Hi all. Several EBBR meetings ago, I introduced the need for allowing OS provided device trees [1]. Please find below the proposal I am delinquent on sending.
Hopefully, we can discuss this in the next meeting.
Thanks
Jon
[1] https://github.com/ARM-software/ebbr/wiki/EBBR-Notes-2024.02.12
Problem statement:
==================
Device trees are in theory a pure description of the hardware, and since the hardware
doesn't change, the device tree describing the hardware likewise never changes.
With this, a device tree could then be burned into the hardware's ROM to be
queried by software for hardware discovery. In practice, though, device trees
evolve over time. They evolve for many reasons, including
- support for previously unsupported hardware
- device driver improvements that require additional hardware information
- bug fixes
Linux's device tree source is maintained with the kernel source, and kernel builds
include building the device trees too. This ensures that the device tree
matching the kernel's usage is always kept in sync. Often, embedded distros will
include the matching device tree blobs.
The EBBR mandates that the device tree blob is provided by the firmware.
Thus it is likely that the device tree provided by the firmware and given to the
operating system is not the matching device tree blob for that kernel. This can
cause hardware to be missing, buggy, or non-functional.
Proposal:
=========
A key goal of the EBBR is to define the contract between the firmware and the OS
so that the OS doesn't need to be built specifically for the hardware, and the
firmware can boot any compliant OS. Thus, any solution that requires the OS to
know specifics about the hardware beyond the EBBR contract would violate the
EBBR goals. This precludes any solution where the OS, having the matching DTBs,
would pick the DTB, because this requires the OS to know what hardware it is
being run on. Likewise, any solution where the firmware is aware of the OS
matching DTBs would require the firmware to be aware of the particular OS it is
booting.
What can be known:
- The firmware knows what board it is running on, and thus knows what device
tree to use. But it doesn't know what version of the device tree to use,
because it doesn't know what OS is being booted.
- The OS knows what version of DTBs matches it's kernel, but does not know which
specific device tree to use.
This proposal then has the firmware choose the device tree by name, or some
other identifier that can be used to match the device tree for the board [1]. It
has the OS-provided OS loader select the location of the matching versions of
DTBs for it.
The firmware would pass the device tree filename/id to the OS loader, instead of
the DTB itself. The OS loader would determine the location of the matching DTBs
based on the chosen OS to boot, load the matching DTB from that location, and
pass to the kernel.
Considerations:
- often a DTB requires fixups. The EFI_DT_FIXUP_PROTOCOL could be utilized.
- device tree overlays could be indicated with a scheme using the device tree ID
passed to the OS loader
- authenticating the DTB would be the responsibility of the OS distribution and
handled in the same way as the kernel itself is authenticated. The OS is the
entity responsible for signing the DTB, as it should be.
This proposal should be in addition to supporting the standard way of passing in
a firmware-provided DT, in cases where the OS doesn't provide or have a need to
provide a matching DT.
[1] Rather than using the device tree source filename, to have more flexibility,
one can conceive an ID or compatible string that the OS could then scan the DTBs
to find a match.
Hi,
I am happy to announce that v2.2.0 of the EBBR specification has now been
released. [1]
The following people (listed in alphabetical order) have participated to this
EBBR release directly or indirectly, with patches, github issues, pull requests,
reviews, comments, suggestions, or by attending the zoom calls:
Ilias Apalodimas
Joakim Bech
Ard Biesheuvel
Etienne Carrière
Wei Chen
Patrick Delaunay
Samuel Dionne-Riel
Samer El-Haj-Mahmoud
Andreas Färber
Ahmad Fatoum
Sughosh Ganu
Simon Glass
Alexander Graf
Rob Herring
Jon Humphreys
Jan Kiszka
Jan Lübbe
Jose Marinho
Loïc Minier
Weizhao Ouyang
François-Frédéric Ozog
Stephen Paulger
Tom Rini
Peter Robinson
Heinrich Schuchardt
Michal Simek
Vincent Stehlé
Daniel Thompson
Ivan T. Ivanov
Stuart Yoder
A big thank you to all the EBBR contributors!
Best regards,
Vincent Stehlé
System Architect - Arm
[1]: https://github.com/ARM-software/ebbr/releases/tag/v2.2.0
On Sat, May 25, 2024 at 05:54:52PM +0100, Conor Dooley wrote:
> On Wed, May 22, 2024 at 04:54:23PM -0700, Elliot Berman wrote:
> > Device manufcturers frequently ship multiple boards or SKUs under a
> > single softwre package. These software packages ship multiple devicetree
> > blobs and require some mechanims to pick the correct DTB for the boards
> > that use the software package. This patch introduces a common language
> > for adding board identifiers to devicetrees.
