Re: [PATCH] mm: memcontrol: fix rcu unbalance in get_non_dying_memcg_end()

[Qi Zheng <qi.zheng@linux.dev>]

On 4/28/26 5:59 PM, Shakeel Butt wrote:
> On Tue, Apr 28, 2026 at 11:06:21AM +0800, Qi Zheng wrote:
>> From: Qi Zheng <zhengqi.arch@bytedance.com>
>>
>> Currently, get_non_dying_memcg_start() and get_non_dying_memcg_end() both
>> evaluate cgroup_subsys_on_dfl(memory_cgrp_subsys) independently to
>> determine whether to acquire or release the RCU read lock.
>>
>> However, the result of cgroup_subsys_on_dfl() can change dynamically at
>> runtime due to cgroup hierarchy rebinding (e.g., when the memory
>> controller is moved between cgroup v1 and v2 hierarchies). This can cause
>> the following warning:
>>
>>   =====================================
>>   WARNING: bad unlock balance detected!
>>   7.0.0-next-20260420+ #83 Tainted: G        W
>>   -------------------------------------
>>   memcg-repro/270 is trying to release lock (rcu_read_lock) at:
>>   [<ffffffff815f57f7>] rcu_read_unlock+0x17/0x60
>>   but there are no more locks to release!
>>
>>   other info that might help us debug this:
>>   1 lock held by memcg-repro/270:
>>    #0: ffff888102fa2088 (vm_lock){++++}-{0:0}, at: do_user_addr_fault+0x285/0x880
>>
>>   stack backtrace:
>>   CPU: 0 UID: 0 PID: 270 Comm: memcg-repro Tainted: G        W           7.0.0-next-20260420+ #
>>   Tainted: [W]=WARN
>>   Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.12.0-1 04/01/2014
>>   Call Trace:
>>    <TASK>
>>    ? rcu_read_unlock+0x17/0x60
>>    dump_stack_lvl+0x77/0xb0
>>    print_unlock_imbalance_bug+0xe0/0xf0
>>    ? rcu_read_unlock+0x17/0x60
>>    lock_release+0x21d/0x2a0
>>    rcu_read_unlock+0x1c/0x60
>>    do_pte_missing+0x233/0xb40
>>    __handle_mm_fault+0x80e/0xcd0
>>    handle_mm_fault+0x146/0x310
>>    do_user_addr_fault+0x303/0x880
>>    exc_page_fault+0x9b/0x270
>>    asm_exc_page_fault+0x26/0x30
>>   RIP: 0033:0x5590e4eb41ea
>>   Code: 61 cc 66 0f 6f e0 66 0f 61 c2 66 0f db cd 66 0f 69 e2 66 0f 6f d0 66 0f 69 d4 66 0f 61 0
>>   RSP: 002b:00007ffcad25f030 EFLAGS: 00010202
>>   RAX: 00005590e4eb8010 RBX: 00007ffcad260f7d RCX: 00007f73c474d44d
>>   RDX: 00005590e4eb80a0 RSI: 00005590e4eb503c RDI: 000000000000000f
>>   RBP: 00005590e4eb70a0 R08: 0000000000000000 R09: 00007f73c483a680
>>   R10: 0000000000000000 R11: 0000000000000246 R12: 0000000000000000
>>   R13: 00007ffcad25f180 R14: 00005590e4eb6dd8 R15: 00007f73c4869020
>>    </TASK>
>>   ------------[ cut here ]------------
>>
>> Fix this by explicitly tracking the RCU lock state, ensuring that
>> rcu_read_unlock() in get_non_dying_memcg_end() is strictly paired with
>> the lock acquisition, regardless of any runtime rebinding events.
>>
>> Fixes: 8285917d6f38 ("mm: memcontrol: prepare for reparenting non-hierarchical stats")
>> Signed-off-by: Qi Zheng <zhengqi.arch@bytedance.com>
>> ---
>>   mm/memcontrol.c | 31 +++++++++++++++++++++----------
>>   1 file changed, 21 insertions(+), 10 deletions(-)
>>
>> diff --git a/mm/memcontrol.c b/mm/memcontrol.c
>> index c3d98ab41f1f1..38f48a45b7ae5 100644
>> --- a/mm/memcontrol.c
>> +++ b/mm/memcontrol.c
>> @@ -805,12 +805,17 @@ static long memcg_state_val_in_pages(int idx, long val)
>>    * Used in mod_memcg_state() and mod_memcg_lruvec_state() to avoid race with
>>    * reparenting of non-hierarchical state_locals.
>>    */
>> -static inline struct mem_cgroup *get_non_dying_memcg_start(struct mem_cgroup *memcg)
>> +static inline struct mem_cgroup *get_non_dying_memcg_start(struct mem_cgroup *memcg,
>> +							   bool *rcu_locked)
>>   {
>> -	if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
>> +	/* Rebinding can cause this value to be changed at runtime */
>> +	if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) {
>> +		*rcu_locked = false;
>>   		return memcg;
>> +	}
>>   
>>   	rcu_read_lock();
>> +	*rcu_locked = true;
>>   
>>   	while (memcg_is_dying(memcg))
>>   		memcg = parent_mem_cgroup(memcg);
>> @@ -818,20 +823,23 @@ static inline struct mem_cgroup *get_non_dying_memcg_start(struct mem_cgroup *me
>>   	return memcg;
>>   }
>>   
>> -static inline void get_non_dying_memcg_end(void)
>> +static inline void get_non_dying_memcg_end(bool rcu_locked)
>>   {
>> -	if (cgroup_subsys_on_dfl(memory_cgrp_subsys))
>> +	if (!rcu_locked)
>>   		return;
>>   
>>   	rcu_read_unlock();
>>   }
>>   #else
>> -static inline struct mem_cgroup *get_non_dying_memcg_start(struct mem_cgroup *memcg)
>> +static inline struct mem_cgroup *get_non_dying_memcg_start(struct mem_cgroup *memcg,
>> +							   bool *rcu_locked)
>>   {
>> +	*rcu_locked = false;
> 
> We don't need to set rcu_locked to false here as we don't access in !V1 build
> option.

Will do.

> 
> With that fixed, you can add:
> 
> Acked-by: Shakeel Butt <shakeel.butt@linux.dev>

Thanks!