[v3.10-rt / v3.12-rt] scheduling while atomic in cgroup code

[v3.10-rt / v3.12-rt] scheduling while atomic in cgroup code

[Nikita Yushchenko]

Hi.

Call Trace:
[e22d5a90] [c0007ea8] show_stack+0x4c/0x168 (unreliable)
[e22d5ad0] [c0618c04] __schedule_bug+0x94/0xb0
[e22d5ae0] [c060b9ec] __schedule+0x530/0x550
[e22d5bf0] [c060bacc] schedule+0x30/0xbc
[e22d5c00] [c060ca24] rt_spin_lock_slowlock+0x180/0x27c
[e22d5c70] [c00b39dc] res_counter_uncharge_until+0x40/0xc4
[e22d5ca0] [c013ca88] drain_stock.isra.20+0x54/0x98
[e22d5cc0] [c01402ac] __mem_cgroup_try_charge+0x2e8/0xbac
[e22d5d70] [c01410d4] mem_cgroup_charge_common+0x3c/0x70
[e22d5d90] [c0117284] __do_fault+0x38c/0x510
[e22d5df0] [c011a5f4] handle_pte_fault+0x98/0x858
[e22d5e50] [c060ed08] do_page_fault+0x42c/0x6fc
[e22d5f40] [c000f5b4] handle_page_fault+0xc/0x80

What happens:

- refill_stock() calls get_cpu_var() and thus disables preemption until
matching put_cpu_var() is called,

- then it calls drain_stock() -> res_counter_uncharge() ->
res_counter_uncharge_until()

- and here we have spin_lock(), which under RT can sleep. Thus we have 
sleeping with preemption disabled.


Any ideas how to fix?

Re: [v3.10-rt / v3.12-rt] scheduling while atomic in cgroup code

[Mike Galbraith]

On Sat, 2014-06-21 at 07:55 +0400, Nikita Yushchenko wrote: 
> Hi.
> 
> Call Trace:
> [e22d5a90] [c0007ea8] show_stack+0x4c/0x168 (unreliable)
> [e22d5ad0] [c0618c04] __schedule_bug+0x94/0xb0
> [e22d5ae0] [c060b9ec] __schedule+0x530/0x550
> [e22d5bf0] [c060bacc] schedule+0x30/0xbc
> [e22d5c00] [c060ca24] rt_spin_lock_slowlock+0x180/0x27c
> [e22d5c70] [c00b39dc] res_counter_uncharge_until+0x40/0xc4
> [e22d5ca0] [c013ca88] drain_stock.isra.20+0x54/0x98
> [e22d5cc0] [c01402ac] __mem_cgroup_try_charge+0x2e8/0xbac
> [e22d5d70] [c01410d4] mem_cgroup_charge_common+0x3c/0x70
> [e22d5d90] [c0117284] __do_fault+0x38c/0x510
> [e22d5df0] [c011a5f4] handle_pte_fault+0x98/0x858
> [e22d5e50] [c060ed08] do_page_fault+0x42c/0x6fc
> [e22d5f40] [c000f5b4] handle_page_fault+0xc/0x80
> 
> What happens:
> 
> - refill_stock() calls get_cpu_var() and thus disables preemption until
> matching put_cpu_var() is called,
> 
> - then it calls drain_stock() -> res_counter_uncharge() ->
> res_counter_uncharge_until()
> 
> - and here we have spin_lock(), which under RT can sleep. Thus we have 
> sleeping with preemption disabled.
> 
> 
> Any ideas how to fix?

The below should work.. though not as well as turning it off.

mm, memcg: make refill_stock()/consume_stock() use get_cpu_light()

Nikita reported the following memcg scheduling while atomic bug:

Call Trace:
[e22d5a90] [c0007ea8] show_stack+0x4c/0x168 (unreliable)
[e22d5ad0] [c0618c04] __schedule_bug+0x94/0xb0
[e22d5ae0] [c060b9ec] __schedule+0x530/0x550
[e22d5bf0] [c060bacc] schedule+0x30/0xbc
[e22d5c00] [c060ca24] rt_spin_lock_slowlock+0x180/0x27c
[e22d5c70] [c00b39dc] res_counter_uncharge_until+0x40/0xc4
[e22d5ca0] [c013ca88] drain_stock.isra.20+0x54/0x98
[e22d5cc0] [c01402ac] __mem_cgroup_try_charge+0x2e8/0xbac
[e22d5d70] [c01410d4] mem_cgroup_charge_common+0x3c/0x70
[e22d5d90] [c0117284] __do_fault+0x38c/0x510
[e22d5df0] [c011a5f4] handle_pte_fault+0x98/0x858
[e22d5e50] [c060ed08] do_page_fault+0x42c/0x6fc
[e22d5f40] [c000f5b4] handle_page_fault+0xc/0x80

