LKML Archive on lore.kernel.org
help / color / mirror / Atom feed
From: Marc Zyngier <maz@kernel.org>
To: Rob Herring <robh@kernel.org>
Cc: Mark Kettenis <mark.kettenis@xs4all.nl>,
	devicetree@vger.kernel.org, Robin Murphy <robin.murphy@arm.com>,
	Sven Peter <sven@svenpeter.dev>,
	Mark Kettenis <kettenis@openbsd.org>,
	Hector Martin <marcan@marcan.st>,
	Bjorn Helgaas <bhelgaas@google.com>,
	linux-arm-kernel <linux-arm-kernel@lists.infradead.org>,
	PCI <linux-pci@vger.kernel.org>,
	"linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>
Subject: Re: [PATCH v3 1/2] dt-bindings: pci: Add DT bindings for apple,pcie
Date: Sun, 15 Aug 2021 17:36:02 +0100	[thread overview]
Message-ID: <8735ra1x8t.wl-maz@kernel.org> (raw)
In-Reply-To: <CAL_JsqLvqWiuib9s4PzX8pOQYJQ0eR7Gxz==J849eVJ5MDq4SA@mail.gmail.com>

Hi Rob,

Apologies for the delay, I somehow misplaced this email...

On Mon, 02 Aug 2021 17:10:39 +0100,
Rob Herring <robh@kernel.org> wrote:
> 
> On Sun, Aug 1, 2021 at 3:31 AM Marc Zyngier <maz@kernel.org> wrote:
> >
> > On Tue, 27 Jul 2021 00:18:48 +0100,
> > Rob Herring <robh@kernel.org> wrote:
> > >
> > > On Mon, Jul 26, 2021 at 10:32:00AM +0200, Mark Kettenis wrote:
> > > > From: Mark Kettenis <kettenis@openbsd.org>
> > > >
> > > > The Apple PCIe host controller is a PCIe host controller with
> > > > multiple root ports present in Apple ARM SoC platforms, including
> > > > various iPhone and iPad devices and the "Apple Silicon" Macs.
> > > >
> > > > Signed-off-by: Mark Kettenis <kettenis@openbsd.org>
> > > > ---
> > > >  .../devicetree/bindings/pci/apple,pcie.yaml   | 166 ++++++++++++++++++
> > > >  MAINTAINERS                                   |   1 +
> > > >  2 files changed, 167 insertions(+)
> > > >  create mode 100644 Documentation/devicetree/bindings/pci/apple,pcie.yaml
> > > >
> > > > diff --git a/Documentation/devicetree/bindings/pci/apple,pcie.yaml b/Documentation/devicetree/bindings/pci/apple,pcie.yaml
> > > > new file mode 100644
> > > > index 000000000000..bfcbdee79c64
> > > > --- /dev/null
> > > > +++ b/Documentation/devicetree/bindings/pci/apple,pcie.yaml
> > > > @@ -0,0 +1,166 @@
> > > > +# SPDX-License-Identifier: (GPL-2.0-only OR BSD-2-Clause)
> > > > +%YAML 1.2
> > > > +---
> > > > +$id: http://devicetree.org/schemas/pci/apple,pcie.yaml#
> > > > +$schema: http://devicetree.org/meta-schemas/core.yaml#
> > > > +
> > > > +title: Apple PCIe host controller
> > > > +
> > > > +maintainers:
> > > > +  - Mark Kettenis <kettenis@openbsd.org>
> > > > +
> > > > +description: |
> > > > +  The Apple PCIe host controller is a PCIe host controller with
> > > > +  multiple root ports present in Apple ARM SoC platforms, including
> > > > +  various iPhone and iPad devices and the "Apple Silicon" Macs.
> > > > +  The controller incorporates Synopsys DesigWare PCIe logic to
> > > > +  implements its root ports.  But the ATU found on most DesignWare
> > > > +  PCIe host bridges is absent.
> > >
> > > blank line
> > >
> > > > +  All root ports share a single ECAM space, but separate GPIOs are
> > > > +  used to take the PCI devices on those ports out of reset.  Therefore
> > > > +  the standard "reset-gpio" and "max-link-speed" properties appear on
> > >
> > > reset-gpios
> > >
> > > > +  the child nodes that represent the PCI bridges that correspond to
> > > > +  the individual root ports.
> > >
> > > blank line
> > >
> > > > +  MSIs are handled by the PCIe controller and translated into regular
> > > > +  interrupts.  A range of 32 MSIs is provided.  These 32 MSIs can be
> > > > +  distributed over the root ports as the OS sees fit by programming
> > > > +  the PCIe controller's port registers.
> > > > +
> > > > +allOf:
> > > > +  - $ref: /schemas/pci/pci-bus.yaml#
> > > > +
> > > > +properties:
> > > > +  compatible:
> > > > +    items:
> > > > +      - const: apple,t8103-pcie
> > > > +      - const: apple,pcie
> > > > +
> > > > +  reg:
> > > > +    minItems: 3
> > > > +    maxItems: 5
> > > > +
> > > > +  reg-names:
> > > > +    minItems: 3
> > > > +    maxItems: 5
> > > > +    items:
> > > > +      - const: config
> > > > +      - const: rc
> > > > +      - const: port0
> > > > +      - const: port1
> > > > +      - const: port2
> > > > +
> > > > +  ranges:
> > > > +    minItems: 2
> > > > +    maxItems: 2
> > > > +
> > > > +  interrupts:
> > > > +    description:
> > > > +      Interrupt specifiers, one for each root port.
> > > > +    minItems: 1
> > > > +    maxItems: 3
> > > > +
> > > > +  msi-controller: true
> > > > +  msi-parent: true
> > > > +
> > > > +  msi-ranges:
> > > > +    description:
> > > > +      A list of pairs <intid span>, where "intid" is the first
> > > > +      interrupt number that can be used as an MSI, and "span" the size
> > > > +      of that range.
> > > > +    $ref: /schemas/types.yaml#/definitions/uint32-matrix
> > > > +    items:
> > > > +      minItems: 2
> > > > +      maxItems: 2
> > >
> > > I still have issues I raised on v1 with this property. It's genericish
> > > looking, but not generic. 'intid' as a single cell can't specify any
> > > parent interrupt such as a GIC which uses 3 cells. You could put in all
> > > the cells, but you'd still be assuming which cell you can increment.
> >
> > The GIC bindings already use similar abstractions, see what we do for
> > both GICv2m and GICv3 MBIs. Other MSI controllers use similar
> > properties (alpine and loongson, for example).
> 
> That's the problem. Everyone making up their own crap.

