Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Content-Type: text/plain;
        charset=utf-8
Mime-Version: 1.0 (Mac OS X Mail 14.0 \(3654.120.0.1.13\))
Subject: Re: [PATCH 1/1] dt-bindings: reg-io-width for SiFive CLINT
From:   Jessica Clarke <jrtc27@jrtc27.com>
In-Reply-To: <49af3dc8-0d1e-e82f-6df2-2bf8338acd3a@canonical.com>
Date:   Fri, 15 Oct 2021 13:15:28 +0100
Cc:     Bin Meng <bmeng.cn@gmail.com>,
        Daniel Lezcano <daniel.lezcano@linaro.org>,
        Thomas Gleixner <tglx@linutronix.de>,
        Guo Ren <guoren@linux.alibaba.com>, Xiang W <wxjstz@126.com>,
        Samuel Holland <samuel@sholland.org>,
        Atish Patra <atish.patra@wdc.com>,
        Rob Herring <robh+dt@kernel.org>,
        Palmer Dabbelt <palmer@dabbelt.com>,
        Paul Walmsley <paul.walmsley@sifive.com>,
        Anup Patel <anup.patel@wdc.com>,
        linux-kernel <linux-kernel@vger.kernel.org>,
        devicetree <devicetree@vger.kernel.org>,
        linux-riscv <linux-riscv@lists.infradead.org>,
        OpenSBI <opensbi@lists.infradead.org>
Content-Transfer-Encoding: quoted-printable
Message-Id: <D976B0F9-6AE4-4C04-8D52-75B24D6E4359@jrtc27.com>
References: <20211015100941.17621-1-heinrich.schuchardt@canonical.com>
 <CAEUhbmUSmi9mbiHpatBjGgRjrtyJPms9BRmyBPs5BWeVb7n1Nw@mail.gmail.com>
 <49af3dc8-0d1e-e82f-6df2-2bf8338acd3a@canonical.com>
To:     Heinrich Schuchardt <heinrich.schuchardt@canonical.com>
Precedence: bulk

On 15 Oct 2021, at 12:54, Heinrich Schuchardt =
<heinrich.schuchardt@canonical.com> wrote:
>=20
> On 10/15/21 12:14, Bin Meng wrote:
>> On Fri, Oct 15, 2021 at 6:09 PM Heinrich Schuchardt
>> <heinrich.schuchardt@canonical.com> wrote:
>>>=20
>>> The CLINT in the T-HEAD 9xx processors do not support 64bit mmio =
access to
>>> the MTIMER device. The current schema does not allow to specify =
this.
>>>=20
>>> OpenSBI currently uses a property 'clint,has-no-64bit-mmio' to =
indicate the
>>> restriction. Samuael Holland suggested in
>>> lib: utils/timer: Use standard property to specify 32-bit I/O
>>> =
https://github.com/smaeul/opensbi/commit/b95e9cf7cf93b0af16fc89204378bc59f=
f30008e
>>> to use "reg-io-width =3D <4>;" as the reg-io-width property is =
generally used
>>> in the devicetree schema for such a condition.
>>>=20
>>> A release candidate of the ACLINT specification is available at
>>> https://github.com/riscv/riscv-aclint/releases
>>>=20
>>> Add reg-io-width as optional property to the SiFive Core Local =
Interruptor.
>>>=20
>>> Signed-off-by: Heinrich Schuchardt =
<heinrich.schuchardt@canonical.com>
>>> ---
>>>  Documentation/devicetree/bindings/timer/sifive,clint.yaml | 7 =
+++++++
>>>  1 file changed, 7 insertions(+)
>>>=20
>>> diff --git =
a/Documentation/devicetree/bindings/timer/sifive,clint.yaml =
b/Documentation/devicetree/bindings/timer/sifive,clint.yaml
>>> index a35952f48742..266012d887b5 100644
>>> --- a/Documentation/devicetree/bindings/timer/sifive,clint.yaml
>>> +++ b/Documentation/devicetree/bindings/timer/sifive,clint.yaml
>>> @@ -41,6 +41,13 @@ properties:
>>>    reg:
>>>      maxItems: 1
>>>=20
>>> +  reg-io-width:
>>> +    description: |
>>> +      Some CLINT implementations, e.g. on the T-HEAD 9xx, only =
support
>>> +      32bit access for MTIMER.
>>> +    $ref: /schemas/types.yaml#/definitions/uint32
>>> +    const: 4
>> But this is not a "sifive,clint" anyway. Should there be a new
>> dt-bindings for T-Head clint variant?
>=20
> I assume by new dt-bindings variant you mean: Add a new compatible =
string in Documentation/devicetree/bindings/timer/sifive,clint.yaml.
>=20
> The vendor Debian image uses:
> compatible =3D "{allwinner,sun20i-d1-clint", "sifive,clint0=E2=80=9D};

I assume Bin means to stop calling it a sifive,clint0 and instead have
something like

    "allwinner,sun20i-d1-clint", "thead,clint0"

as is being done for their non-conforming PLIC. It=E2=80=99s worth =
pointing out
that the same is true here; the SiFive CLINT supports 64-bit accesses
on RV64, so this is not compatible with a SiFive CLINT. Moreover, the
RISC-V spec was clearly written in a way that intended 64-bit accesses
to be supported for RV64, though was not completely explicit about
that, which has now been resolved (see
https://github.com/riscv/riscv-isa-manual/issues/639), so this CLINT
violates the obvious intent of the pre-1.12 privileged specs (see
Andrew=E2=80=99s message in the issue, which agrees with my reading).

Jess