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.
> Signed-off-by: Wander Lairson Costa <wander@xxxxxxxxxx>
Sebastian