And that crap gets approved:

https://lore.kernel.org/lkml/20200512205704.GA10412@bogus/

I'm not trying to be antagonistic here, but it seems that your
position on this very subject has changed recently.

> > > I think you should just list all these under 'interrupts' using
> > > interrupt-names to make your life easier:
> > >
> > > interrupt-names:
> > >   items:
> > >     - const: port0
> > >     - const: port1
> > >     - const: port2
> > >     - const: msi0
> > >     - const: msi1
> > >     - const: msi2
> > >     - const: msi3
> > >     ...
> > >
> > > Yeah, it's kind of verbose, but if the h/w block handles N interrupts,
> > > you should list N interrupts. The worst case for the above is N entries
> > > too if not contiguous.
> >
> > And that's where I beg to differ, again.
> >
> > Specifying interrupts like this gives the false impression that these
> > interrupts are generated by the device that owns them (the RC). Which
> > for MSIs is not the case.
> 
> It's no different than an interrupt controller node having an
> interrupts property. The source is downstream and the interrupt
> controller is combining/translating the interrupts.
> 
> The physical interrupt signals are connected to and originating in
> this block.

Oh, I also object to this, for the same reasons. The only case where
it makes sense IMHO is when the interrupt controller is a multiplexer.

> That sounds like perfectly 'describing the h/w' to me.

I guess we have a different view of about these things. At the end of
the day, I don't care enough as long as we can expose a range of
interrupts one way or another.

