lockdep warning: module_mutex vs. ftrace_lock
Quoted 43 lines · Steven Rostedt
Quoted 41 lines · Frederic Weisbecker
Quoted 40 lines · Miklos Szeredi
=======================================================
[ INFO: possible circular locking dependency detected ] 2.6.29-09854-gd508afb-dirty #6
-------------------------------------------------------
modprobe/3184 is trying to acquire lock:
(ftrace_lock){+.+.+.}, at: [<ffffffff80277e71>]ftrace_convert_nops+0x32/0x267but task is already holding lock:
(module_mutex){+.+.+.}, at: [<ffffffff8026a539>]sys_init_module+0x3f/0x1d3which lock already depends on the new lock.
the existing dependency chain (in reverse order) is:
-> #1 (module_mutex){+.+.+.}:
[<ffffffff8026134a>] __lock_acquire+0x12fe/0x1668 [<ffffffff802617a2>] lock_acquire+0xee/0x112 [<ffffffff804fe0a6>] mutex_lock_nested+0x4f/0x305 [<ffffffff8026709e>] module_update_tracepoints+0x1c/0x73 [<ffffffff80274f74>] tracepoint_update_probes+0x21/0x23 [<ffffffff8027510a>] tracepoint_probe_register+0x4a/0x68 [<ffffffff80277cae>] register_ftrace_graph+0x2a9/0x30f [<ffffffff8027f928>] trace_selftest_startup_function_graph+0x2e/0xbb [<ffffffff8027fdc5>] register_tracer+0x151/0x26e [<ffffffff8070b942>] init_graph_trace+0x10/0x12 [<ffffffff80209066>] do_one_initcall+0x5b/0x135 [<ffffffff806fa5e4>] kernel_init+0x12f/0x185 [<ffffffff8020c2ba>] child_rip+0xa/0x20[<ffffffffffffffff>]
0xffffffffffffffff-> #0 (ftrace_lock){+.+.+.}:
[<ffffffff80261067>] __lock_acquire+0x101b/0x1668 [<ffffffff802617a2>] lock_acquire+0xee/0x112 [<ffffffff804fe0a6>] mutex_lock_nested+0x4f/0x305 [<ffffffff80277e71>] ftrace_convert_nops+0x32/0x267 [<ffffffff802780bd>] ftrace_init_module+0x17/0x19 [<ffffffff8026a011>] load_module+0x1122/0x160b [<ffffffff8026a556>] sys_init_module+0x5c/0x1d3 [<ffffffff8020b21b>] system_call_fastpath+0x16/0x1b[<ffffffffffffffff>]
0xffffffffffffffff
Quoted 4 lines
Quoted 2 lines
Ah that's this ftrace/modules lock dependency that Steven told me about recently. Steven, do you know more about it now?
Yeah, I think I submitted a patch to fix that. I'll check my repo.
This is still live in 2.6.30 proper.
Let me know if you'd like further testing on your patch, or if you can't find locate it.
Thanks,
Daniel
--
Daniel J Blueman