Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Fri, 16 Oct 2020 12:54:16 +0200
From:   Oleg Nesterov <oleg@redhat.com>
To:     Thomas Gleixner <tglx@linutronix.de>
Cc:     Jens Axboe <axboe@kernel.dk>, linux-kernel@vger.kernel.org,
        io-uring@vger.kernel.org, peterz@infradead.org
Subject: Re: [PATCH 4/5] x86: wire up TIF_NOTIFY_SIGNAL
Message-ID: <20201016105415.GA21989@redhat.com>
References: <20201015131701.511523-1-axboe@kernel.dk>
 <20201015131701.511523-5-axboe@kernel.dk>
 <87o8l3a8af.fsf@nanos.tec.linutronix.de>
 <20201015143409.GC24156@redhat.com>
 <87y2k6trzr.fsf@nanos.tec.linutronix.de>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <87y2k6trzr.fsf@nanos.tec.linutronix.de>
User-Agent: Mutt/1.5.24 (2015-08-30)
Precedence: bulk

On 10/16, Thomas Gleixner wrote:
>
> On Thu, Oct 15 2020 at 16:34, Oleg Nesterov wrote:
> > On 10/15, Thomas Gleixner wrote:
> >> Instead of adding this to every architectures signal magic, we can
> >> handle TIF_NOTIFY_SIGNAL in the core code:
> >>
> >> static void handle_singal_work(ti_work, regs)
> >> {
> >> 	if (ti_work & _TIF_NOTIFY_SIGNAL)
> >>         	tracehook_notify_signal();
> >>
> >>         arch_do_signal(ti_work, regs);
> >> }
> >>
> >>       loop {
> >>       		if (ti_work & (SIGPENDING | NOTIFY_SIGNAL))
> >>                 	handle_signal_work(ti_work, regs);
> >>       }
> >
> > To me this looks like unnecessary complication. We need to change
> > every architecture anyway, how can this helper help?
>
> This is about the generic entry code. For the users of that it makes
> absolutely no sense to have that in architecture code.
>
> Something which every architecture needs to do in the exactly same way
> goes into the common code. If not, you can spare the exercise of having
> common code in the first place.
>
> Also arch_do_signal() becomes a misnomer with this new magic.

Well, to me arch_do_signal() paths should handle the signal_pending() == T
case.

But I won't argue, this is subjective.

> static void handle_signal_work(ti_work, regs)
> {
> 	if (ti_work & _TIF_NOTIFY_SIGNAL)
>         	tracehook_notify_signal();
>
>         arch_do_signal_or_restart(ti_work, regs);
> }
>
> which makes it entirely clear what this is about.

In this case I'd prefer to pass the "(ti_work & _TIF_SIGPENDING)" boolen
to arch_do_signal_or_restart().

But again, I won't argue. And to remind, we do not really need to touch
arch_do_signal() at all. We can just add

	if (test_thread_flag(TIF_NOTIFY_SIGNAL))
		tracehook_notify_signal();

	if (!task_sigpending(current))
		return 0;

at the start of get_signal() and avoid the code duplication automatically.

Oleg.