Re: [PATCH] swap 10/13 tmpfs atomics

[Andrew Morton]

Hugh Dickins <[e-mail]> wrote:
>
> move_from_swap_cache and add_to_page_cache_lru are using GFP_ATOMIC,
> which can easily fail in an intermittent way.  Rude if shmem_getpage
> then fails with -ENOMEM: cond_resched to let kswapd in, and repeat.

I think the preferred way of waiting for the IO system and kswapd to catch up
is blk_congestion_wait().  Because it waits for the "right" amount of time.

I'll make that change.

And yes, the name is silly: "what if it's not block-backed"?  It hasn't
caused any problems yet, but maybe one day we'll need to find a way for
network-backed filesystems to deliver a wakeup to sleepers there.