[meta-xilinx] [PATCH] machine-xilinx-default.inc: Default to u-boot for Zynq

[Nathan Rossi]

On 28 April 2017 at 01:36, Manjukumar Harthikote Matha
<manjukumar.harthikote-matha at xilinx.com> wrote:
>
>
>> -----Original Message-----
>> From: Nathan Rossi [mailto:nathan at nathanrossi.com]
>> Sent: Thursday, April 27, 2017 7:11 AM
>> To: Manjukumar Harthikote Matha <MANJUKUM at xilinx.com>
>> Cc: Philip Balister <philip at balister.org>; meta-xilinx at lists.yoctoproject.org
>> Subject: Re: [meta-xilinx] [PATCH] machine-xilinx-default.inc: Default to u-boot for
>> Zynq
>>
>> On 27 April 2017 at 08:21, Manjukumar Harthikote Matha
>> <manjukumar.harthikote-matha at xilinx.com> wrote:
>> >
>> >
>> >> -----Original Message-----
>> >> From: Philip Balister [mailto:philip at balister.org]
>> >> Sent: Wednesday, April 26, 2017 12:38 PM
>> >> To: Manjukumar Harthikote Matha <MANJUKUM at xilinx.com>; Nathan Rossi
>> >> <nathan at nathanrossi.com>
>> >> Cc: meta-xilinx at lists.yoctoproject.org
>> >> Subject: Re: [meta-xilinx] [PATCH] machine-xilinx-default.inc: Default to u-boot
>> for
>> >> Zynq
>> >>
>> >> On 04/26/2017 03:06 PM, Manjukumar Harthikote Matha wrote:
>> >> >
>> >> >
>> >> >> -----Original Message-----
>> >> >> From: Nathan Rossi [mailto:nathan at nathanrossi.com]
>> >> >> Sent: Wednesday, April 26, 2017 10:54 AM
>> >> >> To: Manjukumar Harthikote Matha <MANJUKUM at xilinx.com>
>> >> >> Cc: meta-xilinx at lists.yoctoproject.org
>> >> >> Subject: Re: [meta-xilinx] [PATCH] machine-xilinx-default.inc:
>> >> >> Default to u-boot for Zynq
>> >> >>
>> >> >> On 27 April 2017 at 02:41, Manjukumar Harthikote Matha
>> >> >> <manjukumar.harthikote- matha at xilinx.com> wrote:
>> >> >>>
>> >> >>>
>> >> >>>> -----Original Message-----
>> >> >>>> From: meta-xilinx-bounces at yoctoproject.org [mailto:meta-xilinx-
>> >> >>>> bounces at yoctoproject.org] On Behalf Of Nathan Rossi
>> >> >>>> Sent: Wednesday, April 26, 2017 4:57 AM
>> >> >>>> To: meta-xilinx at lists.yoctoproject.org
>> >> >>>> Subject: [meta-xilinx] [PATCH] machine-xilinx-default.inc: Default
>> >> >>>> to u-boot for Zynq
>> >> >>>>
>> >> >>>> Upstream U-Boot provides an almost complete environment for the
>> >> >>>> majority of Zynq targets and specifically covers all the boot
>> >> >>>> functionality required for the boards in the meta-xilinx layer. As
>> >> >>>> such default to
>> >> >> the mainline version of U-Boot.
>> >> >>>>
>> >> >>>> For users that require or prefer to use u-boot-xlnx this can be
>> >> >>>> selected on a per- machine basis using:
>> >> >>>>
>> >> >>>>   PREFERRED_PROVIDER_virtual/bootloader = "u-boot-xlnx"
>> >> >>>>
>> >> >>>> Signed-off-by: Nathan Rossi <nathan at nathanrossi.com>
>> >> >>>> ---
>> >> >>>>  conf/machine/include/machine-xilinx-default.inc | 1 +
>> >> >>>>  1 file changed, 1 insertion(+)
>> >> >>>>
>> >> >>>> diff --git a/conf/machine/include/machine-xilinx-default.inc
>> >> >>>> b/conf/machine/include/machine-xilinx-default.inc
>> >> >>>> index 13e4df5746..f9e7e3a33f 100644
>> >> >>>> --- a/conf/machine/include/machine-xilinx-default.inc
>> >> >>>> +++ b/conf/machine/include/machine-xilinx-default.inc
>> >> >>>> @@ -19,6 +19,7 @@ PREFERRED_VERSION_linux-xlnx ?= "4.6-xilinx-
>> v2016.4%"
>> >> >>>>
>> >> >>>>  # U-Boot Configuration
>> >> >>>>  XILINX_DEFAULT_UBOOT := "u-boot-xlnx"
>> >> >>>> +XILINX_DEFAULT_UBOOT_zynq := "u-boot"
>> >> >>>>  XILINX_DEFAULT_UBOOT_zynqmp := "u-boot"
>> >> >>>
>> >> >>> Why have any preferred_provider? Distro can provide these settings.
>> >> >>> For ex: meta-petalinux can provide u-boot-xlnx, oe-core can provide
>> >> >>> u-boot
>> >> >>>
>> >> >>> Any thoughts?
>> >> >>
>> >> >> Unfortunately not picking a provider for "virtual/bootloader" will
>> >> >> result in bitbake attempting to build all providers (since
>> >> >> EXTRA_IMAGEDEPENDS is depending on virtual/bootloader), as bitbake
>> >> >> has no idea which one is desired (and they are all compatible).
>> >> >>
>> >> >> --
>> >> >> NOTE: multiple providers are available for virtual/bootloader
>> >> >> (u-boot-xlnx, u-boot- xlnx-dev, u-boot)
>> >> >> NOTE: consider defining a PREFERRED_PROVIDER entry to match
>> >> >> virtual/bootloader ...
>> >> >> ERROR: Multiple .bb files are due to be built which each provide
>> >> >> virtual/bootloader ...
>> >> >> <various failures due to overlapping files/etc.>
>> >> >> --
>> >> >>
>> >> >> Also note, meta-petalinux or other layers should already be able to
>> >> >> override this default (by setting with ??=) either in a distro conf
>> >> >> or in a machine conf e.g. zynq- generic (for meta-petalinux).
>> >> >>
>> >> > If this is the case then why meta-xilinx should peg to upstream u-boot by
>> default?
>> >> It should peg to u-boot-xlnx or linux-xlnx by default. We know that there are
>> >> patches/drivers which are not upstreamed yet, and only available in Xilinx specific
>> >> tree. This layer is specific to Xilinx updates and should stick to defaults supported
>> by
>> >> Xilinx. Other distros or layer stack which use meta-xilinx should override
>> depending
>> >> on their requirements not the other way round.
>>
>> It should be noted that Xilinx does work with upstream U-Boot, since
>> Xilinx has an active maintainer (and active contributors). I would
>> call that support, but that is open to your own interpretation.
>>
>> >> > PREFERRED_PROVIDER_virtual/bootloader = "u-boot" should be specific to
>> boards
>> >> other than the Xilinx eval boards, for example in zybo/microzed.
>> >>
>> >> I'd be happiest if meta-xilinx used upstream u-boot and a bbappend to add
>> patches
>> >> that are not upstream yet. This way we (the consumers) know exactly what the
>> delta
>> >> is to upstream.
>> >>
>> > Yes agreed on the methodology, that it would be the best if we maintained
>> patchset on top upstream master for u-boot or kernel.
>> > But currently we don't have this model, and pinning down meta-xilinx to upstream
>> u-boot by default and not fetching Xilinx specific trees is not the right answer as well.
>> > meta-xilinx should ideally point to Xilinx trees, we have an option either via distro
>> or local.conf to change it to upstream at any given point.
>> > We also can point certain evaluation boards (zybo/microzed/picozed etc) to
>> upstream u-boot or kernel.
>> >
>> >> In this day and age, using non-mainline u-boot and kernel leads to future pain.
>> (Yes, I
>> >> have some old pain from using linux-xlnx)
>> > I agree, but certain drivers are not up-streamed and we are doing our best to get
>> them up-streamed.
>>
>> So the reason I have posted this patch for discussion is for this
>> exact reason. Since u-boot v2017.01 (mainline), the majority of
>> features and drivers are upstream (for zynq).
>>
>> The delta between xilinx/master
>> (92e3dd638b50ad22dd90072673c80d8730903e95) and u-boot v2017.01 (which
>> is the version in oe-core, as well as the base of the current
>> xilinx/master) is very small for zynq.
>>
>> Looking at the differences, ignoring the obvious zynqmp/microblaze
>> changes there are only a few features that are available in
>> u-boot-xlnx that are not in mainline:
>>  * partial bitstream loading support
>>  * secure/encrypted bitstream loading support
>>  * support for rsa verification of images
>>  * board configs/devicetrees for various Xilinx internal boards
>>
>> Given that these features have very specific requirements, and that
>> they rely on external Xilinx tools and software it is very unlikely
>> that users are interested in these features by default when using
>> meta-xilinx.
>
> This is totally debatable correct?

