On Mon, Dec 20, 2021 at 7:05 AM Sebastian Andrzej Siewior
<bigeasy@xxxxxxxxxxxxx> wrote:
>
> On 2021-12-17 11:16:56 [-0300], Wander Lairson Costa wrote:
>
> Assuming neither Steven nor Jens object,
>
> > TRACE_EVENT disables preemption before calling the callback. Because of
> > that blktrace triggers the following bug under PREEMPT_RT:
>
> The tracepoint is invoked with disabled preemption.
>
> > BUG: sleeping function called from invalid context at kernel/locking/spinlock_rt.c:35
> > in_atomic(): 1, irqs_disabled(): 0, non_block: 0, pid: 119, name: kworker/u2:2
> > 5 locks held by kworker/u2:2/119:
> > #0: ffff8c2e4a88f538 ((wq_completion)xfs-cil/dm-0){+.+.}-{0:0}, at: process_one_work+0x200/0x450
> > #1: ffffab3840ac7e68 ((work_completion)(&cil->xc_push_work)){+.+.}-{0:0}, at: process_one_work+0x200/0x450
> > #2: ffff8c2e4a887128 (&cil->xc_ctx_lock){++++}-{3:3}, at: xlog_cil_push_work+0xb7/0x670 [xfs]
> > #3: ffffffffa6a63780 (rcu_read_lock){....}-{1:2}, at: blk_add_trace_bio+0x0/0x1f0
> > #4: ffffffffa6610620 (running_trace_lock){+.+.}-{2:2}, at: __blk_add_trace+0x3ef/0x480
> > Preemption disabled at:
> > [<ffffffffa4d35c05>] migrate_enable+0x45/0x140
> > CPU: 0 PID: 119 Comm: kworker/u2:2 Kdump: loaded Not tainted 5.14.0-25.rt21.25.light.el9.x86_64+debug #1
> > Hardware name: Red Hat KVM, BIOS 0.5.1 01/01/2011
> > Workqueue: xfs-cil/dm-0 xlog_cil_push_work [xfs]
> > Call Trace:
> > ? migrate_enable+0x45/0x140
> > dump_stack_lvl+0x57/0x7d
> > ___might_sleep.cold+0xe3/0xf7
> > rt_spin_lock+0x3a/0xd0
> > ? __blk_add_trace+0x3ef/0x480
> > __blk_add_trace+0x3ef/0x480
> > blk_add_trace_bio+0x18d/0x1f0
> > trace_block_bio_queue+0xb5/0x150
> > submit_bio_checks+0x1f0/0x520
> > ? sched_clock_cpu+0xb/0x100
> > submit_bio_noacct+0x30/0x1d0
> > ? bio_associate_blkg+0x66/0x190
> > xlog_cil_push_work+0x1b6/0x670 [xfs]
> > ? register_lock_class+0x43/0x4f0
> > ? xfs_swap_extents+0x5f0/0x5f0 [xfs]
> > process_one_work+0x275/0x450
> > ? process_one_work+0x200/0x450
> > worker_thread+0x55/0x3c0
> > ? process_one_work+0x450/0x450
> > kthread+0x188/0x1a0
> > ? set_kthread_struct+0x40/0x40
> > ret_from_fork+0x22/0x30
>
> The above fills 90% of my screen with _no_ additional information. What
> about:
>
> The running_trace_lock protects running_trace_list and is acquired
> within the tracepoint which implies disabled preemption. The spinlock_t
> typed lock can not be acquired with disabled preemption on PREEMPT_RT
> because it becomes a sleeping lock.
> The runtime of the tracepoint depends on the number of entries in
> running_trace_list and has no limit. The blk-tracer is considered debug
> code and higher latencies here are okay.
>
> Make running_trace_lock a raw_spinlock_t
>
> > To avoid this bug, we switch the trace lock to a raw spinlock.
>
> Basically I want to give rationale _why_ changing a lock to
> raw_spinlock_t _here_ is okay. I want to avoid that people slap a
> s/spinlock_t/raw_spinlock_t/ each time they see warning of this kind.
>
Thanks, this sounds great. I am going to send v4 with the modified
commit message right now.
> > Signed-off-by: Wander Lairson Costa <wander@xxxxxxxxxx>
>
> Sebastian
>
