Skip to contentSkip to conversation

Fix spanned_pages is not updated at a case of memory hot-add.

PATCH In review Yasunori Goto 2 participants 6 messages May 23, 2006 – May 26, 2006

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 Fix spanned_pages is not updated at a case of memory hot-add.

Hello.

I found there is a bug in grow_zone_span() and grow_pgdat_span().

Please apply.

---------
If hot-added memory's address is smaller than old area, spanned_pages will not be updated. It must be fixed.

example) Old zone_start_pfn = 0x60000, and spanned_pages = 0x10000

         Added new memory's start_pfn = 0x50000, and end_pfn = 0x60000

new spanned_pages will be still 0x10000 by old code.
(It should be updated to 0x20000.) Because old_zone_end_pfn will be 0x70000, and end_pfn smaller than it. So, spanned_pages will not be updated.

In current code, spanned_pages is updated only when end_pfn is updated. But, it should be updated even if end_pfn is not updated, because start_pfn might be changed.

This is for 2.6.17-rc4-mm3.
I tested this patch on Tiger4 with my node emulation.

------------------------------------------------

 mm/memory_hotplug.c |   18 +++++++++++-------
 1 files changed, 11 insertions(+), 7 deletions(-)
mm/memory_hotplug.c +11 -7
Index: pgdat15/mm/memory_hotplug.c
===================================================================
@@ -95,16 +95,18 @@ EXPORT_SYMBOL_GPL(__add_pages);
 static void grow_zone_span(struct zone *zone,
 		unsigned long start_pfn, unsigned long end_pfn)
 {
-	unsigned long old_zone_end_pfn;
+	unsigned long new_zone_end_pfn;
 
 	zone_span_writelock(zone);
 
-	old_zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages;
+	new_zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages;
 	if (start_pfn < zone->zone_start_pfn)
 		zone->zone_start_pfn = start_pfn;
 
-	if (end_pfn > old_zone_end_pfn)
-		zone->spanned_pages = end_pfn - zone->zone_start_pfn;
+	if (end_pfn > new_zone_end_pfn)
+		 new_zone_end_pfn = end_pfn;
+
+	zone->spanned_pages = new_zone_end_pfn - zone->zone_start_pfn;
 
 	zone_span_writeunlock(zone);
 }
@@ -112,14 +114,16 @@ static void grow_zone_span(struct zone *
 static void grow_pgdat_span(struct pglist_data *pgdat,
 		unsigned long start_pfn, unsigned long end_pfn)
 {
-	unsigned long old_pgdat_end_pfn =
+	unsigned long new_pgdat_end_pfn =
 		pgdat->node_start_pfn + pgdat->node_spanned_pages;
 
 	if (start_pfn < pgdat->node_start_pfn)
 		pgdat->node_start_pfn = start_pfn;
 
-	if (end_pfn > old_pgdat_end_pfn)
-		pgdat->node_spanned_pages = end_pfn - pgdat->node_start_pfn;
+	if (end_pfn > new_pgdat_end_pfn)
+		new_pgdat_end_pfn = end_pfn;
+
+	pgdat->node_spanned_pages = new_pgdat_end_pfn - pgdat->node_start_pfn;
 }
 
 int online_pages(unsigned long pfn, unsigned long nr_pages)
Signature · 1 line
Yasunori Goto 
#1 lore raw
Quoted 3 lines · Yasunori Goto

Hello.

I found there is a bug in grow_zone_span() and grow_pgdat_span().

...

Review EXPORT_SYMBOL_GPL(__add_pages); · line 102

Quoted 11 lines
(quoted diff) +2 -2
@@ -95,16 +95,18 @@ EXPORT_SYMBOL_GPL(__add_pages);
 static void grow_zone_span(struct zone *zone,
 		unsigned long start_pfn, unsigned long end_pfn)
 {
-	unsigned long old_zone_end_pfn;
+	unsigned long new_zone_end_pfn;
 
 	zone_span_writelock(zone);
 
-	old_zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages;
+	new_zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages;

I really don't like the idea of having this variable called "new_" something. That implies that this is what the new end_pfn is going to be. The *new* one. In reality, it is what it _might_ have been. How about "tmp_zone_end_pfn"?

This practice of dealing with spanned_pages is a real pain.