> >
> > Signed-off-by: Elliot Berman <quic_eberman(a)quicinc.com>
> > ---
> > .../devicetree/bindings/board/board-id.yaml | 71 ++++++++++++++++++++++
> > 1 file changed, 71 insertions(+)
> >
> > diff --git a/Documentation/devicetree/bindings/board/board-id.yaml b/Documentation/devicetree/bindings/board/board-id.yaml
> > new file mode 100644
> > index 000000000000..894c1e310cbd
> > --- /dev/null
> > +++ b/Documentation/devicetree/bindings/board/board-id.yaml
> > @@ -0,0 +1,71 @@
> > +# SPDX-License-Identifier: (GPL-2.0 OR BSD-2-Clause)
> > +%YAML 1.2
> > +---
> > +$id: http://devicetree.org/schemas/board/board-id.yaml#
> > +$schema: http://devicetree.org/meta-schemas/core.yaml#
> > +
> > +title: Board identifiers
> > +description: |
> > + This node contains a list of identifier values for the board(s) supported by
> > + this devicetree. Identifier values are either N-tuples of integers or a
> > + string. The number of items for an N-tuple identifer is determined by the
> > + property name. String identifiers must be suffixed with "-string".
> > +
> > + Every identifier in the devicetree must have a matching value from the board
> > + to be considered a valid devicetree for the board. In other words: if
> > + multiple identifiers are present in the board-id and one identifier doesn't
> > + match against the board, then the devicetree is not applicable. Note this is
> > + not the case where the the board can provide more identifiers than the
> > + devicetree describes: those additional identifers can be ignored.
> > +
> > + Identifiers in the devicetree can describe multiple possible valid values,
> > + such as revision 1 and revision 2.
> > +
> > +maintainers:
> > + - Elliot Berman <quic_eberman(a)quicinc.com>
> > +
> > +properties:
> > + $nodename:
> > + const: '/'
> > + board-id:
>
>
> Does this need to be
> properties:
> $nodename:
> const: board-id
> ? That's the pattern I see for all top level nodes.
>
> > + type: object
> > + patternProperties:
> > + "^.*(?<!-string)$":
>
> At least this regex now actually works :)
>
> > + $ref: /schemas/types.yaml#/definitions/uint32-matrix
> > + description: |
> > + List of values that match boards this devicetree applies to.
> > + A bootloader checks whether any of the values in this list
> > + match against the board's value.
> > +
> > + The number of items per tuple is determined by the property name,
> > + see the vendor-specific board-id bindings.
> > + "^.*-string$":
> > + $ref: /schemas/types.yaml#/definitions/string-array
>
> Your description above doesn't match a string-array, just a single
> string. That said I'm far from sold on the "thou shalt have -string"
> edict. If every vendor is expected to go and define their own set of
> properties (and provide their own callback in your libfdt PoC) there's
> little to no reason to inflict property naming on them, AFAICT all that
> is gained is a being able to share
> if (string) {
> return fdt_stringlist_contains(prop->data,
> fdt32_to_cpu(prop->len),
> data);
> } else {
> // exact data comparison. data_len is the size of each entry
> if (fdt32_to_cpu(prop->len) % data_len || data_len % 4)
> return -FDT_ERR_BADVALUE;
>
> for (int i = 0; i < fdt32_to_cpu(prop->len); i += data_len) {
> if (!memcmp(&prop->data[i], data, data_len))
> return 1;
> }
>
> return 0;
> }
> in the libfdt PoC? I'd be expecting that a common mechanism would use
> the same "callback" for boards shipped by both Qualcomm and
> $other_vendor. Every vendor having different properties and only sharing
> the board-id node name seems a wee bit like paying lip-service to a
> common mechanism to me. What am I missing?
One way I thought to get the real board-id values from firmware to OS
loader is via DT itself. A firmware-provided DT provides the real
board-id values. In this case, firmware doesn't have any way to say the
board-id property is a string or a number, so I put that info in the DT
property name.
Another way I thought to get the real board-id values from firmware is
via a UEFI protocol. In that case, we could easily share whether the
value is a string or number and we can drop the "-string" suffix bit.