> > This is not only verbose, this is
> > semantically dubious. And what should we do when the number of
> > possible interrupt is ridiculously large, as it is for the GICv3 ITS?
> 
> I don't disagree with the verbose part. But that's not really an issue
> in this case.
> 
> > I wish we had a standard way to express these constraints. Until we
> > do, I don't think enumerating individual interrupts is a practical
> > thing to do, nor that it actually represents the topology of the
> > system.
> 
> The only way a standard way will happen is to stop accepting the
> custom properties.
> 
> All the custom properties suffer from knowledge of what the parent
> interrupt controller is. To fix that, I think we need something like
> this:
> 
> msi-ranges = <intspec base>, <intspec step>, <intspec end>;
> 
> 'intspec' is defined by the parent interrupt-controller cells. step is
> the value to add. And end is what to match on to stop aka the last
> interrupt in the range. For example, if the GIC is the parent, we'd
> have something like this:
> 
> <GIC_SPI 123 0>, <0 1 0>, <GIC_SPI 124 0>
> 
> Does this apply to cases other than MSI? I think so as don't we have
> the same type of properties with the low power mode shadow interrupt
> controllers?  So 'interrupt-ranges'?

This would work, though the increment seems a bit over-engineered. You
also may need this property to accept multiple ranges.

> It looks to me like there's an assumption in the kernel that an MSI
> controller has a linear range of parent interrupts? Is that correct
> and something that's guaranteed? That assumption leaks into the
> existing bindings.

Depends on how the controller works. In general, the range maps to the
MultiMSI requirements where the message is an offset from the base of
the interrupt range. So you generally end-up with ranges of at least
32 contiguous MSIs. Anything under that is sub-par and probably not
worth supporting.

Of course, the controller may have some mapping facilities, which
makes things more... interesting.

> It's fine for the kernel to assume that until there's a case that's
> not linear, but a common binding needs to be able handle a
> non-linear case.

Fair enough. I can probably work with Mark to upgrade the binding and
the M1 PCIe code. Could you come up with a more formalised proposal?

Thanks,

	M.

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

  reply	other threads:[~2021-08-15 16:36 UTC|newest]

Thread overview: 14+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2021-07-26  8:31 [PATCH v3 0/2] Apple M1 PCIe DT bindings Mark Kettenis
2021-07-26  8:32 ` [PATCH v3 1/2] dt-bindings: pci: Add DT bindings for apple,pcie Mark Kettenis
2021-07-26 23:18   ` Rob Herring
2021-07-31  9:44     ` Mark Kettenis
2021-08-01  9:31     ` Marc Zyngier
2021-08-02 16:10       ` Rob Herring
2021-08-15 16:36         ` Marc Zyngier [this message]
2021-08-15 19:19           ` Rob Herring
2021-08-18 19:56             ` Mark Kettenis
2021-08-18 20:51               ` Rob Herring
2021-08-22 17:44                 ` Mark Kettenis
2021-08-23 15:24                   ` Rob Herring
2021-07-26  8:32 ` [PATCH v3 2/2] arm64: apple: Add PCIe node Mark Kettenis
2021-07-26 10:05 ` [PATCH v3 0/2] Apple M1 PCIe DT bindings Marc Zyngier

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=8735ra1x8t.wl-maz@kernel.org \
    --to=maz@kernel.org \
    --cc=bhelgaas@google.com \
    --cc=devicetree@vger.kernel.org \
    --cc=kettenis@openbsd.org \
    --cc=linux-arm-kernel@lists.infradead.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=linux-pci@vger.kernel.org \
    --cc=marcan@marcan.st \
    --cc=mark.kettenis@xs4all.nl \
    --cc=robh@kernel.org \
    --cc=robin.murphy@arm.com \
    --cc=sven@svenpeter.dev \
    --subject='Re: [PATCH v3 1/2] dt-bindings: pci: Add DT bindings for apple,pcie' \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link

This is a public inbox, see mirroring instructions
on how to clone and mirror all data and code used for this inbox