposed output does indeed look a lot better.
The columns don't line up, but I guess we can live with that ;)
> --- a/mm/page_alloc.c
> +++ b/mm/page_alloc.c
> @@ -3327,7 +3313,7 @@ void show_free_areas(unsigned int filter)
>
> for_each_populated_zone(zone) {
> unsigned long nr[MAX_ORDER], flags, order, total = 0;
> - unsigned char types[MAX_ORDER];
> + unsigned long nr_free[MAX_ORDER][MIGRATE_TYPES], mtype;
>
> if (skip_free_areas_node(filter, zone_to_nid(zone)))
> continue;
nr_free[][] is an 8x11 array of 8, I think? That's 704 bytes of stack,
and show_free_areas() is called from very deep call stacks - from the
oom-killer, for example. We shouldn't do this.
I think we can eliminate nr_free[][]:
> + for (mtype = 0; mtype < MIGRATE_TYPES; mtype++) {
> + printk("%12s: ", migratetype_names[mtype]);
> + for (order = 0; order < MAX_ORDER; order++)
> + printk("%6lu ", nr_free[order][mtype]);
> + printk("\n");
> + }
In the above loop, take zone->lock and calculate the nr_free for this
particular order/mtype, then release zone->lock.
That will be slower, but show_free_areas() doesn't need to be fast.
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxx. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>