2017-06-08 04:03:41

by David Rientjes

Subject: [patch -mm] mm, hugetlb: schedule when potentially allocating many hugepages

A few hugetlb allocators loop while calling the page allocator and can
potentially prevent rescheduling if the page allocator slowpath is not
utilized.

Conditionally schedule when large numbers of hugepages can be allocated.

Signed-off-by: David Rientjes <[email protected]>
---
Based on -mm only to prevent merge conflicts with
"mm/hugetlb.c: warn the user when issues arise on boot due to hugepages"

mm/hugetlb.c | 3 +++
1 file changed, 3 insertions(+)

diff --git a/mm/hugetlb.c b/mm/hugetlb.c
--- a/mm/hugetlb.c
+++ b/mm/hugetlb.c
@@ -1754,6 +1754,7 @@ static int gather_surplus_pages(struct hstate *h, int delta)
break;
}
list_add(&page->lru, &surplus_list);
+ cond_resched();
}
allocated += i;

@@ -2222,6 +2223,7 @@ static void __init hugetlb_hstate_alloc_pages(struct hstate *h)
} else if (!alloc_fresh_huge_page(h,
&node_states[N_MEMORY]))
break;
+ cond_resched();
}
if (i < h->max_huge_pages) {
char buf[32];
@@ -2364,6 +2366,7 @@ static unsigned long set_max_huge_pages(struct hstate *h, unsigned long count,
ret = alloc_fresh_gigantic_page(h, nodes_allowed);
else
ret = alloc_fresh_huge_page(h, nodes_allowed);
+ cond_resched();
spin_lock(&hugetlb_lock);
if (!ret)
goto out;

2017-06-08 04:32:07

by Mike Kravetz

Subject: Re: [patch -mm] mm, hugetlb: schedule when potentially allocating many hugepages

On 06/07/2017 09:03 PM, David Rientjes wrote:
> A few hugetlb allocators loop while calling the page allocator and can
> potentially prevent rescheduling if the page allocator slowpath is not
> utilized.
>
> Conditionally schedule when large numbers of hugepages can be allocated.
>
> Signed-off-by: David Rientjes <[email protected]>
> ---
> Based on -mm only to prevent merge conflicts with
> "mm/hugetlb.c: warn the user when issues arise on boot due to hugepages"
>
> mm/hugetlb.c | 3 +++
> 1 file changed, 3 insertions(+)
>
> diff --git a/mm/hugetlb.c b/mm/hugetlb.c
> --- a/mm/hugetlb.c
> +++ b/mm/hugetlb.c
> @@ -1754,6 +1754,7 @@ static int gather_surplus_pages(struct hstate *h, int delta)
> break;
> }
> list_add(&page->lru, &surplus_list);
> + cond_resched();
> }
> allocated += i;
>
> @@ -2222,6 +2223,7 @@ static void __init hugetlb_hstate_alloc_pages(struct hstate *h)
> } else if (!alloc_fresh_huge_page(h,
> &node_states[N_MEMORY]))
> break;
> + cond_resched();
> }
> if (i < h->max_huge_pages) {
> char buf[32];
> @@ -2364,6 +2366,7 @@ static unsigned long set_max_huge_pages(struct hstate *h, unsigned long count,
> ret = alloc_fresh_gigantic_page(h, nodes_allowed);
> else
> ret = alloc_fresh_huge_page(h, nodes_allowed);
> + cond_resched();

Are not the following lines immediately before the above huge page allocation
in set_max_huge_pages, or am I looking at an incorrect version of the file?

/* yield cpu to avoid soft lockup */
cond_resched();

--
Mike Kravetz

> spin_lock(&hugetlb_lock);
> if (!ret)
> goto out;
>
> --
> To unsubscribe, send a message with 'unsubscribe linux-mm' in
> the body to [email protected]. For more info on Linux MM,
> see: http://www.linux-mm.org/ .
> Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
>

2017-06-09 22:35:30

by David Rientjes

Subject: Re: [patch -mm] mm, hugetlb: schedule when potentially allocating many hugepages

On Wed, 7 Jun 2017, Mike Kravetz wrote:

> > @@ -2364,6 +2366,7 @@ static unsigned long set_max_huge_pages(struct hstate *h, unsigned long count,
> > ret = alloc_fresh_gigantic_page(h, nodes_allowed);
> > else
> > ret = alloc_fresh_huge_page(h, nodes_allowed);
> > + cond_resched();
>
> Are not the following lines immediately before the above huge page allocation
> in set_max_huge_pages, or am I looking at an incorrect version of the file?
>
> /* yield cpu to avoid soft lockup */
> cond_resched();

Ahh, we don't have this in our tree, thanks for catching it. The other
two cond_resched()'s are needed because we have reproduced them, so I'll
send a v2.

