Re: [PATCH 2/4] locking: Apply contention tracepoints in the slow path

From: Steven Rostedt
Date: Tue Mar 01 2022 - 09:45:39 EST


On Tue, 1 Mar 2022 10:03:54 +0100
Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:

> diff --git a/kernel/locking/rtmutex.c b/kernel/locking/rtmutex.c
> index 8555c4efe97c..18b9f4bf6f34 100644
> --- a/kernel/locking/rtmutex.c
> +++ b/kernel/locking/rtmutex.c
> @@ -1579,6 +1579,8 @@ static int __sched __rt_mutex_slowlock(struct rt_mutex_base *lock,
>
> set_current_state(state);
>
> + trace_contention_begin(lock, _RET_IP_, LCB_F_RT);

I guess one issue with this is that _RET_IP_ will return the rt_mutex
address if this is not inlined, making the _RET_IP_ rather useless.

Now, if we can pass the _RET_IP_ into __rt_mutex_slowlock() then we could
reference that.

-- Steve


> +
> ret = task_blocks_on_rt_mutex(lock, waiter, current, ww_ctx, chwalk);
> if (likely(!ret))
> ret = rt_mutex_slowlock_block(lock, ww_ctx, state, NULL, waiter);
> @@ -1601,6 +1603,9 @@ static int __sched __rt_mutex_slowlock(struct rt_mutex_base *lock,
> * unconditionally. We might have to fix that up.
> */
> fixup_rt_mutex_waiters(lock);
> +
> + trace_contention_end(lock, ret);
> +
> return ret;
> }