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;
From:   Thomas Gleixner <tglx@linutronix.de>
To:     Jens Axboe <axboe@kernel.dk>, linux-kernel@vger.kernel.org,
        io-uring@vger.kernel.org
Cc:     peterz@infradead.org, oleg@redhat.com, Jens Axboe <axboe@kernel.dk>
Subject: Re: [PATCH 3/4] kernel: add support for TIF_NOTIFY_SIGNAL
In-Reply-To: <20201008152752.218889-4-axboe@kernel.dk>
References: <20201008152752.218889-1-axboe@kernel.dk> <20201008152752.218889-4-axboe@kernel.dk>
Date:   Wed, 14 Oct 2020 01:42:53 +0200
Message-ID: <878sc9d75u.fsf@nanos.tec.linutronix.de>
MIME-Version: 1.0
Content-Type: text/plain
Precedence: bulk

On Thu, Oct 08 2020 at 09:27, Jens Axboe wrote:
> This adds TIF_NOTIFY_SIGNAL handling in the generic code, which if set,
> will return true if signal_pending() is used in a wait loop. That causes
> an exit of the loop so that notify_signal tracehooks can be run. If the
> wait loop is currently inside a system call, the system call is restarted
> once task_work has been processed.
>
> x86 is using the generic entry code, add the necessary TIF_NOTIFY_SIGNAL
> definitions for it.

Can you please split that into core changes and a patch which adds
support for x86?

>  static inline int signal_pending(struct task_struct *p)
>  {
> +#ifdef TIF_NOTIFY_SIGNAL

As I said in the other thread, plase make this

#if defined(CONFIG_GENERIC_ENTRY) && defined(TIF_NOTIFY_SIGNAL)

> +/*
> + * called by exit_to_user_mode_loop() if ti_work & _TIF_NOTIFY_SIGNAL. This
> + * is currently used by TWA_SIGNAL based task_work, which requires breaking
> + * wait loops to ensure that task_work is noticed and run.
> + */
> +static inline void tracehook_notify_signal(void)
> +{
> +#ifdef TIF_NOTIFY_SIGNAL

Ditto.

> +	clear_thread_flag(TIF_NOTIFY_SIGNAL);
> +	smp_mb__after_atomic();
> +	if (current->task_works)
> +		task_work_run();
> +#endif
> +}
> +
> +/*
> + * Called when we have work to process from exit_to_user_mode_loop()
> + */
> +static inline void set_notify_signal(struct task_struct *task)
> +{
> +#ifdef TIF_NOTIFY_SIGNAL

And this one.

Other than that, this approach of using arch_do_signal() addresses my
earlier concerns about the syscall restart machinery.

Thanks,

        tglx