Re: [PATCH v7 1/1] RISC-V: Use SBI SRST extension when available

From: Heinrich Schuchardt
Date: Tue Nov 09 2021 - 10:19:31 EST

On 7/29/21 08:10, Atish Patra wrote:
On Wed, Jul 28, 2021 at 9:30 PM Palmer Dabbelt <palmer@xxxxxxxxxxx> wrote:

On Sun, 11 Jul 2021 11:59:33 PDT (-0700), Palmer Dabbelt wrote:
On Fri, 09 Jul 2021 22:01:02 PDT (-0700), Anup Patel wrote:


On 08/07/21, 9:22 AM, "Anup Patel" <anup@xxxxxxxxxxxxxx> wrote:

On Wed, Jul 7, 2021 at 1:57 AM Palmer Dabbelt <palmerdabbelt@xxxxxxxxxx> wrote:
>
> On Mon, 21 Jun 2021 21:46:46 PDT (-0700), anup@xxxxxxxxxxxxxx wrote:
> > Hi Palmer,
> >
> > On Wed, Jun 9, 2021 at 5:43 PM Anup Patel <anup.patel@xxxxxxx> wrote:
> >>
> >> The SBI SRST extension provides a standard way to poweroff and
> >> reboot the system irrespective to whether Linux RISC-V S-mode
> >> is running natively (HS-mode) or inside Guest/VM (VS-mode).
> >>
> >> The SBI SRST extension is available in the SBI v0.3 specification.
> >> (Refer, https://github.com/riscv/riscv-sbi-doc/releases/tag/v0.3.0-rc1)
> >
> > Can you please consider this patch for Linux-5.14-rc1 ?
> >
> > The SBI v0.3 spec is already frozen and this patch has been
> > floating on LKML for quite a few months now.
>
> I didn't realize that SBI-0.3 had been frozed. That link is to a RC,
> the cooresponding v0.3.0 tag isn't in that repo. Can you give me a
> pointer to the frozen spec?

Here's the link to SBI v0.3.0 tag:
https://github.com/riscv/riscv-sbi-doc/releases/tag/v0.3.0

We treat RC tags as frozen in SBI spec because no functional
changes are done in SBI spec after it is tagged as RC. We only
do typo fixes and clarifications on SBI spec RC release.

Treating the 0.3.0-rc1 as frozen as soon as it's released is a
terrifying policy: some of the fixes I sent in after I saw rc1 released
change the actual meaning of the text, even if they were meant to change
them to what I thought the intended meaning was supposed to be. That
means the actual text of 0.3.0-rc1 and 0.3.0 conflict with each other.
Given that frozen comes with a guarntee of backwards compatibility, does
that mean that the behavior allowed by 0.3.0-rc1 is compliant with the
SBI, even if it was likely just allowed by a wording mistake?

If you're going to freeze things at rc1 then you really need to be quite
explicit about that, as generally the point of RCs is to elicit
review/testing. Looks like I was the only person to have provided any
review, so I guess I was the only one who assumed "We don't expect any
significant functional changes. We will wait for any further feedback
and release the official v0.3 in a month or so." actually meant "this is
frozen".

Can you take this patch for Linux-5.14 ??

No, sorry, it's way too late for that. Please be specific about when
you freeze specifications in the future, so we can all stay on the same
page.

I went and talked to Krste, and he says that there's a whole process for
freezing extensions that this hasn't gone through. They don't have
anything written down that I can point to, but can you guys please just
get on the same page about this? It seems like every time I talk to

Absolutely. The freezing extensions process is documented right now[1]
but that is only meant
for ISA/hardware/platform specifications. There is no process defined
for a SBI specification which is purely
a software specification because SBI specification release
processes(v0.1 and v0.2) predate these documented processes.
The SBI specification is owned by the Platform HSC which falls under
the purview of software HC.
You can see a detailed chart of the RVI organization at [2]. All the
aspects of SBI specification are discussed
in platform meetings[3] and frozen only after public review[4] and
approval from the platform working group
and the software HC. The official SBI specification(v0.3) will also be
available along with all other RISC-V specifications
once they figure out how to structure non-ISA specifications.

I have cc'd Kumar (chair of the Platform HSC) and Philip (chair of the
software HC) in case they want to add anything.
I was not aware of the fact that Krste/Andrew are not aware of the
progress of the SBI specification.
I will raise this topic during the next meeting and make sure they are
in the loop as well.

someone from the RISC-V foundation I get a conflicting description of
what's going on, and I'm entirely out of patience when it comes to
getting blamed for all the chaos over there.

I agree the RVI process has not been very clear in the past. However,
that has changed a lot in recent times thanks to Mark and
other working group chairs. I don't think anybody is blaming you for
the delay in ratification of the RVI specifications.
There is a clear path for all the specifications to be ratified e.g.
the AIA and H extensions are planned to be frozen by the end of this
year.
Let me know if you want to see the timeline of each specification and
I can point you to the correct sheet.

[1] https://docs.google.com/presentation/d/1nQ5uFb39KA6gvUi5SReWfIQSiRN7hp6z7ZPfctE4mKk/edit#slide=id.ga0a994c3c8_0_6
[2] https://docs.google.com/presentation/d/1eEVuu6lRZd9iiDnZQSZME7Q7svtTG3pGIKHPmZ79B8E/edit#slide=id.ga275a504df_0_9
[3] https://github.com/riscv/riscv-platform-specs/wiki
[4] https://lists.riscv.org/g/tech-unixplatformspec/message/1042

https://github.com/riscv-non-isa/riscv-sbi-doc/releases/tag/v0.3.1-rc1
has:

"This tag the release candidate of version 0.3.1 of the RISC-V SBI specification. It doesn't have any significant changes other than typos.
A new release is created to adapt the ratification process for non-ISA specifications defined by RVI recently."

Has this patch to wait until release 0.3.1 of the SBI specification is ratified?
What is the timeline?

Could you, please, provide a link the the non-ISA ratification process description.

Best regards

Heinrich