Thanks,
Elliot
On Sat, May 25, 2024 at 06:21:32PM +0100, Conor Dooley wrote:
> On Tue, May 21, 2024 at 11:38:01AM -0700, Elliot Berman wrote:
> > #define QCOM_BOARD_ID(a, major, minor) \
> > - (((major & 0xff) << 16) | ((minor & 0xff) << 8) | QCOM_BOARD_ID_##a)
> > + (((major & 0xff) << 16) | ((minor & 0xff) << 8) | ((QCOM_BOARD_ID_##a) & 0xff))
>
> I assume there's no devices that have a >8 bit QCOM_BOARD_ID that would
> end up with a different value in their dtb due to this change?
That's correct.
On Mon, May 27, 2024 at 09:19:59AM +0200, Michal Simek wrote:
> Hi,
>
> thanks for CCing me.
>
> On 5/24/24 17:51, Konrad Dybcio wrote:
> > On 21.05.2024 9:00 PM, Dmitry Baryshkov wrote:
> > > Hi Elliot,
> > >
> > > On Tue, 21 May 2024 at 21:41, Elliot Berman <quic_eberman(a)quicinc.com> wrote:
> > > >
> > > > Device manufacturers frequently ship multiple boards or SKUs under a
> > > > single software package. These software packages will ship multiple
> > > > devicetree blobs and require some mechanism to pick the correct DTB for
> > > > the board the software package was deployed. Introduce a common
> > > > definition for adding board identifiers to device trees. board-id
> > > > provides a mechanism for bootloaders to select the appropriate DTB which
> > > > is vendor/OEM-agnostic.
> > >
> > > This is a v3 of the RFC, however it is still a qcom-only series. Might
> > > I suggest gaining an actual interest from any other hardware vendor
> > > (in the form of the patches) before posting v4? Otherwise it might
> > > still end up being a Qualcomm solution which is not supported and/or
> > > used by other hardware vendors.
> >
> > AMD should be onboard [1].
> >
> > Konrad
> >
> > [1] https://resources.linaro.org/en/resource/q7U3Rr7m3ZbZmXzYK7A9u3
>
> I am trying to wrap my head around this and I have also looked at that EOSS
> presentation.
> I don't think I fully understand your case.
> There are multiple components which you need to detect. SOC - I expect
> reading by some regs, board - I expect you have any eeprom, OTP, adc, gpio,
> etc way how to detect board ID and revision.
> And then you mentioned displays - how do you detect them?
We have a similar mechanism to what you mention below: we have a ROM
which encodes information about the platform and that can be read by
firmware/bootloader/OS.
>
> In our Kria platform we have eeproms on SOM and CC cards (or FMC/extension
> cards) which we read and decode and based on information from it we are
> composing "unique" string. And then we are having DTBs in FIT image where
> description of configuration it taken as regular expression. That's why it
> is up to you how you want to combine them.
I don't think this is a fundamentally different approach from my
proposal. Instead of composing a "unique" string and using regex to
match, I'm proposing that the information (bytes) that are in your
eeprom can be matched without going through regex/string conversion.
Instead of firmware/bootloader doing a conversion to the strings, it
provides the values via board-id. I have concerns about having
bootloaders to contain a regex library -- probably easily addressed by
standardizing what terms the regex processor needs to support. I'm also
not sure if regex strings are an appropriate use of compatible strings.
Using strings limits the usefulness of comaptible strings to the
consumers of DTB, since the compatible string has to describe only the
boards the DTB is applicable to, you can't mention compatible strings
"this board is like" such as some generic SoC compatible.
> Currently we are merging them offline and we are not applying any DT overlay
> at run time but can be done (we are missing one missing piece in U-Boot for
> it).
>
> In presentation you mentioned also that applying overlay can fail but not
> sure how you can reach it. Because Linux kernel has the whole infrastructure
> to cover all combinations with base DT + overlays. It means if you cover all
> working combinations there you should see if they don't apply properly.
Mostly, I was referring to a situation where firmware provides an
overlay. Firmware doesn't know the DTB that OS has and I don't see any
way to gaurantee that firmware knows how to fix up the OS DTB.
>
> Also do you really need to detect everything from firmware side? Or isn't it
> enough to have just "some" devices and then load the rest where you are in
> OS?
> I think that's pretty much another way to go to have bare minimum
> functionality provided by firmware and deal with the rest in OS.
Agreed, although not all devices can be loaded once you are in the OS.
All nondiscoverable devices would need to be desribed in the DT.
