Re: [PATCH] signal/x86: Delay calling signals in atomic

From: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
To: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: x86@kernel.org, linux-kernel@vger.kernel.org,
	Oleg Nesterov <oleg@redhat.com>, "H. Peter Anvin" <hpa@zytor.com>,
	Andy Lutomirski <luto@kernel.org>,
	Ben Segall <bsegall@google.com>, Borislav Petkov <bp@alien8.de>,
	Daniel Bristot de Oliveira <bristot@redhat.com>,
	Dave Hansen <dave.hansen@linux.intel.com>,
	Dietmar Eggemann <dietmar.eggemann@arm.com>,
	Ingo Molnar <mingo@redhat.com>,
	Juri Lelli <juri.lelli@redhat.com>, Mel Gorman <mgorman@suse.de>,
	Peter Zijlstra <peterz@infradead.org>,
	Steven Rostedt <rostedt@goodmis.org>,
	Thomas Gleixner <tglx@linutronix.de>,
	Vincent Guittot <vincent.guittot@linaro.org>
Subject: Re: [PATCH] signal/x86: Delay calling signals in atomic
Date: Mon, 28 Mar 2022 18:17:38 +0200	[thread overview]
Message-ID: <YkHfokLO1UGCISGc@linutronix.de> (raw)
In-Reply-To: <8735j2xigt.fsf@email.froward.int.ebiederm.org>

On 2022-03-28 09:25:06 [-0500], Eric W. Biederman wrote:
> Sebastian Andrzej Siewior <bigeasy@linutronix.de> writes:
> 
> Folks I really would have appreciated being copied on a signal handling
> patch like this.

Sorry for that. For the whole ptrace/signal part is no maintainer listed
and I got the feeling that Oleg knows these bits.

> It is too late to nack, but I think this buggy patch deserved one.  Can
> we please fix PREEMPT_RT instead?

Sure.

> As far as I can tell this violates all of rules from
> implementing/maintaining the RT kernel.  Instead of coming up with new
> abstractions that makes sense and can use by everyone this introduces
> a hack only for PREEMPT_RT and a pretty horrible one at that.
>
> This talks about int3, but the code looks for in_atomic().  Which means
> that essentially every call of force_sig will take this path as they
> almost all come from exception handlers.  It is the nature of signals
> that report on faults.  An exception is raised and the kernel reports it
> to userspace with a fault signal (aka force_sig_xxx).

The int3 is invoked with disabled interrupts. There are also a few
others path which are explicit with disabled interrupts or with a
raw_spinlock_t which lead to an atomic section on PREEMPT_RT. Call
chains with spinlock_t or a rwlock_t don't lead to a atomic section on
PREEMPT_RT. Therefore I don't think this is "essentially every call of
force_sig" that is going to use that.

> Further this code is buggy.  TIF_NOTIFY_RESUME is not the correct
> flag to set to enter into exit_to_usermode_loop.  TIF_NOTIFY_RESUME is
> about that happens after signal handling.  This very much needs to be
> TIF_SIGPENDING with recalc_sigpending and friends updated to know about
> "task->force_info".
> 
> Does someone own this problem?  Can that person please fix this
> properly?

Sure. Instead setting TIF_NOTIFY_RESUME you want the code updated to use
recalc_sigpending() only. Or do you have other suggestions regarding
fixing this properly?

> I really don't think it is going to be maintainable for PREEMPT_RT to
> maintain a separate signal delivery path for faults from the rest of
> linux.

Okay.

> Eric

Sebastian