Re: [PATCH RESEND 2/2] tracing: syscall_regfunc() should not skip kernel threads

From: Oleg Nesterov <oleg@redhat.com>
To: Steven Rostedt <rostedt@goodmis.org>
Cc: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>,
	Frederic Weisbecker <fweisbec@gmail.com>,
	LKML <linux-kernel@vger.kernel.org>,
	Andrew Morton <akpm@linux-foundation.org>,
	Ingo Molnar <mingo@kernel.org>,
	Hendrik Brueckner <brueckner@linux.vnet.ibm.com>
Subject: Re: [PATCH RESEND 2/2] tracing: syscall_regfunc() should not skip kernel threads
Date: Fri, 11 Apr 2014 14:03:47 +0200	[thread overview]
Message-ID: <20140411120347.GA19753@redhat.com> (raw)
In-Reply-To: <20140410155554.44f88536@gandalf.local.home>

On 04/10, Steven Rostedt wrote:
>
> On Thu, 10 Apr 2014 21:38:20 +0200
> Oleg Nesterov <oleg@redhat.com> wrote:
>
> > I _think_ that the kernel thread can only return from "call *%rbx" if
> > it is no longer a kernel thread, iow, do_execve() was called.
>
> Ah right. But only in special cases.
>
> Actually, it only returns if the function in kernel_thread() returns,
> and in the case of ____call_usermodehelper, it does an exec and
> returns on success. But if it fails, it needs to call do_exit().

Yes. If it simply returns the kernel will crash, without start_thread/etc
"int_ret_from_sys_call" can do nothing good.

So we can conclude that "there is only one case when a kernel thread can
reach the usual syscall exit tracing path:" from cc3b13c11c567c69 is no
longer valid and we can kille the PF_KTHREAD check in regfuncs.

> Thus, it's not sufficient to just say "only if do_execve is called" but
> to say, if the handler that is called is only allowed to return iff it
> did a execve first and succeeded.

I do remember, I already told you are pedant!

Oleg.