mm,page_owner: Defer enablement of static branch
Subsystems (2)
PATCH 1/1 mm,page_owner: Defer enablement of static branch
Kefeng Wang reported that he was seeing some memory leaks with kmemleak with page_owner enabled.
The reason behind is that we enable the page_owner_inited static branch and then proceed with the linking of stack_list struct to dummy_stack, which means that exists a race window between these two steps where we can have pages already being allocated calling add_stack_record_to_list(), allocating objects and linking them to stack_list, but then we set stack_list pointing to dummy_stack in init_page_owner.
Which means that the objects that have been allocated during that time window are unreferenced and lost.
Fix this by deferring the enablement of the branch until we have properly set up the list.
---
Special thanks and kudos go to Kefeng Wang for helping me out chasing down this bug, as I could not reproduce it with any of my machines, and to Vlastimil to bring another pair of eyes, which was very helpful.
mm/page_owner.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
mm/page_owner.c +1 -1
@@ -118,7 +118,6 @@ static __init void init_page_owner(void)register_dummy_stack ();register_failure_stack ();register_early_stack (); -static_branch_enable (&page_owner_inited);init_early_allocated_pages ();/* Initialize dummy and failure stacks and link them to stack_list */ dummy_stack.stack_record =__stack_depot_get_stack_record (dummy_handle);
@@ -129,6 +128,7 @@ static __init void init_page_owner(void)refcount_set (&failure_stack.stack_record->count,1 ); dummy_stack.next = &failure_stack; stack_list = &dummy_stack; +static_branch_enable (&page_owner_inited); }struct page_ext_operations page_owner_ops = {
Signature · 1 line
2.44.0
Quoted 11 lines · Oscar Salvador
Kefeng Wang reported that he was seeing some memory leaks with kmemleak with
page_ownerenabled.
The reason behind is that we enable thepage_owner_initedstatic branch and then proceed with the linking ofstack_liststruct todummy_stack, which means that exists a race window between these two steps where we can have pages already being allocated callingadd_stack_record_to_list(), allocating objects and linking them tostack_list, but then we setstack_listpointing todummy_stackininit_page_owner.
Which means that the objects that have been allocated during that time window are unreferenced and lost.
Quoted 7 lines
Fix this by deferring the enablement of the branch until we have properly set up the list.
Reported-by Kefeng Wang <[email protected]>Tested-by Kefeng Wang <[email protected]>Fixes 4bedfb314bdd ("mm,page_owner: maintain own list of stack_records structs")
That's in rc1 so ideally mm-hotfixes.
Quoted 1 line
Signed-off-by Oscar Salvador <[email protected]>
Quoted 28 lines
---
Special thanks and kudos go to Kefeng Wang for helping me out chasing down this bug, as I could not reproduce it with any of my machines, and to Vlastimil to bring another pair of eyes, which was very helpful.mm/page_owner.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)mm/page_owner.c +1
-1@@ -118,7 +118,6 @@ static __init void init_page_owner(void)register_dummy_stack ();register_failure_stack ();register_early_stack (); -static_branch_enable (&page_owner_inited);init_early_allocated_pages ();/* Initialize dummy and failure stacks and link them to stack_list */ dummy_stack.stack_record =__stack_depot_get_stack_record (dummy_handle);@@ -129,6 +128,7 @@ static __init void init_page_owner(void)refcount_set (&failure_stack.stack_record->count,1 ); dummy_stack.next = &failure_stack; stack_list = &dummy_stack; +static_branch_enable (&page_owner_inited); }struct page_ext_operations page_owner_ops = {
Quoted 19 lines · Oscar Salvador
Kefeng Wang reported that he was seeing some memory leaks with kmemleak with
page_ownerenabled.
The reason behind is that we enable thepage_owner_initedstatic branch and then proceed with the linking ofstack_liststruct todummy_stack, which means that exists a race window between these two steps where we can have pages already being allocated callingadd_stack_record_to_list(), allocating objects and linking them tostack_list, but then we setstack_listpointing todummy_stackininit_page_owner.
Which means that the objects that have been allocated during that time window are unreferenced and lost.Fix this by deferring the enablement of the branch until we have properly set up the list.
Reported-by Kefeng Wang <[email protected]>Tested-by Kefeng Wang <[email protected]>Fixes 4bedfb314bdd ("mm,page_owner: maintain own list of stack_records structs")Signed-off-by Oscar Salvador <[email protected]>
Quoted 5 lines
---
Special thanks and kudos go to Kefeng Wang for helping me out chasing down this bug, as I could not reproduce it with any of my machines, and to Vlastimil to bring another pair of eyes, which was very helpful.
The issue is found by accident when test my migrate changes, thanks for your great job and quick response, this does pass my test, thanks ;)
Quoted 23 lines
mm/page_owner.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)mm/page_owner.c +1
-1@@ -118,7 +118,6 @@ static __init void init_page_owner(void)register_dummy_stack ();register_failure_stack ();register_early_stack (); -static_branch_enable (&page_owner_inited);init_early_allocated_pages ();/* Initialize dummy and failure stacks and link them to stack_list */ dummy_stack.stack_record =__stack_depot_get_stack_record (dummy_handle);@@ -129,6 +128,7 @@ static __init void init_page_owner(void)refcount_set (&failure_stack.stack_record->count,1 ); dummy_stack.next = &failure_stack; stack_list = &dummy_stack; +static_branch_enable (&page_owner_inited); }struct page_ext_operations page_owner_ops = {