Re: [PATCH] mm: compaction: early termination in compact_nodes()

[Andrew Morton <akpm@linux-foundation.org>]

On Sat, 17 Feb 2024 10:55:10 +0800 Kefeng Wang <wangkefeng.wang@huawei.com> wrote:

> On 2024/2/12 22:22, David Hildenbrand wrote:
> > On 08.02.24 22:14, Andrew Morton wrote:
> >> On Thu, 8 Feb 2024 10:25:08 +0800 Kefeng Wang 
> >> <wangkefeng.wang@huawei.com> wrote:
> >>
> >>> No need to continue try compact memory if pending fatal signal,
> >>> allow loop termination earlier in compact_nodes().
> >>
> >> Seems sensible, but...  why?  Is there some problem which we can
> >> demonstrate with the existing code?  In other words, does this change
> >> provide any observable benefit under any circumstances?
> > 
> > I'd also be curious why the existing fatal_signal_pending() calls are 
> > insufficient.
> 
> The existing fatal_signal_pending() does make compact_zone() breakout of
> the while loop, but it still enter the next zone/next nid, and some 
> unnecessary functions(eg, lru_add_drain) called, no observable benefit
> from test, it is just found from code inspection when refactor 

Fair enough.  I added the above words to the changelog (this material
should have been communicated in the original!) and I'll plan to move
this change into mm-stable next week unless someone stops me.