mm, page_alloc: Fix a division by zero error when boosting watermarks
diff mbox series

Message ID 20190213131923.GQ9565@techsingularity.net
State New
Headers show
Series
  • mm, page_alloc: Fix a division by zero error when boosting watermarks
Related show

Commit Message

Mel Gorman Feb. 13, 2019, 1:19 p.m. UTC
Yury Norov reported that an arm64 KVM instance could not boot since after
v5.0-rc1 and could addressed by reverting the patches

1c30844d2dfe272d58c ("mm: reclaim small amounts of memory when an external
73444bc4d8f92e46a20 ("mm, page_alloc: do not wake kswapd with zone lock held")

The problem is that a division by zero error is possible if boosting occurs
either very early in boot or if the high watermark is very small. This
patch checks for the conditions and avoids boosting in those cases.

Fixes: 1c30844d2dfe ("mm: reclaim small amounts of memory when an external fragmentation event occurs")
Reported-and-tested-by: Yury Norov <yury.norov@gmail.com>
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
---
 mm/page_alloc.c | 5 +++++
 1 file changed, 5 insertions(+)

Comments

Will Deacon Feb. 13, 2019, 1:20 p.m. UTC | #1
On Wed, Feb 13, 2019 at 01:19:23PM +0000, Mel Gorman wrote:
> Yury Norov reported that an arm64 KVM instance could not boot since after
> v5.0-rc1 and could addressed by reverting the patches
> 
> 1c30844d2dfe272d58c ("mm: reclaim small amounts of memory when an external
> 73444bc4d8f92e46a20 ("mm, page_alloc: do not wake kswapd with zone lock held")
> 
> The problem is that a division by zero error is possible if boosting occurs
> either very early in boot or if the high watermark is very small. This
> patch checks for the conditions and avoids boosting in those cases.
> 
> Fixes: 1c30844d2dfe ("mm: reclaim small amounts of memory when an external fragmentation event occurs")
> Reported-and-tested-by: Yury Norov <yury.norov@gmail.com>
> Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
> ---
>  mm/page_alloc.c | 5 +++++
>  1 file changed, 5 insertions(+)
> 
> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> index d295c9bc01a8..ae7e4ba5b9f5 100644
> --- a/mm/page_alloc.c
> +++ b/mm/page_alloc.c
> @@ -2170,6 +2170,11 @@ static inline void boost_watermark(struct zone *zone)
>  
>  	max_boost = mult_frac(zone->_watermark[WMARK_HIGH],
>  			watermark_boost_factor, 10000);
> +
> +	/* high watermark be be uninitialised or very small */
> +	if (!max_boost)
> +		return;
> +
>  	max_boost = max(pageblock_nr_pages, max_boost);
>  
>  	zone->watermark_boost = min(zone->watermark_boost + pageblock_nr_pages,

I can confirm that this also allows my KVM guest to boot:

Tested-by: Will Deacon <will.deacon@arm.com>

Will
Vlastimil Babka Feb. 13, 2019, 1:42 p.m. UTC | #2
On 2/13/19 2:19 PM, Mel Gorman wrote:
> Yury Norov reported that an arm64 KVM instance could not boot since after
> v5.0-rc1 and could addressed by reverting the patches
> 
> 1c30844d2dfe272d58c ("mm: reclaim small amounts of memory when an external
> 73444bc4d8f92e46a20 ("mm, page_alloc: do not wake kswapd with zone lock held")
> 
> The problem is that a division by zero error is possible if boosting occurs
> either very early in boot or if the high watermark is very small. This
> patch checks for the conditions and avoids boosting in those cases.

Hmm is it really a division by zero? The following line sets max_boost to
pageblock_nr_pages if it's zero. And where would the division happen anyway?

So I wonder what's going on, your patch should AFAICS only take effect when
zone->_watermark[WMARK_HIGH] is 0 or 1 to begin with, otherwise max_boost is at
least 2?

Also upon closer look, I think that (prior to the patch), boost_watermark()
could be reduced (thanks to the max+min capping) to

zone->watermark_boost = pageblock_nr_pages

?

> 
> Fixes: 1c30844d2dfe ("mm: reclaim small amounts of memory when an external fragmentation event occurs")
> Reported-and-tested-by: Yury Norov <yury.norov@gmail.com>
> Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
> ---
>  mm/page_alloc.c | 5 +++++
>  1 file changed, 5 insertions(+)
> 
> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> index d295c9bc01a8..ae7e4ba5b9f5 100644
> --- a/mm/page_alloc.c
> +++ b/mm/page_alloc.c
> @@ -2170,6 +2170,11 @@ static inline void boost_watermark(struct zone *zone)
>  
>  	max_boost = mult_frac(zone->_watermark[WMARK_HIGH],
>  			watermark_boost_factor, 10000);
> +
> +	/* high watermark be be uninitialised or very small */
> +	if (!max_boost)
> +		return;
> +
>  	max_boost = max(pageblock_nr_pages, max_boost);
>  
>  	zone->watermark_boost = min(zone->watermark_boost + pageblock_nr_pages,
>
Mel Gorman Feb. 13, 2019, 2:15 p.m. UTC | #3
On Wed, Feb 13, 2019 at 02:42:36PM +0100, Vlastimil Babka wrote:
> On 2/13/19 2:19 PM, Mel Gorman wrote:
> > Yury Norov reported that an arm64 KVM instance could not boot since after
> > v5.0-rc1 and could addressed by reverting the patches
> > 
> > 1c30844d2dfe272d58c ("mm: reclaim small amounts of memory when an external
> > 73444bc4d8f92e46a20 ("mm, page_alloc: do not wake kswapd with zone lock held")
> > 
> > The problem is that a division by zero error is possible if boosting occurs
> > either very early in boot or if the high watermark is very small. This
> > patch checks for the conditions and avoids boosting in those cases.
> 
> Hmm is it really a division by zero? The following line sets max_boost to
> pageblock_nr_pages if it's zero. And where would the division happen anyway?
> 
> So I wonder what's going on, your patch should AFAICS only take effect when
> zone->_watermark[WMARK_HIGH] is 0 or 1 to begin with, otherwise max_boost is at
> least 2?
> 

The issue can occur if pageblock_nr_pages is also zero or not yet
initialised. It means the changelog is misleading because it  has to
trigger very early in boot as happened with Yury.

> Also upon closer look, I think that (prior to the patch), boost_watermark()
> could be reduced (thanks to the max+min capping) to
> 
> zone->watermark_boost = pageblock_nr_pages
> 

I don't think it's worth being fancy about it if we're hitting
fragmentation issues that early in boot.

Patch
diff mbox series

diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index d295c9bc01a8..ae7e4ba5b9f5 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -2170,6 +2170,11 @@  static inline void boost_watermark(struct zone *zone)
 
 	max_boost = mult_frac(zone->_watermark[WMARK_HIGH],
 			watermark_boost_factor, 10000);
+
+	/* high watermark be be uninitialised or very small */
+	if (!max_boost)
+		return;
+
 	max_boost = max(pageblock_nr_pages, max_boost);
 
 	zone->watermark_boost = min(zone->watermark_boost + pageblock_nr_pages,