Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Thu, 22 Sep 2022 14:27:41 +0100
From:   Will Deacon <will@kernel.org>
To:     Pierre Gondois <pierre.gondois@arm.com>
Cc:     linux-kernel@vger.kernel.org,
        Dietmar Eggemann <dietmar.eggemann@arm.com>,
        Valentin Schneider <vschneid@redhat.com>,
        Thomas Gleixner <tglx@linutronix.de>,
        Sebastian Andrzej Siewior <bigeasy@linutronix.de>,
        Mark Rutland <mark.rutland@arm.com>,
        linux-arm-kernel@lists.infradead.org
Subject: Re: [PATCH v1 0/1] arm_pmu: acpi: Pre-allocate pmu structures
Message-ID: <20220922132740.GB12095@willie-the-truck>
References: <20220912155105.1443303-1-pierre.gondois@arm.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20220912155105.1443303-1-pierre.gondois@arm.com>
User-Agent: Mutt/1.10.1 (2018-07-13)
Precedence: bulk

On Mon, Sep 12, 2022 at 05:51:03PM +0200, Pierre Gondois wrote:
> Sleeping inside arm_pmu_acpi_cpu_starting() while running a preemp_rt
> kernel was reported at [1], and a solution was suggested at [2].
> 
> It seems [2] doesn't work because of the following:
> """PREPARE: The callbacks are invoked on a control CPU before the
> hotplugged CPU is started up or after the hotplugged CPU has died."""
> 
> Indeed the 'prepare' callbacks are executed on the control CPU and
> this CPU cannot remotely read the cpuid (cf read_cpuid_id()) of the
> other CPUs.
> 
> Another solution:
> 1. count the number of different PMU IRQs (#IRQs)
> 2. allocate #IRQs pmu structures. There is at most #IRQs PMUs
> 3. in arm_pmu_acpi_cpu_starting() subcalls, after checking the cpuid,
>    decide to use or re-use a pre-allocated pmu structure.
>    Thus the pre-allocated pmu struct can be seen as a pool and all
>    the struct might not be used.
> 4. free the unused pmu structures when probing
> is suggested in this patch.
> 
> Freeing the unused pmu structures (4.) could maybe be done in a
> CPUHP_AP_ONLINE_DYN hotplug callback instead, but only one CPU is
> needed to search/free unused pre-allocated pmu structures.

I vaguely remember Mark R having a look at this at the time, so I'd like
his Ack on your approach before I queue anything.

Will