Re: [PATCH] trace/pid_list: optimize pid_list->lock contention

[Yongliang Gao]

Hi Steven, Sebastian,

Thank you for your review and explanation.
You are absolutely right. I overlooked the case of PREEMPT_RT. I will
look into other possible optimization methods later.

Best regards,
Yongliang

On Tue, Nov 11, 2025 at 4:13 PM Sebastian Andrzej Siewior
<bigeasy@xxxxxxxxxxxxx> wrote:
>
> On 2025-11-10 18:38:54 [-0500], Steven Rostedt wrote:
> > On Wed, 15 Oct 2025 19:49:52 +0800
> > Yongliang Gao <leonylgao@xxxxxxxxx> wrote:
> >
> > > diff --git a/kernel/trace/pid_list.c b/kernel/trace/pid_list.c
> > > index 090bb5ea4a19..62082a4f60db 100644
> > > --- a/kernel/trace/pid_list.c
> > > +++ b/kernel/trace/pid_list.c
> > > @@ -138,14 +138,14 @@ bool trace_pid_list_is_set(struct trace_pid_list *pid_list, unsigned int pid)
> > >     if (pid_split(pid, &upper1, &upper2, &lower) < 0)
> > >             return false;
> > >
> > > -   raw_spin_lock_irqsave(&pid_list->lock, flags);
> > > +   read_lock_irqsave(&pid_list->lock, flags);
> > >     upper_chunk = pid_list->upper[upper1];
> > >     if (upper_chunk) {
> > >             lower_chunk = upper_chunk->data[upper2];
> > >             if (lower_chunk)
> > >                     ret = test_bit(lower, lower_chunk->data);
> > >     }
> > > -   raw_spin_unlock_irqrestore(&pid_list->lock, flags);
> > > +   read_unlock_irqrestore(&pid_list->lock, flags);
> > >
> > >     return ret;
> > >  }
> >
> > Unfortunately you cannot do this because this is called while holding the
> > task pi_lock and rq locks. In PREEMPT_RT() the read/write_lock_* turn into
> > mutexes.
> >
> > Sebastian, is there any equivalent of raw_read/write_locks() that can be
> > used?
>
> Nope, no read-write lock that can be used in atomic sections. Well,
> there is RCU.
>
> > -- Steve
>
> Sebastian