What happens:

   refill_stock()
      get_cpu_var()
      drain_stock()
         res_counter_uncharge()
            res_counter_uncharge_until()
               spin_lock() <== boom

Fix it by replacing get/put_cpu_var() with get/put_cpu_light().

Reported-by: Nikita Yushchenko <nyushchenko@dev.rtsoft.ru>
Signed-off-by: Mike Galbraith <umgwanakikbuti@gmail.com>
---
 mm/memcontrol.c |   13 +++++++++----
 1 file changed, 9 insertions(+), 4 deletions(-)

--- a/mm/memcontrol.c
+++ b/mm/memcontrol.c
@@ -2398,16 +2398,18 @@ static bool consume_stock(struct mem_cgr
 {
 	struct memcg_stock_pcp *stock;
 	bool ret = true;
+	int cpu;
 
 	if (nr_pages > CHARGE_BATCH)
 		return false;
 
-	stock = &get_cpu_var(memcg_stock);
+	cpu = get_cpu_light();
+	stock = &per_cpu(memcg_stock, cpu);
 	if (memcg == stock->cached && stock->nr_pages >= nr_pages)
 		stock->nr_pages -= nr_pages;
 	else /* need to call res_counter_charge */
 		ret = false;
-	put_cpu_var(memcg_stock);
+	put_cpu_light();
 	return ret;
 }
 
@@ -2457,14 +2459,17 @@ static void __init memcg_stock_init(void
  */
 static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
 {
-	struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock);
+	struct memcg_stock_pcp *stock;
+	int cpu = get_cpu_light();
+
+	stock = &per_cpu(memcg_stock, cpu);
 
 	if (stock->cached != memcg) { /* reset if necessary */
 		drain_stock(stock);
 		stock->cached = memcg;
 	}
 	stock->nr_pages += nr_pages;
-	put_cpu_var(memcg_stock);
+	put_cpu_light();
 }
 
 /*

Re: [v3.10-rt / v3.12-rt] scheduling while atomic in cgroup code

[Sebastian Andrzej Siewior]

* Mike Galbraith | 2014-06-21 10:09:48 [+0200]:

>--- a/mm/memcontrol.c
>+++ b/mm/memcontrol.c
>@@ -2398,16 +2398,18 @@ static bool consume_stock(struct mem_cgr
> {
> 	struct memcg_stock_pcp *stock;
> 	bool ret = true;
>+	int cpu;
> 
> 	if (nr_pages > CHARGE_BATCH)
> 		return false;
> 
>-	stock = &get_cpu_var(memcg_stock);
>+	cpu = get_cpu_light();
>+	stock = &per_cpu(memcg_stock, cpu);
> 	if (memcg == stock->cached && stock->nr_pages >= nr_pages)
> 		stock->nr_pages -= nr_pages;
> 	else /* need to call res_counter_charge */
> 		ret = false;
>-	put_cpu_var(memcg_stock);
>+	put_cpu_light();
> 	return ret;
> }

I am not taking this chunk. That preempt_disable() is lower weight
and there is nothing happening that does not work with it.

>@@ -2457,14 +2459,17 @@ static void __init memcg_stock_init(void
>  */
> static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages)
> {
>-	struct memcg_stock_pcp *stock = &get_cpu_var(memcg_stock);
>+	struct memcg_stock_pcp *stock;
>+	int cpu = get_cpu_light();
>+
>+	stock = &per_cpu(memcg_stock, cpu);
> 
> 	if (stock->cached != memcg) { /* reset if necessary */
> 		drain_stock(stock);
> 		stock->cached = memcg;
> 	}

I am a little more worried that drain_stock() could be called more than
once on the same CPU.
- memcg_cpu_hotplug_callback() doesn't disable preemption
- drain_local_stock() doesn't as well

so maybe it doesn't matter.

> 	stock->nr_pages += nr_pages;
>-	put_cpu_var(memcg_stock);
>+	put_cpu_light();
> }

Sebastian