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:   Fri, 14 Jul 2023 14:35:12 +0100
Message-ID: <86cz0uvcof.wl-maz@kernel.org>
From:   Marc Zyngier <maz@kernel.org>
To:     Anup Patel <apatel@ventanamicro.com>
Cc:     Saravana Kannan <saravanak@google.com>,
        Palmer Dabbelt <palmer@dabbelt.com>,
        Paul Walmsley <paul.walmsley@sifive.com>,
        Thomas Gleixner <tglx@linutronix.de>,
        Rob Herring <robh+dt@kernel.org>,
        Krzysztof Kozlowski <krzysztof.kozlowski+dt@linaro.org>,
        Atish Patra <atishp@atishpatra.org>,
        Andrew Jones <ajones@ventanamicro.com>,
        Sunil V L <sunilvl@ventanamicro.com>,
        Conor Dooley <conor@kernel.org>,
        Anup Patel <anup@brainfault.org>,
        linux-riscv@lists.infradead.org, linux-kernel@vger.kernel.org,
        devicetree@vger.kernel.org
Subject: Re: [PATCH v5 7/9] irqchip: Add RISC-V advanced PLIC driver
In-Reply-To: <CAK9=C2ULxsXednwnoyzYKjQFpe_zBSZ4v0FqUYFnxK2TpbiMtg@mail.gmail.com>
References: <20230710094321.1378351-1-apatel@ventanamicro.com>
        <20230710094321.1378351-8-apatel@ventanamicro.com>
        <CAGETcx8kH8cJVdhcv5K4qNUo58godFZEBnOfTGKUUQ6VuUguvQ@mail.gmail.com>
        <86jzv2vpdb.wl-maz@kernel.org>
        <CAK9=C2ULxsXednwnoyzYKjQFpe_zBSZ4v0FqUYFnxK2TpbiMtg@mail.gmail.com>
User-Agent: Wanderlust/2.15.9 (Almost Unreal) SEMI-EPG/1.14.7 (Harue)
 FLIM-LB/1.14.9 (=?UTF-8?B?R29qxY0=?=) APEL-LB/10.8 EasyPG/1.0.0 Emacs/28.2
 (aarch64-unknown-linux-gnu) MULE/6.0 (HANACHIRUSATO)
MIME-Version: 1.0 (generated by SEMI-EPG 1.14.7 - "Harue")
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

On Fri, 14 Jul 2023 10:35:34 +0100,
Anup Patel <apatel@ventanamicro.com> wrote:
>=20
> On Fri, Jul 14, 2023 at 2:31=E2=80=AFPM Marc Zyngier <maz@kernel.org> wro=
te:
> >
> > Anup,
> >
> > On Fri, 14 Jul 2023 00:56:22 +0100,
> > Saravana Kannan <saravanak@google.com> wrote:
> > >
> > > On Mon, Jul 10, 2023 at 2:44=E2=80=AFAM Anup Patel <apatel@ventanamic=
ro.com> wrote:
> > > >
> > > > The RISC-V advanced interrupt architecture (AIA) specification defi=
nes
> > > > a new interrupt controller for managing wired interrupts on a RISC-V
> > > > platform. This new interrupt controller is referred to as advanced
> > > > platform-level interrupt controller (APLIC) which can forward wired
> > > > interrupts to CPUs (or HARTs) as local interrupts OR as message
> > > > signaled interrupts.
> > > > (For more details refer https://github.com/riscv/riscv-aia)
> > > >
> > > > This patch adds an irqchip driver for RISC-V APLIC found on RISC-V
> > > > platforms.
> > > >
> > > > Signed-off-by: Anup Patel <apatel@ventanamicro.com>
> >
> > [...]
> >
> > > > +static int __init aplic_dt_init(struct device_node *node,
> > > > +                               struct device_node *parent)
> > > > +{
> > > > +       /*
> > > > +        * The APLIC platform driver needs to be probed early
> > > > +        * so for device tree:
> > > > +        *
> > > > +        * 1) Set the FWNODE_FLAG_BEST_EFFORT flag in fwnode which
> > > > +        *    provides a hint to the device driver core to probe the
> > > > +        *    platform driver early.
> > > > +        * 2) Clear the OF_POPULATED flag in device_node because
> > > > +        *    of_irq_init() sets it which prevents creation of
> > > > +        *    platform device.
> > > > +        */
> > > > +       node->fwnode.flags |=3D FWNODE_FLAG_BEST_EFFORT;
> > >
> > > Please stop spamming us with broken patches. Already told you this is
> > > not an option.
> > >
> > > Nack.
> >
> > What puzzles me here is that *no other arch* requires this sort of
> > hack. What is so special about the APLIC that it requires it? I see
> > nothing in this patch that even hints at it, despite the "discussion"
> > in the last round.
> >
> > The rules are simple:
> >
> > - either the APLIC is so fundamental to the system that it has to be
> >   initialised super early, much like the GIC on arm64, at which point
> >   it cannot be a platform device, and the story is pretty simple.
> >
> > - or it isn't that fundamental, and it can be probed as a platform
> >   device using the dependency infrastructure that is already used by
> >   multiple other interrupt controller drivers, without any need to
> >   mess with internal flags. Again, this should be simple enough.
>=20
> APLIC manages all wired interrupts whereas IMSIC manages all
> MSIs. Both APLIC and IMSIC are fundamental devices which need
> to be probed super early.
>=20
> Now APLIC has two modes of operations:
> 1) Direct mode where there is no IMSIC in the system and APLIC
>     directly injects interrupt to CPUs
> 2) MSI mode where IMSIC is present in the system and APLIC
>     converts wired interrupts into MSIs
>=20
> The APLIC driver added by this patch is a common driver for
> both above modes.

Which it doesn't need to be. You are pointlessly making life difficult
for yourself, and everyone else. The MSI bridge behaviour has *zero*
reason to be the same driver as the main "I need it super early"
driver. They may be called the same, but they *are* different things
in the system.

They can share code, but they are fundamentally a different thing in
the system. And I guess this silly approach has other ramifications:
the IMSIC is also some early driver when it really doesn't need to be.
Who needs MSIs that early in the life of the system? I don't buy this
for even a second.

Frankly, this whole thing needs to be taken apart and rebuilt from the
ground up.

> For #2, APLIC needs to be a platform device to create a device
> MSI domain using platform_msi_create_device_domain() which
> is why the APLIC driver is a platform driver.

You can't have your cake and eat it. If needed super early, and it
cannot be a platform driver. End of the story.

And to my earlier point: IMSIC and APLIC-as-MSI-bridge have no purpose
being early drivers. They must be platform drivers, and only that.

> > If these rules don't apply to your stuff, please explain what is so
> > different. And I mean actually explain the issue. Which isn't telling
> > us "it doesn't work without it". Because as things stand, there is no
> > way I will even consider taking this ugly mix of probing methods.
>=20
> Yes, I don't want this ugly FWNODE_FLAG_BEST_EFFORT hack
> in this driver.

And yet you are hammering it even when told this is wrong.

> I tried several things but setting the FWNODE_FLAG_BEST_EFFORT
> flag is the only thing which works right now.

How about you take a step back and realise that the way you've
architected your drivers makes little sense? I don't think you have
tried *that*.

Thanks,

	M.

--=20
Without deviation from the norm, progress is not possible.