Yes, that is the point of this discussion.

Remember this patch is about changing the default for the machines in
meta-xilinx. However if someone uses meta-xilinx and creates a custom
machine the features might be desired, however in this case that
machine (or distro) can set to use u-boot-xlnx.

> When a feature is introduced there must have been a requirement from some set of customers for it be implemented.

Maybe, though that may not necessarily be the case. Also that does not
mean it was because all users wanted/needed said feature. Is the logic
that because these features exist it means that meta-xilinx machines
should be using the vendor tree despite not needing/using them?
Essentially throwing away all the benefits of using the mainline
source for no reason?

>
>  The fact that the delta is almost gone for Zynq suggests
>> that Xilinx has almost achieved that goal of fully upstreaming Zynq
>> support.
>>
>> In my opinion at least, I think that the benefits of using mainline
>> u-boot now greatly out weigh the benefits of u-boot-xlnx. As such I
>> think now is the best time to make this switch.
>>
>
> I disagree, meta-xilinx should not switch to mainline till delta is completely zero.

It would be great if you could provide some more information as to why
you disagree, it seems there is something I am missing?

Whilst it will be great to see zynq support reach a zero delta with
upstream, that might take years due to bug fixes and development for
zynq still occurring on u-boot-xlnx before upstream. And that is
ignoring the delta due to zynqmp/microblaze and driver overlap. Also
it can be argued that Xilinx might not have the incentive to get to a
almost zero delta since not enough users/customers are using mainline,
of which making this change could help provide a nudge for completing
it.

> The option to switch to mainline is available either through distro or local.conf.

Sure, but the same is also true of the reverse.

Regards,
Nathan