Hey, Peter.

On Thu, Dec 03, 2015 at 09:43:13PM +0100, Peter Zijlstra wrote:
> I'm not sure about using PF_MEMALLOC for detecting reclaim. There appear
> to be more sites setting this than reclaim. See:
>
> drivers/block/nbd.c:	current->flags |= PF_MEMALLOC;
> drivers/mmc/card/queue.c:	current->flags |= PF_MEMALLOC;
> drivers/mtd/nand/nandsim.c:	current->flags |= PF_MEMALLOC;
> drivers/scsi/iscsi_tcp.c:	current->flags |= PF_MEMALLOC;
> drivers/staging/lustre/include/linux/libcfs/linux/linux-mem.h:#define memory_pressure_set() do { current->flags |= PF_MEMALLOC; } while (0)
> fs/cifs/connect.c:	current->flags |= PF_MEMALLOC;
> fs/xfs/libxfs/xfs_btree.c:		new_pflags |= PF_MEMALLOC | PF_SWAPWRITE | PF_KSWAPD;
> fs/xfs/xfs_trans_ail.c:	current->flags |= PF_MEMALLOC;
> include/linux/sched.h:	current->flags |= PF_MEMALLOC_NOIO;
> mm/page_alloc.c:	current->flags |= PF_MEMALLOC;
> mm/page_alloc.c:	current->flags |= PF_MEMALLOC;
> mm/vmscan.c:	tsk->flags |= PF_MEMALLOC | PF_SWAPWRITE | PF_KSWAPD;
> mm/vmscan.c:	p->flags |= PF_MEMALLOC;
> mm/vmscan.c:	p->flags |= PF_MEMALLOC | PF_SWAPWRITE;
> net/core/dev.c:		current->flags |= PF_MEMALLOC;
> net/core/sock.c:	current->flags |= PF_MEMALLOC;
> 
> 
> The actual reclaim sites in page_alloc and vmscan set
> current->reclaim_state. So testing against that might be more accurate.

So, if I'm not mistaken, those are all marking tasks which can be
depended upon during memory reclaim and we do want to catch them all.
PF_MEMALLOC shouldn't depend on something which require memory to be
reclaimed to guarantee forward progress.

Thanks.

-- 
tejun