[PATCH] hugetlb: vmstat events for huge page allocations v3

[PATCH] hugetlb: vmstat events for huge page allocations v3

[Eric B Munson]

[-- Attachment #1: Type: text/plain, Size: 2863 bytes --]

From: Adam Litke <agl@us.ibm.com>

Following feedback here is v3

Allocating huge pages directly from the buddy allocator is not guaranteed
to succeed.  Success depends on several factors (such as the amount of
physical memory available and the level of fragmentation).  With the
addition of dynamic hugetlb pool resizing, allocations can occur much more
frequently.  For these reasons it is desirable to keep track of huge page
allocation successes and failures.

Add two new vmstat entries to track huge page allocations that succeed and
fail.  The presence of the two entries is contingent upon
CONFIG_HUGETLB_PAGE being enabled.

This patch was created against linux-2.6.25-rc5

Signed-off-by: Adam Litke <agl@us.ibm.com>
Signed-off-by: Eric Munson <ebmunson@us.ibm.com>

 include/linux/vmstat.h |    8 +++++++-
 mm/hugetlb.c           |    7 +++++++
 mm/vmstat.c            |    4 ++++
 3 files changed, 18 insertions(+), 1 deletions(-)

diff --git a/include/linux/vmstat.h b/include/linux/vmstat.h
index 9f1b4b4..f68f538 100644
--- a/include/linux/vmstat.h
+++ b/include/linux/vmstat.h
@@ -25,6 +25,12 @@
 #define HIGHMEM_ZONE(xx)
 #endif
 
+#ifdef CONFIG_HUGETLB_PAGE
+#define HTLB_STATS	HTLB_BUDDY_PGALLOC, HTLB_BUDDY_PGALLOC_FAIL,
+#else
+#define HTLB_STATS
+#endif
+
 #define FOR_ALL_ZONES(xx) DMA_ZONE(xx) DMA32_ZONE(xx) xx##_NORMAL HIGHMEM_ZONE(xx) , xx##_MOVABLE
 
 enum vm_event_item { PGPGIN, PGPGOUT, PSWPIN, PSWPOUT,
@@ -36,7 +42,7 @@ enum vm_event_item { PGPGIN, PGPGOUT, PSWPIN, PSWPOUT,
 		FOR_ALL_ZONES(PGSCAN_KSWAPD),
 		FOR_ALL_ZONES(PGSCAN_DIRECT),
 		PGINODESTEAL, SLABS_SCANNED, KSWAPD_STEAL, KSWAPD_INODESTEAL,
-		PAGEOUTRUN, ALLOCSTALL, PGROTATED,
+		PAGEOUTRUN, ALLOCSTALL, PGROTATED, HTLB_STATS
 		NR_VM_EVENT_ITEMS
 };
 
diff --git a/mm/hugetlb.c b/mm/hugetlb.c
index 74c1b6b..dd20cb0 100644
--- a/mm/hugetlb.c
+++ b/mm/hugetlb.c
@@ -239,6 +239,11 @@ static int alloc_fresh_huge_page(void)
 		hugetlb_next_nid = next_nid;
 	} while (!page && hugetlb_next_nid != start_nid);
 
+	if (ret)
+		count_vm_event(HTLB_BUDDY_PGALLOC);
+	else
+		count_vm_event(HTLB_BUDDY_PGALLOC_FAIL);
+
 	return ret;
 }
 
@@ -299,9 +304,11 @@ static struct page *alloc_buddy_huge_page(struct vm_area_struct *vma,
 		 */
 		nr_huge_pages_node[nid]++;
 		surplus_huge_pages_node[nid]++;
+		__count_vm_event(HTLB_BUDDY_PGALLOC);
 	} else {
 		nr_huge_pages--;
 		surplus_huge_pages--;
+		__count_vm_event(HTLB_BUDDY_PGALLOC_FAIL);
 	}
 	spin_unlock(&hugetlb_lock);
 
diff --git a/mm/vmstat.c b/mm/vmstat.c
index 422d960..bbe728d 100644
--- a/mm/vmstat.c
+++ b/mm/vmstat.c
@@ -644,6 +644,10 @@ static const char * const vmstat_text[] = {
 	"allocstall",
 
 	"pgrotated",
+#ifdef CONFIG_HUGETLB_PAGE
+	"htlb_alloc_success",
+	"htlb_alloc_fail",
+#endif
 #endif
 };
 

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

Re: [PATCH] hugetlb: vmstat events for huge page allocations v3

[Mel Gorman]

