On Tue 03-11-15 14:16:03, Andrew Morton wrote: [...] > > [1188431.177410] apache2: page allocation failure: order:1, mode:0x204020 > > An order-1 page, __GFP_COMP|__GFP_HIGH. ie: GFP_ATOMIC. __GFP_HIGH doesn't really work well on a small zone like DMA I am afraid. [...] > > [1188431.177521] Node 0 DMA free:7968kB min:40kB low:48kB high:60kB [...] > > [1188431.177527] lowmem_reserve[]: 0 1988 1988 1988 [...] > > [1188431.177555] Node 0 DMA: 44*4kB (UE) 94*8kB (UEM) 76*16kB (UE) 42*32kB > > (UEM) 22*64kB (UEM) 6*128kB (UE) 3*256kB (UE) 1*512kB (E) 1*1024kB (U) 0*2048kB > > 0*4096kB = 7968kB > > The DMA zone has lots and lots of higher-order pages available which > could satisfy this allocation. min = 10 - 10/2 = 5 # __GFP_HIGH min = 5 - 5/4 = 4 # ALLOC_HARDER free_pages = 1992 - ((1<<1) - 1) = 1991 free_cma = 0 1991 <= 4 + 1988 So we do not pass lowmem reserves check here... [...] > The kernel could and should have satisfied this order-1 GFP_ATOMIC > IRQ-context allocation from the DMA zone. But it did not do so. Bug. I am not really sure this is a bug to be honest. It seems that we are not coping with the non sleeping allocation pressure. I would suggest increasing min_free_kbytes. -- Michal Hocko SUSE Labs -- 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>