Re: [PATCH v2] kmemleak: Turn kmemleak_lock to raw spinlock on RT

[Sebastian Andrzej Siewior]

On 2018-11-22 17:04:19 [+0800], zhe.he@xxxxxxxxxxxxx wrote:
> From: He Zhe <zhe.he@xxxxxxxxxxxxx>
> 
> kmemleak_lock, as a rwlock on RT, can possibly be held in atomic context and
> causes the follow BUG.
> 
> BUG: scheduling while atomic: migration/15/132/0x00000002
â
> Preemption disabled at:
> [<ffffffff8c927c11>] cpu_stopper_thread+0x71/0x100
> CPU: 15 PID: 132 Comm: migration/15 Not tainted 4.19.0-rt1-preempt-rt #1
> Hardware name: Intel Corp. Harcuvar/Server, BIOS HAVLCRB1.X64.0015.D62.1708310404 08/31/2017
> Call Trace:
>  dump_stack+0x4f/0x6a
>  ? cpu_stopper_thread+0x71/0x100
>  __schedule_bug.cold.16+0x38/0x55
>  __schedule+0x484/0x6c0
>  schedule+0x3d/0xe0
>  rt_spin_lock_slowlock_locked+0x118/0x2a0
>  rt_spin_lock_slowlock+0x57/0x90
>  __rt_spin_lock+0x26/0x30
>  __write_rt_lock+0x23/0x1a0
>  ? intel_pmu_cpu_dying+0x67/0x70
>  rt_write_lock+0x2a/0x30
>  find_and_remove_object+0x1e/0x80
>  delete_object_full+0x10/0x20
>  kmemleak_free+0x32/0x50
>  kfree+0x104/0x1f0
>  ? x86_pmu_starting_cpu+0x30/0x30
>  intel_pmu_cpu_dying+0x67/0x70
>  x86_pmu_dying_cpu+0x1a/0x30
>  cpuhp_invoke_callback+0x92/0x700
>  take_cpu_down+0x70/0xa0
>  multi_cpu_stop+0x62/0xc0
>  ? cpu_stop_queue_work+0x130/0x130
>  cpu_stopper_thread+0x79/0x100
>  smpboot_thread_fn+0x20f/0x2d0
>  kthread+0x121/0x140
>  ? sort_range+0x30/0x30
>  ? kthread_park+0x90/0x90
>  ret_from_fork+0x35/0x40

If this is the only problem? kfree() from a preempt-disabled section
should cause a warning even without kmemleak.

> And on v4.18 stable tree the following call trace, caused by grabbing
> kmemleak_lock again, is also observed.
> 
> kernel BUG at kernel/locking/rtmutex.c:1048! 
> invalid opcode: 0000 [#1] PREEMPT SMP PTI 
> CPU: 5 PID: 689 Comm: mkfs.ext4 Not tainted 4.18.16-rt9-preempt-rt #1 
â
> Call Trace: 
>  ? preempt_count_add+0x74/0xc0 
>  rt_spin_lock_slowlock+0x57/0x90 
>  ? __kernel_text_address+0x12/0x40 
>  ? __save_stack_trace+0x75/0x100 
>  __rt_spin_lock+0x26/0x30 
>  __write_rt_lock+0x23/0x1a0 
>  rt_write_lock+0x2a/0x30 
>  create_object+0x17d/0x2b0 
â

is this an RT-only problem? Because mainline should not allow read->read
locking or read->write locking for reader-writer locks. If this only
happens on v4.18 and not on v4.19 then something must have fixed it.
 

Sebastian