2017-06-09 22:36:30

by David Rientjes

Subject: [patch v2 -mm] mm, hugetlb: schedule when potentially allocating many hugepages

A few hugetlb allocators loop while calling the page allocator and can
potentially prevent rescheduling if the page allocator slowpath is not
utilized.

Conditionally schedule when large numbers of hugepages can be allocated.

Signed-off-by: David Rientjes <[email protected]>
---
Based on -mm only to prevent merge conflicts with
"mm/hugetlb.c: warn the user when issues arise on boot due to hugepages"

v2: removed redundant cond_resched() per Mike

mm/hugetlb.c | 2 ++
1 file changed, 2 insertions(+)

diff --git a/mm/hugetlb.c b/mm/hugetlb.c
--- a/mm/hugetlb.c
+++ b/mm/hugetlb.c
@@ -1754,6 +1754,7 @@ static int gather_surplus_pages(struct hstate *h, int delta)
break;
}
list_add(&page->lru, &surplus_list);
+ cond_resched();
}
allocated += i;

@@ -2222,6 +2223,7 @@ static void __init hugetlb_hstate_alloc_pages(struct hstate *h)
} else if (!alloc_fresh_huge_page(h,
&node_states[N_MEMORY]))
break;
+ cond_resched();
}
if (i < h->max_huge_pages) {
char buf[32];

2017-06-09 22:43:51

by Mike Kravetz

Subject: Re: [patch v2 -mm] mm, hugetlb: schedule when potentially allocating many hugepages

On 06/09/2017 03:36 PM, David Rientjes wrote:
> A few hugetlb allocators loop while calling the page allocator and can
> potentially prevent rescheduling if the page allocator slowpath is not
> utilized.
>
> Conditionally schedule when large numbers of hugepages can be allocated.
>
> Signed-off-by: David Rientjes <[email protected]>

Thanks for doing this.

Reviewed-by: Mike Kravetz <[email protected]>

--
Mike Kravetz

> ---
> Based on -mm only to prevent merge conflicts with
> "mm/hugetlb.c: warn the user when issues arise on boot due to hugepages"
>
> v2: removed redundant cond_resched() per Mike
>
> mm/hugetlb.c | 2 ++
> 1 file changed, 2 insertions(+)
>
> diff --git a/mm/hugetlb.c b/mm/hugetlb.c
> --- a/mm/hugetlb.c
> +++ b/mm/hugetlb.c
> @@ -1754,6 +1754,7 @@ static int gather_surplus_pages(struct hstate *h, int delta)
> break;
> }
> list_add(&page->lru, &surplus_list);
> + cond_resched();
> }
> allocated += i;
>
> @@ -2222,6 +2223,7 @@ static void __init hugetlb_hstate_alloc_pages(struct hstate *h)
> } else if (!alloc_fresh_huge_page(h,
> &node_states[N_MEMORY]))
> break;
> + cond_resched();
> }
> if (i < h->max_huge_pages) {
> char buf[32];
>
> --
> To unsubscribe, send a message with 'unsubscribe linux-mm' in
> the body to [email protected]. For more info on Linux MM,
> see: http://www.linux-mm.org/ .
> Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
>

2017-07-06 06:08:18

by Anshuman Khandual

Subject: Re: [patch v2 -mm] mm, hugetlb: schedule when potentially allocating many hugepages

On 06/10/2017 04:06 AM, David Rientjes wrote:
> A few hugetlb allocators loop while calling the page allocator and can
> potentially prevent rescheduling if the page allocator slowpath is not
> utilized.
>
> Conditionally schedule when large numbers of hugepages can be allocated.
>
> Signed-off-by: David Rientjes <[email protected]>

Fixes a task which was getting hung while writing like 10000
hugepages (16MB on POWER8) into /proc/sys/vm/nr_hugepages.

Tested-by: Anshuman Khandual <[email protected]>