Thanks,
Elliot
On Sat, May 25, 2024 at 06:08:46PM +0100, Conor Dooley wrote:
> On Tue, May 21, 2024 at 11:38:02AM -0700, Elliot Berman wrote:
> > +
> > +allOf:
> > + # either describe soc or soc-version; it's confusing to have both
>
> Why not just use the one that has the most information and discard the
> others? If your dtb picker for this platform doesn't care about the soc
> version, then just don't look at that cell?
The dtb picker for the platform doesn't know whether to care about the
SoC version/platform version/whatever. That's a property of the DTB
itself and I don't think it makes much sense to bake that into the DTB
picker which would otherwise be unaware of this.
>
> Likewise for platform and PMIC, why can't you ignore the cells you don't
> care about, rather than having a new property for each variant? Nothing
> in this patch explains why multiple variants are required rather than
> just dealing with the most informational.
>
Sure, I will explain in future revision.
- Elliot
Hi Conor,
Thanks for taking the time to look at the patch.
On Tue, May 21, 2024 at 08:21:45PM +0100, Conor Dooley wrote:
> On Tue, May 21, 2024 at 11:37:59AM -0700, Elliot Berman wrote:
> > Device manufcturers frequently ship multiple boards or SKUs under a
> > single softwre package. These software packages ship multiple devicetree
> > blobs and require some mechanims to pick the correct DTB for the boards
> > that use the software package.
>
> Okay, you've got the problem statement here, nice.
>
> > This patch introduces a common language
> > for adding board identifiers to devicetrees.
>
> But then a completely useless remainder of the commit message.
> I open this patch, see the regexes, say "wtf", look at the commit
> message and there is absolutely no explanation of what these properties
> are for. That's quite frankly just not good enough - even for an RFC.
>
Understood, I've been trying to walk the line of getting the idea across
to have conversation about the board-ids, while not getting into too
much of the weeds. I was hoping the example and the matching code in the
first patch would get enough of the idea across, but I totally
empathize that might not be enough. I'll reply here shortly with a
version of this patch which adds more details.
> >
> > Signed-off-by: Elliot Berman <quic_eberman(a)quicinc.com>
> > ---
> > .../devicetree/bindings/board/board-id.yaml | 24 ++++++++++++++++++++++
> > 1 file changed, 24 insertions(+)
> >
> > diff --git a/Documentation/devicetree/bindings/board/board-id.yaml b/Documentation/devicetree/bindings/board/board-id.yaml
> > new file mode 100644
> > index 000000000000..99514aef9718
> > --- /dev/null
> > +++ b/Documentation/devicetree/bindings/board/board-id.yaml
> > @@ -0,0 +1,24 @@
> > +# SPDX-License-Identifier: BSD-2-Clause
> > +%YAML 1.2
> > +---
> > +$id: http://devicetree.org/schemas/board/board-id.yaml#
> > +$schema: http://devicetree.org/meta-schemas/core.yaml#
> > +
> > +title: board identifiers
> > +description: Common property for board-id subnode
>
> s/property/properties/
>
> > +
> > +maintainers:
> > + - Elliot Berman <quic_eberman(a)quicinc.com>
> > +
> > +properties:
> > + $nodename:
> > + const: '/'
> > + board-id:
> > + type: object
> > + patternProperties:
> > + "^.*(?!_str)$":
>
> Does this regex even work? Take "foo_str" as an example - doesn't "^.*"
> consume all of the string, leaving the negative lookahead with nothing
> to object to? I didn't properly test this with an example and the dt
> tooling, but I lazily threw it into regex101 and both the python and
> emcascript versions agree with me. Did you test this?
Right, it should be a lookbehind, not a lookahead.
>
> And while I am here, no underscores in property names please. And if
> "str" means string, I suggest not saving 3 characters.
>
> > + $ref: /schemas/types.yaml#/definitions/uint32-matrix
> > + "^.*_str$":
> > + $ref: /schemas/types.yaml#/definitions/string-array
>
> Why do we even need two methods? Commit message tells me nothing and
> there's no description at all... Why do we need regexes here, rather
> than explicitly defined properties? Your commit message should explain
> the justification for that and the property descriptions (as comments if
> needs be for patternProperties) should explain why this is intended to
> be used.
>
> How is anyone supposed to look at this binding and understand how it
> should be used?
I was thinking that firmware may only provide the data without being
able to provide the context whether the value is a number or a string.
I think this is posisble if firmware provides the device's board
identifier in the format of a DT itself. It seems natural to me in the
EBBR flow. There is example of this in example in patches 3
(fdt-select-board) and 9 (the test suite). DTB doesn't inherently
provide instruction on how to interpret a property's value, so I created
a rule that strings have to be suffixed with "-string".
