Skip to contentSkip to conversation

mm: Do not walk all of system memory during show_mem

PATCH In review Mel Gorman 3 participants 6 messages Oct 16, 2013 – Dec 3, 2013

Subsystems (6)
lore
Share & tools
lore

Mel Gorman

author
#0 lore raw

PATCH 1/1 mm: Do not walk all of system memory during show_mem

It has been reported on very large machines that show_mem is taking almost 5 minutes to display information. This is a serious problem if there is an OOM storm. The bulk of the cost is in show_mem doing a very expensive PFN walk to give us the following information

Total RAM: Also available as totalram_pages
Highmem pages: Also available as totalhigh_pages
Reserved pages: Can be inferred from the zone structure
Shared pages: PFN walk required
Unshared pages: PFN walk required
Quick pages: Per-cpu walk required

Only the shared/unshared pages requires a full PFN walk but that information is useless. It is also inaccurate as page pins of unshared pages would be accounted for as shared. Even if the information was accurate, I'm struggling to think how the shared/unshared information could be useful for debugging OOM conditions. Maybe it was useful before rmap existed when reclaiming shared pages was costly but it is less relevant today.

The PFN walk could be optimised a bit but why bother as the information is useless. This patch deletes the PFN walker and infers the total RAM, highmem and reserved pages count from struct zone. It omits the shared/unshared page usage on the grounds that it is useless. It also corrects the reporting of HighMem as HighMem/MovableOnly as ZONE_MOVABLE has similar problems to HighMem with respect to lowmem/highmem exhaustion.

---
 lib/show_mem.c | 39 +++++++++++----------------------------
 1 file changed, 11 insertions(+), 28 deletions(-)
lib/show_mem.c +11 -28
index b7c7231..5847a49 100644
@@ -12,8 +12,7 @@
 void show_mem(unsigned int filter)
 {
 	pg_data_t *pgdat;
-	unsigned long total = 0, reserved = 0, shared = 0,
-		nonshared = 0, highmem = 0;
+	unsigned long total = 0, reserved = 0, highmem = 0;
 
 	printk("Mem-Info:\n");
 	show_free_areas(filter);
@@ -22,43 +21,27 @@ void show_mem(unsigned int filter)
 		return;
 
 	for_each_online_pgdat(pgdat) {
-		unsigned long i, flags;
+		unsigned long flags;
+		int zoneid;
 
 		pgdat_resize_lock(pgdat, &flags);
-		for (i = 0; i < pgdat->node_spanned_pages; i++) {
-			struct page *page;
-			unsigned long pfn = pgdat->node_start_pfn + i;
-
-			if (unlikely(!(i % MAX_ORDER_NR_PAGES)))
-				touch_nmi_watchdog();
-
-			if (!pfn_valid(pfn))
+		for (zoneid = 0; zoneid < MAX_NR_ZONES; zoneid++) {
+			struct zone *zone = &pgdat->node_zones[zoneid];
+			if (!populated_zone(zone))
 				continue;
 
-			page = pfn_to_page(pfn);
-
-			if (PageHighMem(page))
-				highmem++;
+			total += zone->present_pages;
+			reserved = zone->present_pages - zone->managed_pages;
 
-			if (PageReserved(page))
-				reserved++;
-			else if (page_count(page) == 1)
-				nonshared++;
-			else if (page_count(page) > 1)
-				shared += page_count(page) - 1;
-
-			total++;
+			if (is_highmem_idx(zoneid))
+				highmem += zone->present_pages;
 		}
 		pgdat_resize_unlock(pgdat, &flags);
 	}
 
 	printk("%lu pages RAM\n", total);
-#ifdef CONFIG_HIGHMEM
-	printk("%lu pages HighMem\n", highmem);
-#endif
+	printk("%lu pages HighMem/MovableOnly\n", highmem);
 	printk("%lu pages reserved\n", reserved);
-	printk("%lu pages shared\n", shared);
-	printk("%lu pages non-shared\n", nonshared);
 #ifdef CONFIG_QUICKLIST
 	printk("%lu pages in pagetable cache\n",
 		quicklist_total_size());
#1 lore raw
Quoted 19 lines · Mel Gorman

It has been reported on very large machines that show_mem is taking almost 5 minutes to display information. This is a serious problem if there is an OOM storm. The bulk of the cost is in show_mem doing a very expensive PFN walk to give us the following information

Total RAM: Also available as totalram_pages
Highmem pages: Also available as totalhigh_pages
Reserved pages: Can be inferred from the zone structure
Shared pages: PFN walk required
Unshared pages: PFN walk required
Quick pages: Per-cpu walk required

Only the shared/unshared pages requires a full PFN walk but that information is useless. It is also inaccurate as page pins of unshared pages would be accounted for as shared. Even if the information was accurate, I'm struggling to think how the shared/unshared information could be useful for debugging OOM conditions. Maybe it was useful before rmap existed when reclaiming shared pages was costly but it is less relevant today.

