Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.36 as permitted sender) client-ip=23.128.96.36;
Date:   Wed, 29 Nov 2023 10:56:00 +0200
From:   Mika Westerberg <mika.westerberg@linux.intel.com>
To:     "Rafael J. Wysocki" <rjw@rjwysocki.net>
Cc:     Linux ACPI <linux-acpi@vger.kernel.org>,
        LKML <linux-kernel@vger.kernel.org>,
        Zhang Rui <rui.zhang@intel.com>,
        Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>,
        Michal Wilczynski <michal.wilczynski@intel.com>,
        Hans de Goede <hdegoede@redhat.com>,
        Andy Shevchenko <andriy.shevchenko@linux.intel.com>,
        Heikki Krogerus <heikki.krogerus@linux.intel.com>,
        Mario Limonciello <mario.limonciello@amd.com>
Subject: Re: [RFT][PATCH v1] ACPI: OSL: Use a threaded interrupt handler for
 SCI
Message-ID: <20231129085600.GQ1074920@black.fi.intel.com>
References: <5745568.DvuYhMxLoT@kreacher>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
In-Reply-To: <5745568.DvuYhMxLoT@kreacher>
Precedence: bulk

Hi Rafael,

On Mon, Nov 27, 2023 at 08:57:43PM +0100, Rafael J. Wysocki wrote:
> From: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
> 
> In the current arrangement, all of the acpi_ev_sci_xrupt_handler() code
> is run as an interrupt handler for the SCI, in interrupt context.  Among
> other things, this causes it to run with local interrupts off which
> can be problematic if many GPEs are enabled and they are located in the
> I/O address space, for example (because in that case local interrupts
> will be off for the duration of all of the GPE hardware accesses carried
> out while handling an SCI combined and that may be quite a bit of time
> in extreme scenarios).
> 
> However, there is no particular reason why the code in question really
> needs to run in interrupt context and in particular, it has no specific
> reason to run with local interrupts off.  The only real requirement is
> to prevent multiple instences of it from running in parallel with each
> other, but that can be achieved regardless.
> 
> For this reason, use request_threaded_irq() instead of request_irq() for
> the ACPI SCI and pass IRQF_ONESHOT to it in flags to indicate that the
> interrupt needs to be masked while its handling thread is running so as
> to prevent it from re-triggering while it is being handled (and in
> particular until the final handled/not handled outcome is determined).
> 
> While at it, drop a redundant local variable from acpi_irq().
> 
> Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
> ---
> 
> The code inspection and (necessarily limited) testing carried out by me
> are good indications that this should just always work, but there may
> be still some really odd platform configurations I'm overlooking, so if
> you have a way to give it a go, please do so.

Tried this on ADL-S and ADL-P systems that I have here and both work
just fine with the patch applied. I can see SCI interrupt count
increases in /proc/interrupts as expected. Did a couple of s2idle cycles
too, all good.

Tested-by: Mika Westerberg <mika.westerberg@linux.intel.com>