On (17/03/08 13:02), Eric B Munson didst pronounce:
> From: Adam Litke <agl@us.ibm.com>
> 
> Following feedback here is v3
> 
> Allocating huge pages directly from the buddy allocator is not guaranteed
> to succeed.  Success depends on several factors (such as the amount of
> physical memory available and the level of fragmentation).  With the
> addition of dynamic hugetlb pool resizing, allocations can occur much more
> frequently.  For these reasons it is desirable to keep track of huge page
> allocation successes and failures.
> 
> Add two new vmstat entries to track huge page allocations that succeed and
> fail.  The presence of the two entries is contingent upon
> CONFIG_HUGETLB_PAGE being enabled.
> 
> This patch was created against linux-2.6.25-rc5
> 

It applied cleanly to 2.6.25-rc5-mm1 as well so I tested using a small x86
machine and found no problems. There was one curiousity with libhugetlbfs
with dynamic pool resizing was that two failures were accounted for when the
mapping was larger than the number of huge pages could be allocated. However,
it turned out that the first failure was libhugetlbfs calling mlock() but
not existing on failure and the second was the page fault. So, while I didn't
expect it in advance, it is still correct accounting.

Tested-by: Mel Gorman <mel@csn.ul.ie>

> Signed-off-by: Adam Litke <agl@us.ibm.com>
> Signed-off-by: Eric Munson <ebmunson@us.ibm.com>
> 
>  include/linux/vmstat.h |    8 +++++++-
>  mm/hugetlb.c           |    7 +++++++
>  mm/vmstat.c            |    4 ++++
>  3 files changed, 18 insertions(+), 1 deletions(-)
> 
> diff --git a/include/linux/vmstat.h b/include/linux/vmstat.h
> index 9f1b4b4..f68f538 100644
> --- a/include/linux/vmstat.h
> +++ b/include/linux/vmstat.h
> @@ -25,6 +25,12 @@
>  #define HIGHMEM_ZONE(xx)
>  #endif
>  
> +#ifdef CONFIG_HUGETLB_PAGE
> +#define HTLB_STATS	HTLB_BUDDY_PGALLOC, HTLB_BUDDY_PGALLOC_FAIL,
> +#else
> +#define HTLB_STATS
> +#endif
> +
>  #define FOR_ALL_ZONES(xx) DMA_ZONE(xx) DMA32_ZONE(xx) xx##_NORMAL HIGHMEM_ZONE(xx) , xx##_MOVABLE
>  
>  enum vm_event_item { PGPGIN, PGPGOUT, PSWPIN, PSWPOUT,
> @@ -36,7 +42,7 @@ enum vm_event_item { PGPGIN, PGPGOUT, PSWPIN, PSWPOUT,
>  		FOR_ALL_ZONES(PGSCAN_KSWAPD),
>  		FOR_ALL_ZONES(PGSCAN_DIRECT),
>  		PGINODESTEAL, SLABS_SCANNED, KSWAPD_STEAL, KSWAPD_INODESTEAL,
> -		PAGEOUTRUN, ALLOCSTALL, PGROTATED,
> +		PAGEOUTRUN, ALLOCSTALL, PGROTATED, HTLB_STATS
>  		NR_VM_EVENT_ITEMS
>  };
>  
> diff --git a/mm/hugetlb.c b/mm/hugetlb.c
> index 74c1b6b..dd20cb0 100644
> --- a/mm/hugetlb.c
> +++ b/mm/hugetlb.c
> @@ -239,6 +239,11 @@ static int alloc_fresh_huge_page(void)
>  		hugetlb_next_nid = next_nid;
>  	} while (!page && hugetlb_next_nid != start_nid);
>  
> +	if (ret)
> +		count_vm_event(HTLB_BUDDY_PGALLOC);
> +	else
> +		count_vm_event(HTLB_BUDDY_PGALLOC_FAIL);
> +
>  	return ret;
>  }
>  
> @@ -299,9 +304,11 @@ static struct page *alloc_buddy_huge_page(struct vm_area_struct *vma,
>  		 */
>  		nr_huge_pages_node[nid]++;
>  		surplus_huge_pages_node[nid]++;
> +		__count_vm_event(HTLB_BUDDY_PGALLOC);
>  	} else {
>  		nr_huge_pages--;
>  		surplus_huge_pages--;
> +		__count_vm_event(HTLB_BUDDY_PGALLOC_FAIL);
>  	}
>  	spin_unlock(&hugetlb_lock);
>  
> diff --git a/mm/vmstat.c b/mm/vmstat.c
> index 422d960..bbe728d 100644
> --- a/mm/vmstat.c
> +++ b/mm/vmstat.c
> @@ -644,6 +644,10 @@ static const char * const vmstat_text[] = {
>  	"allocstall",
>  
>  	"pgrotated",
> +#ifdef CONFIG_HUGETLB_PAGE
> +	"htlb_alloc_success",
> +	"htlb_alloc_fail",
> +#endif
>  #endif
>  };
>  



-- 
Mel Gorman
Part-time Phd Student                          Linux Technology Center
University of Limerick                         IBM Dublin Software Lab

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>

Re: [PATCH] hugetlb: vmstat events for huge page allocations v3

[Andy Whitcroft]