Quoted 7 lines

The PFN walk could be optimised a bit but why bother as the information is useless. This patch deletes the PFN walker and infers the total RAM, highmem and reserved pages count from struct zone. It omits the shared/unshared page usage on the grounds that it is useless. It also corrects the reporting of HighMem as HighMem/MovableOnly as ZONE_MOVABLE has similar problems to HighMem with respect to lowmem/highmem exhaustion.

We haven't been hit by this for the oom killer, but we did get hit with this for page allocation failure warnings as a result of having irqs disabled and passing GFP_ATOMIC to the page allocator without GFP_NOWARN. That was the intention of passing SHOW_MEM_FILTER_PAGE_COUNT into show_mem() in 4b59e6c47309 ("mm, show_mem: suppress page counts in non-blockable contexts").

With this, I assume we can just remove SHOW_MEM_FILTER_PAGE_COUNT entirely?

--
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>

author
#2 lore raw
Quoted 36 lines · David Rientjes
Quoted 19 lines · Mel Gorman

It has been reported on very large machines that show_mem is taking almost 5 minutes to display information. This is a serious problem if there is an OOM storm. The bulk of the cost is in show_mem doing a very expensive PFN walk to give us the following information

Total RAM: Also available as totalram_pages
Highmem pages: Also available as totalhigh_pages
Reserved pages: Can be inferred from the zone structure
Shared pages: PFN walk required
Unshared pages: PFN walk required
Quick pages: Per-cpu walk required

Only the shared/unshared pages requires a full PFN walk but that information is useless. It is also inaccurate as page pins of unshared pages would be accounted for as shared. Even if the information was accurate, I'm struggling to think how the shared/unshared information could be useful for debugging OOM conditions. Maybe it was useful before rmap existed when reclaiming shared pages was costly but it is less relevant today.

Quoted 7 lines

The PFN walk could be optimised a bit but why bother as the information is useless. This patch deletes the PFN walker and infers the total RAM, highmem and reserved pages count from struct zone. It omits the shared/unshared page usage on the grounds that it is useless. It also corrects the reporting of HighMem as HighMem/MovableOnly as ZONE_MOVABLE has similar problems to HighMem with respect to lowmem/highmem exhaustion.

We haven't been hit by this for the oom killer, but we did get hit with this for page allocation failure warnings as a result of having irqs disabled and passing GFP_ATOMIC to the page allocator without GFP_NOWARN. That was the intention of passing SHOW_MEM_FILTER_PAGE_COUNT into show_mem() in 4b59e6c47309 ("mm, show_mem: suppress page counts in non-blockable contexts").

Quoted 2 lines

With this, I assume we can just remove SHOW_MEM_FILTER_PAGE_COUNT entirely?

We could once all the per-arch show_mem functions were updated similar to lib/show_mem.c. I've added a todo item to do just that. Thanks.

Signature · 2 lines
Mel Gorman
SUSE Labs
author
#3 lore raw

PATCH 1/1 mm, show_mem: Remove SHOW_MEM_FILTER_PAGE_COUNT

In reply to David Rientjes

Commit 4b59e6c4 (mm, show_mem: suppress page counts in non-blockable contexts) introduced SHOW_MEM_FILTER_PAGE_COUNT to suppress PFN walks on large memory machines. Commit c78e9363 (mm: do not walk all of system memory during show_mem) avoided a PFN walk in the generic show_mem helper which removes the requirement for SHOW_MEM_FILTER_PAGE_COUNT in that case.

This patch removes PFN walkers from the arch-specific implementations that report on a per-node or per-zone granularity. ARM and unicore32 still do a PFN walk as they report memory usage on each bank which is a much finer granularity where the debugging information may still be of use. As the remaining arches doing PFN walks have relatively small amounts of memory, this patch simply removes SHOW_MEM_FILTER_PAGE_COUNT.

---
 arch/arm/mm/init.c       |  3 ---
 arch/ia64/mm/contig.c    | 68 ------------------------------------------------
 arch/ia64/mm/discontig.c | 63 --------------------------------------------
 arch/ia64/mm/init.c      | 48 ++++++++++++++++++++++++++++++++++
 arch/parisc/mm/init.c    | 62 ++++++++++++++-----------------------------
 arch/unicore32/mm/init.c |  3 ---
 include/linux/mm.h       |  1 -
 lib/show_mem.c           |  3 ---
 mm/page_alloc.c          |  7 -----
 9 files changed, 68 insertions(+), 190 deletions(-)
arch/arm/mm/init.c +0 -3
index 3e8f106..2e71e24 100644
@@ -92,9 +92,6 @@ void show_mem(unsigned int filter)
 	printk("Mem-info:\n");
 	show_free_areas(filter);
 
-	if (filter & SHOW_MEM_FILTER_PAGE_COUNT)
-		return;
-
 	for_each_bank (i, mi) {
 		struct membank *bank = &mi->bank[i];
 		unsigned int pfn1, pfn2;
arch/ia64/mm/contig.c +0 -68

