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;
MIME-Version: 1.0
References: <20220809074522.2444672-1-chenhuacai@loongson.cn> <874jylx0ad.wl-maz@kernel.org>
In-Reply-To: <874jylx0ad.wl-maz@kernel.org>
From:   Huacai Chen <chenhuacai@kernel.org>
Date:   Tue, 9 Aug 2022 17:19:31 +0800
Message-ID: <CAAhV-H50HERxjrmfDkSaHAVY7Q_ufpZP-e+qdZWVd=qEuQVXBg@mail.gmail.com>
Subject: Re: [PATCH] LoongArch: Don't disable EIOINTC master core
To:     Marc Zyngier <maz@kernel.org>
Cc:     Huacai Chen <chenhuacai@loongson.cn>,
        Thomas Gleixner <tglx@linutronix.de>,
        linux-arch <linux-arch@vger.kernel.org>,
        LKML <linux-kernel@vger.kernel.org>,
        Xuerui Wang <kernel@xen0n.name>,
        Xuefeng Li <lixuefeng@loongson.cn>,
        Jiaxun Yang <jiaxun.yang@flygoat.com>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

Hi, Marc,

On Tue, Aug 9, 2022 at 4:56 PM Marc Zyngier <maz@kernel.org> wrote:
>
> On Tue, 09 Aug 2022 08:45:22 +0100,
> Huacai Chen <chenhuacai@loongson.cn> wrote:
> >
> > This patch fix a CPU hotplug issue. The EIOINTC master core (the first
> > core of an EIOINTC node) should not be disabled at runtime, since it has
> > the responsibility of dispatching I/O interrupts.
> >
> > Signed-off-by: Huacai Chen <chenhuacai@loongson.cn>
> > ---
> >  arch/loongarch/kernel/smp.c            | 9 +++++++++
> >  drivers/irqchip/irq-loongson-eiointc.c | 5 +++++
> >  2 files changed, 14 insertions(+)
> >
> > diff --git a/arch/loongarch/kernel/smp.c b/arch/loongarch/kernel/smp.c
> > index 09743103d9b3..54901716f8de 100644
> > --- a/arch/loongarch/kernel/smp.c
> > +++ b/arch/loongarch/kernel/smp.c
> > @@ -242,9 +242,18 @@ void loongson3_smp_finish(void)
> >
> >  static bool io_master(int cpu)
> >  {
> > +     int i, node, master;
> > +
> >       if (cpu == 0)
> >               return true;
> >
> > +     for (i = 1; i < loongson_sysconf.nr_io_pics; i++) {
> > +             node = eiointc_get_node(i);
> > +             master = cpu_number_map(node * CORES_PER_EIO_NODE);
> > +             if (cpu == master)
> > +                     return true;
> > +     }
> > +
> >       return false;
> >  }
> >
> > diff --git a/drivers/irqchip/irq-loongson-eiointc.c b/drivers/irqchip/irq-loongson-eiointc.c
> > index 170dbc96c7d3..6c99a2ff95f5 100644
> > --- a/drivers/irqchip/irq-loongson-eiointc.c
> > +++ b/drivers/irqchip/irq-loongson-eiointc.c
> > @@ -56,6 +56,11 @@ static void eiointc_enable(void)
> >       iocsr_write64(misc, LOONGARCH_IOCSR_MISC_FUNC);
> >  }
> >
> > +int eiointc_get_node(int id)
> > +{
> > +     return eiointc_priv[id]->node;
> > +}
> > +
> >  static int cpu_to_eio_node(int cpu)
> >  {
> >       return cpu_logical_map(cpu) / CORES_PER_EIO_NODE;
>
>
> I don't understand why it has to be this complex and make any use of
> the node number.
>
> As I understand it, CPU-0 in any EIOINTC block is a master. So all you
> need to find out is whether the CPU number is a multiple of
> CORES_PER_EIO_NODE.
CPU-0 in any EIOINTC block may be a master, but not absolutely be a
master to dispatch I/O interrupts. If there is no bridge under a
EIOINTC, then this EIOINTC doesn't handle I/O interrupts, and it can
be disabled at runtime.

Huacai
>
> Something like:
>
>         return (cpu_logical_map(cpu) % CORES_PER_EIO_NODE) == 0;
>
> which doesn't require exporting anything from anywhere?
>
>         M.
>
> --
> Without deviation from the norm, progress is not possible.