Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Date:   Wed, 24 Apr 2019 22:45:54 +0200 (CEST)
From:   Thomas Gleixner <tglx@linutronix.de>
To:     Daniel Drake <drake@endlessm.com>
cc:     mingo@redhat.com, bp@alien8.de, hpa@zytor.com, x86@kernel.org,
        linux-kernel@vger.kernel.org, len.brown@intel.com,
        rafael.j.wysocki@intel.com, linux@endlessm.com, hdegoede@redhat.com
Subject: Re: [PATCH 1/2] x86/time: check usability of IRQ0 PIT timer
In-Reply-To: <20190423050354.8025-1-drake@endlessm.com>
Message-ID: <alpine.DEB.2.21.1904242114380.1762@nanos.tec.linutronix.de>
References: <20190423050354.8025-1-drake@endlessm.com>
User-Agent: Alpine 2.21 (DEB 202 2017-01-01)
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Tue, 23 Apr 2019, Daniel Drake wrote:
>  /* Default timer init function */
>  void __init hpet_time_init(void)
>  {
> -	if (!hpet_enable())
> -		setup_pit_timer();
> +	if (hpet_enable()) {
> +		setup_default_timer_irq();
> +		return;
> +	}
> +
> +	/* Fall back on legacy 8253 PIT */
> +	setup_pit_timer();
>  	setup_default_timer_irq();
> +
> +	/*
> +	 * Intel SoCs like ApolloLake, Skylake and newer can have
> +	 * their PIT "gated" by the BIOS such that IRQ0 does not
> +	 * tick. Check for that situation here.
> +	 */
> +	if (!irq0_timer_works()) {

So you rely on the fact that the legacy PIC delivery is working
here. That's fragile at best especially when this is a boot of a crash
kernel.

> +		pr_info("HPET is not available, and 8253 timer is not working. "
> +			"Continuing without IRQ0 timer.\n");
> +		remove_default_timer_irq();
> +		remove_pit_timer();
> +	}
> +
> +void __init clockevent_i8253_exit(void)
> +{
> +	clockevents_switch_state(&i8253_clockevent, CLOCK_EVT_STATE_DETACHED);

Smart, but broken. The PIT clockevent is still referenced in the
clockevents core code after the unbind. The unbind logic is there for a
reason and just because the above duct tape does not explode in your face
does not make it more correct. :)

> +	clockevents_unbind_device(&i8253_clockevent, smp_processor_id());

We really want to avoid the whole register and whatever dance at all for
these devices. Let me think about it some more.

Thanks,

	tglx