74 lines not on this page · raw message

arch/ia64/mm/discontig.c +0 -63

69 lines not on this page · raw message

arch/ia64/mm/init.c +48 -0

51 lines not on this page · raw message

arch/parisc/mm/init.c +20 -42

75 lines not on this page · raw message

arch/unicore32/mm/init.c +0 -3
index ae6bc03..be2bde9 100644
@@ -66,9 +66,6 @@ void show_mem(unsigned int filter)
 	printk(KERN_DEFAULT "Mem-info:\n");
 	show_free_areas(filter);
 
-	if (filter & SHOW_MEM_FILTER_PAGE_COUNT)
-		return;
-
 	for_each_bank(i, mi) {
 		struct membank *bank = &mi->bank[i];
 		unsigned int pfn1, pfn2;
include/linux/mm.h +0 -1
index 1cedd00..42b2486 100644
@@ -984,7 +984,6 @@ extern void pagefault_out_of_memory(void);
  * various contexts.
  */
 #define SHOW_MEM_FILTER_NODES		(0x0001u)	/* disallowed nodes */
-#define SHOW_MEM_FILTER_PAGE_COUNT	(0x0002u)	/* page type count */
 
 extern void show_free_areas(unsigned int flags);
 extern bool skip_free_areas_node(unsigned int flags, int nid);
lib/show_mem.c +0 -3
index 5847a49..f58689f 100644
@@ -17,9 +17,6 @@ void show_mem(unsigned int filter)
 	printk("Mem-Info:\n");
 	show_free_areas(filter);
 
-	if (filter & SHOW_MEM_FILTER_PAGE_COUNT)
-		return;
-
 	for_each_online_pgdat(pgdat) {
 		unsigned long flags;
 		int zoneid;
mm/page_alloc.c +0 -7
index 580a5f0..dcb61e8 100644
@@ -2073,13 +2073,6 @@ void warn_alloc_failed(gfp_t gfp_mask, int order, const char *fmt, ...)
 		return;
 
 	/*
-	 * Walking all memory to count page types is very expensive and should
-	 * be inhibited in non-blockable contexts.
-	 */
-	if (!(gfp_mask & __GFP_WAIT))
-		filter |= SHOW_MEM_FILTER_PAGE_COUNT;
-
-	/*
 	 * This documents exceptions given to allocations in certain
 	 * contexts that are allowed to allocate outside current's set
 	 * of allowed nodes.
Acked-by on v1 1/1
#4 lore raw

Re: [PATCH] mm, show_mem: Remove SHOW_MEM_FILTER_PAGE_COUNT

Quoted 6 lines · Mel Gorman

Commit 4b59e6c4 (mm, show_mem: suppress page counts in non-blockable contexts) introduced SHOW_MEM_FILTER_PAGE_COUNT to suppress PFN walks on large memory machines. Commit c78e9363 (mm: do not walk all of system memory during show_mem) avoided a PFN walk in the generic show_mem helper which removes the requirement for SHOW_MEM_FILTER_PAGE_COUNT in that case.

Quoted 8 lines

This patch removes PFN walkers from the arch-specific implementations that report on a per-node or per-zone granularity. ARM and unicore32 still do a PFN walk as they report memory usage on each bank which is a much finer granularity where the debugging information may still be of use. As the remaining arches doing PFN walks have relatively small amounts of memory, this patch simply removes SHOW_MEM_FILTER_PAGE_COUNT.

Thanks!

Acked-by on v1 1/1
#5 lore raw
In reply to Mel Gorman
Quoted 19 lines · Mel Gorman

It has been reported on very large machines that show_mem is taking almost 5 minutes to display information. This is a serious problem if there is an OOM storm. The bulk of the cost is in show_mem doing a very expensive PFN walk to give us the following information

Total RAM: Also available as totalram_pages
Highmem pages: Also available as totalhigh_pages
Reserved pages: Can be inferred from the zone structure
Shared pages: PFN walk required
Unshared pages: PFN walk required
Quick pages: Per-cpu walk required

Only the shared/unshared pages requires a full PFN walk but that information is useless. It is also inaccurate as page pins of unshared pages would be accounted for as shared. Even if the information was accurate, I'm struggling to think how the shared/unshared information could be useful for debugging OOM conditions. Maybe it was useful before rmap existed when reclaiming shared pages was costly but it is less relevant today.

Quoted 8 lines

The PFN walk could be optimised a bit but why bother as the information is useless. This patch deletes the PFN walker and infers the total RAM, highmem and reserved pages count from struct zone. It omits the shared/unshared page usage on the grounds that it is useless. It also corrects the reporting of HighMem as HighMem/MovableOnly as ZONE_MOVABLE has similar problems to HighMem with respect to lowmem/highmem exhaustion.

That's ok. I haven't used such information on my long oom debugging history.