I generally try to avoid max/min in code, but this struck me as possibly being useful. Do you find this easier to read, or your patch?

mm/memory_hotplug.c renamed from mm/memory_hotplug.c~fix-spanned-pages +4 -4
diff -puN mm/memory_hotplug.c~fix-spanned-pages mm/memory_hotplug.c
@@ -91,8 +91,8 @@ static void grow_zone_span(struct zone *
 	if (start_pfn < zone->zone_start_pfn)
 		zone->zone_start_pfn = start_pfn;
 
-	if (end_pfn > old_zone_end_pfn)
-		zone->spanned_pages = end_pfn - zone->zone_start_pfn;
+	zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
+				zone->zone_start_pfn);
 
 	zone_span_writeunlock(zone);
 }
@@ -106,8 +106,8 @@ static void grow_pgdat_span(struct pglis
 	if (start_pfn < pgdat->node_start_pfn)
 		pgdat->node_start_pfn = start_pfn;
 
-	if (end_pfn > old_pgdat_end_pfn)
-		pgdat->node_spanned_pages = end_pfn - pgdat->node_start_pfn;
+	pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
+					pgdat->node_start_pfn;
 }
 
 int online_pages(unsigned long pfn, unsigned long nr_pages)

_

-- Dave

#2 lore raw
Quoted 9 lines

I really don't like the idea of having this variable called "new_" something. That implies that this is what the new end_pfn is going to be. The *new* one. In reality, it is what it _might_ have been. How about "tmp_zone_end_pfn"?

This practice of dealing with spanned_pages is a real pain.

I generally try to avoid max/min in code, but this struck me as possibly being useful. Do you find this easier to read, or your patch?

Ah, I think using max is better due to smaller code, indeed.

mm/memory_hotplug.c static void grow_zone_span(struct zone * · line 95

Quoted 12 lines

diff -puN mm/memory_hotplug.c~fix-spanned-pages mm/memory_hotplug.c

mm/memory_hotplug.c renamed from mm/memory_hotplug.c~fix-spanned-pages +2 -2
@@ -91,8 +91,8 @@ static void grow_zone_span(struct zone *
 	if (start_pfn < zone->zone_start_pfn)
 		zone->zone_start_pfn = start_pfn;
 
-	if (end_pfn > old_zone_end_pfn)
-		zone->spanned_pages = end_pfn - zone->zone_start_pfn;
+	zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
+				zone->zone_start_pfn);

^
this parentheses is redundant. :-)

Could you fix and repost it? Or should I?

Signature · 1 line
Yasunori Goto 
#3 lore raw
Quoted 1 line · Yasunori Goto

Could you fix and repost it? Or should I?

I'd appreciate it if you could test it, and forward it along afterword.

-- Dave

#4 lore raw
Quoted 5 lines · Yasunori Goto
Quoted 1 line

Could you fix and repost it? Or should I?

I'd appreciate it if you could test it, and forward it along afterword.

Ok. I'll do it. :-)

Signature · 1 line
Yasunori Goto 
#5 lore raw

PATCH 1/1 Fix spanned_pages is not updated at a case of memory hot-add take 2..

In reply to Dave Hansen

I fixed redundant parentheses and tested this fix updating spanned_pages patch.

Andrew-san.
Please apply.

------------------
If hot-added memory's address is smaller than old area, spanned_pages will not be updated. It must be fixed.

example) Old zone_start_pfn = 0x60000, and spanned_pages = 0x10000

         Added new memory's start_pfn = 0x50000, and end_pfn = 0x60000

new spanned_pages will be still 0x10000 by old code.
(It should be updated to 0x20000.) Because old_zone_end_pfn will be 0x70000, and end_pfn smaller than it. So, spanned_pages will not be updated.

In current code, spanned_pages is updated only when end_pfn is updated. But, it should be updated by subtraction between bigger end_pfn and new zone_start_pfn.

This is for 2.6.17-rc4-mm3.
I tested this patch on Tiger4 with my node emulation.

 mm/memory_hotplug.c |    8 ++++----
 1 files changed, 4 insertions(+), 4 deletions(-)
mm/memory_hotplug.c +4 -4
Index: pgdat15/mm/memory_hotplug.c
===================================================================
@@ -103,8 +103,8 @@ static void grow_zone_span(struct zone *
 	if (start_pfn < zone->zone_start_pfn)
 		zone->zone_start_pfn = start_pfn;
 
-	if (end_pfn > old_zone_end_pfn)
-		zone->spanned_pages = end_pfn - zone->zone_start_pfn;
+	zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
+				zone->zone_start_pfn;
 
 	zone_span_writeunlock(zone);
 }
@@ -118,8 +118,8 @@ static void grow_pgdat_span(struct pglis
 	if (start_pfn < pgdat->node_start_pfn)
 		pgdat->node_start_pfn = start_pfn;
 
-	if (end_pfn > old_pgdat_end_pfn)
-		pgdat->node_spanned_pages = end_pfn - pgdat->node_start_pfn;
+	pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
+					pgdat->node_start_pfn;
 }
 
 int online_pages(unsigned long pfn, unsigned long nr_pages)
Signature · 1 line
Yasunori Goto