On Mon, Mar 17, 2008 at 01:02:55PM -0700, Eric B Munson wrote:
> From: Adam Litke <agl@us.ibm.com>
> 
> Following feedback here is v3
> 
> Allocating huge pages directly from the buddy allocator is not guaranteed
> to succeed.  Success depends on several factors (such as the amount of
> physical memory available and the level of fragmentation).  With the
> addition of dynamic hugetlb pool resizing, allocations can occur much more
> frequently.  For these reasons it is desirable to keep track of huge page
> allocation successes and failures.
> 
> Add two new vmstat entries to track huge page allocations that succeed and
> fail.  The presence of the two entries is contingent upon
> CONFIG_HUGETLB_PAGE being enabled.
> 
> This patch was created against linux-2.6.25-rc5
> 
> Signed-off-by: Adam Litke <agl@us.ibm.com>
> Signed-off-by: Eric Munson <ebmunson@us.ibm.com>
> 
>  include/linux/vmstat.h |    8 +++++++-
>  mm/hugetlb.c           |    7 +++++++
>  mm/vmstat.c            |    4 ++++
>  3 files changed, 18 insertions(+), 1 deletions(-)
> 
> diff --git a/include/linux/vmstat.h b/include/linux/vmstat.h
> index 9f1b4b4..f68f538 100644
> --- a/include/linux/vmstat.h
> +++ b/include/linux/vmstat.h
> @@ -25,6 +25,12 @@
>  #define HIGHMEM_ZONE(xx)
>  #endif
>  
> +#ifdef CONFIG_HUGETLB_PAGE
> +#define HTLB_STATS	HTLB_BUDDY_PGALLOC, HTLB_BUDDY_PGALLOC_FAIL,
> +#else
> +#define HTLB_STATS
> +#endif
> +
>  #define FOR_ALL_ZONES(xx) DMA_ZONE(xx) DMA32_ZONE(xx) xx##_NORMAL HIGHMEM_ZONE(xx) , xx##_MOVABLE
>  
>  enum vm_event_item { PGPGIN, PGPGOUT, PSWPIN, PSWPOUT,
> @@ -36,7 +42,7 @@ enum vm_event_item { PGPGIN, PGPGOUT, PSWPIN, PSWPOUT,
>  		FOR_ALL_ZONES(PGSCAN_KSWAPD),
>  		FOR_ALL_ZONES(PGSCAN_DIRECT),
>  		PGINODESTEAL, SLABS_SCANNED, KSWAPD_STEAL, KSWAPD_INODESTEAL,
> -		PAGEOUTRUN, ALLOCSTALL, PGROTATED,
> +		PAGEOUTRUN, ALLOCSTALL, PGROTATED, HTLB_STATS
>  		NR_VM_EVENT_ITEMS
>  };
>  
> diff --git a/mm/hugetlb.c b/mm/hugetlb.c
> index 74c1b6b..dd20cb0 100644
> --- a/mm/hugetlb.c
> +++ b/mm/hugetlb.c
> @@ -239,6 +239,11 @@ static int alloc_fresh_huge_page(void)
>  		hugetlb_next_nid = next_nid;
>  	} while (!page && hugetlb_next_nid != start_nid);
>  
> +	if (ret)
> +		count_vm_event(HTLB_BUDDY_PGALLOC);
> +	else
> +		count_vm_event(HTLB_BUDDY_PGALLOC_FAIL);
> +
>  	return ret;
>  }
>  
> @@ -299,9 +304,11 @@ static struct page *alloc_buddy_huge_page(struct vm_area_struct *vma,
>  		 */
>  		nr_huge_pages_node[nid]++;
>  		surplus_huge_pages_node[nid]++;
> +		__count_vm_event(HTLB_BUDDY_PGALLOC);
>  	} else {
>  		nr_huge_pages--;
>  		surplus_huge_pages--;
> +		__count_vm_event(HTLB_BUDDY_PGALLOC_FAIL);
>  	}
>  	spin_unlock(&hugetlb_lock);
>  
> diff --git a/mm/vmstat.c b/mm/vmstat.c
> index 422d960..bbe728d 100644
> --- a/mm/vmstat.c
> +++ b/mm/vmstat.c
> @@ -644,6 +644,10 @@ static const char * const vmstat_text[] = {
>  	"allocstall",
>  
>  	"pgrotated",
> +#ifdef CONFIG_HUGETLB_PAGE
> +	"htlb_alloc_success",
> +	"htlb_alloc_fail",

I think I was expecting these to follow the names events.
htlb_buddy_alloc_{success,fail}.  Just in case that we do ever want to
add consumer stats on htlb_alloc_{success,fail}.

> +#endif
>  #endif
>  };

Other than the output tags, that looks pretty good.

Reviewed-by: Andy Whitcroft <apw@shadowen.org>

-apw

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>