One other note -- I (QCOM) don't currently have a need for board-ids to
be strings. I thought it was likely that someone might want that though.
Thanks,
Elliot
On Tue, May 21, 2024 at 08:28:01PM +0100, Conor Dooley wrote:
> On Tue, May 21, 2024 at 11:37:58AM -0700, Elliot Berman wrote:
> > The devicetree spec introduced a mechanism to match devicetree blobs to
> > boards using firmware-provided identifiers.
>
> Can you share a link to where the devicetree spec introduced this
> mechanism? I don't recall seeing a PR to dt-schema for it nor did a
> quick check of the devicetree specification repo show a PR adding it.
>
> What am I missing?
My thinking is that the next patch would probably go to dt-schema or
devicetree specification repo.
Thanks,
Elliot
On Tue, May 21, 2024 at 2:25 PM Conor Dooley <conor(a)kernel.org> wrote:
>
> On Tue, May 21, 2024 at 08:21:45PM +0100, Conor Dooley wrote:
> > On Tue, May 21, 2024 at 11:37:59AM -0700, Elliot Berman wrote:
> > > Device manufcturers frequently ship multiple boards or SKUs under a
> > > single softwre package. These software packages ship multiple devicetree
> > > blobs and require some mechanims to pick the correct DTB for the boards
> > > that use the software package.
> >
> > Okay, you've got the problem statement here, nice.
> >
> > > This patch introduces a common language
> > > for adding board identifiers to devicetrees.
> >
> > But then a completely useless remainder of the commit message.
> > I open this patch, see the regexes, say "wtf", look at the commit
> > message and there is absolutely no explanation of what these properties
> > are for. That's quite frankly just not good enough - even for an RFC.
> >
> > >
> > > Signed-off-by: Elliot Berman <quic_eberman(a)quicinc.com>
> > > ---
> > > .../devicetree/bindings/board/board-id.yaml | 24 ++++++++++++++++++++++
> > > 1 file changed, 24 insertions(+)
> > >
> > > diff --git a/Documentation/devicetree/bindings/board/board-id.yaml b/Documentation/devicetree/bindings/board/board-id.yaml
> > > new file mode 100644
> > > index 000000000000..99514aef9718
> > > --- /dev/null
> > > +++ b/Documentation/devicetree/bindings/board/board-id.yaml
> > > @@ -0,0 +1,24 @@
> > > +# SPDX-License-Identifier: BSD-2-Clause
> > > +%YAML 1.2
> > > +---
> > > +$id: http://devicetree.org/schemas/board/board-id.yaml#
> > > +$schema: http://devicetree.org/meta-schemas/core.yaml#
> > > +
> > > +title: board identifiers
> > > +description: Common property for board-id subnode
> >
> > s/property/properties/
> >
> > > +
> > > +maintainers:
> > > + - Elliot Berman <quic_eberman(a)quicinc.com>
> > > +
> > > +properties:
> > > + $nodename:
> > > + const: '/'
> > > + board-id:
> > > + type: object
> > > + patternProperties:
> > > + "^.*(?!_str)$":
> >
> > Does this regex even work? Take "foo_str" as an example - doesn't "^.*"
> > consume all of the string, leaving the negative lookahead with nothing
> > to object to? I didn't properly test this with an example and the dt
> > tooling, but I lazily threw it into regex101 and both the python and
> > emcascript versions agree with me. Did you test this?
> >
> > And while I am here, no underscores in property names please. And if
> > "str" means string, I suggest not saving 3 characters.
> >
> > > + $ref: /schemas/types.yaml#/definitions/uint32-matrix
> > > + "^.*_str$":
> > > + $ref: /schemas/types.yaml#/definitions/string-array
> >
> > Why do we even need two methods? Commit message tells me nothing and
> > there's no description at all... Why do we need regexes here, rather
> > than explicitly defined properties? Your commit message should explain
> > the justification for that and the property descriptions (as comments if
> > needs be for patternProperties) should explain why this is intended to
> > be used.
> >
> > How is anyone supposed to look at this binding and understand how it
> > should be used?
>
> Also, please do not CC private mailing lists on your postings, I do not
> want to get spammed by linaro's mailman :(
boot-architecture is not private[0]. It is where EBBR gets discussed
amongst other things. This came up in a thread there[1].
Rob
[0] https://lists.linaro.org/mailman3/lists/boot-architecture.lists.linaro.org/
[1] https://lists.linaro.org/archives/list/boot-architecture@lists.linaro.org/t…
