Catalin Marinas wrote:
> Hi Shailabh,
>
> Michal was running some kmemleak tests and there are about 20 orphan
> pointers reported in delayacct.c. The allocation backtrace is:
>
> orphan pointer 0xf548fde0 (size 76):
> c0174674: <kmem_cache_zalloc>
> c01591ee: <__delayacct_tsk_init>
> c0127e06: <copy_process>
> c0128cd2: <do_fork>
> c0104d39: <sys_clone>
>
> I'm not sure whether the leak occurs but there might be a path where
> task_struct is freed and the task->delays pointer is lost. Could you
> please have a look at this? Thanks.
>
One possibility for the leak is a missing free for tsk->delays on a
failed fork. Were the kmemleak tests causing fork failures to happen ?
What was being run in userspace ? Since the tsk->delays get allocated
from a slab, it should be easy enough to detect.
Could you try the patch below ? Its also being used for an oops reported
for delay accounting.
Thanks,
Shailabh