Re: [kprobes/x86] a19b2e3d78: WARNING:at_kernel/locking/lockdep.c:#trace_hardirqs_off_caller

From: Masami Hiramatsu
Date: Mon Oct 02 2017 - 22:26:37 EST

On Mon, 2 Oct 2017 09:19:10 -0700
Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:

> On Mon, Oct 2, 2017 at 8:46 AM, Masami Hiramatsu <mhiramat@xxxxxxxxxx> wrote:
> >
> > I'm considering to remove disabling-irq itself from jprobe.
> > (Frankly to say, I would like to remove jprobe itself...)
> Please please please...
> That would be lovely. The jprobe thing is really nasty, and despite
> the thing having been around forever (looking at history, it does back
> to 2004) there are very few users and they all look dubious to me.
> I seriously doubt anybody uses them, and I suspect our current tracing
> infrastructure is just *so* much better and more powerful than jprobes
> was.

I completely agree. Moreover, jprobe can not handle the functions
which is optimized and modified function type by compiler nowadays.

> So I'd heartily recommend just getting rid of jprobes. Or at least
> trying, and seeing if anybody actually even notices (and then
> reverting the removal and looking at what the usage ends up actually
> being).

OK, should I just make a series to remove jprobes and its few users,
or mark APIs obsolete and remove it after next version?

Thank you,

> Linus

Masami Hiramatsu <mhiramat@xxxxxxxxxx>