Re: [PATCH] mm, page_alloc: only use per-cpu allocator for irq-safe requests -fix v2

[Thomas Gleixner]

On Wed, 8 Feb 2017, Mel Gorman wrote:

> preempt_enable_no_resched() was used based on review feedback that had
> no strong objection at the time. The thinking was that it avoided adding
> a preemption point where one didn't exist before so the feedback was
> applied. This reasoning was wrong.
> 
> There was an indirect preemption point as explained by Thomas Gleixner where
> an interrupt could set_need_resched() followed by preempt_enable being
> a preemption point that matters. This use of preempt_enable_no_resched
> is bad from both a mainline and RT perspective and a violation of the
> preemption mechanism. Peter Zijlstra noted that "the only acceptable use
> of preempt_enable_no_resched() is if the next statement is a schedule()
> variant".
> 
> The usage was outright broken and I should have stuck to preempt_enable()
> as it was originally developed. It's known from previous tests
> that there was no detectable difference to the performance by using
> preempt_enable_no_resched().
> 
> This is a fix to the mmotm patch
> mm-page_alloc-only-use-per-cpu-allocator-for-irq-safe-requests.patch
> 
> Signed-off-by: Mel Gorman <mgorman@xxxxxxxxxxxxxxxxxxx>

Reviewed-by: Thomas Gleixner <tglx@xxxxxxxxxxxxx>

> ---
>  mm/page_alloc.c | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
> 
> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> index eaecb4b145e6..2a36dad03dac 100644
> --- a/mm/page_alloc.c
> +++ b/mm/page_alloc.c
> @@ -2520,7 +2520,7 @@ void free_hot_cold_page(struct page *page, bool cold)
>  	}
>  
>  out:
> -	preempt_enable_no_resched();
> +	preempt_enable();
>  }
>  
>  /*
> @@ -2686,7 +2686,7 @@ static struct page *rmqueue_pcplist(struct zone *preferred_zone,
>  		__count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order);
>  		zone_statistics(preferred_zone, zone);
>  	}
> -	preempt_enable_no_resched();
> +	preempt_enable();
>  	return page;
>  }
>  
> 
> -- 
> Mel Gorman
> SUSE Labs
>