Unify the 64MiB flash layout

Unify the 64MiB flash layout

[Lei Yu]

This email is to share some information about the OpenBMC 64MiB flash layout.

In current Linux kernel, there is openbmc-flash-layout-64.dtsi that is
being used on meta-bytedance/meta-g220a system, there is on-going
change[1] for aspeed-bmc-ampere-mtjade.dts to use the layout as well.

However, according to ast2600's change[2], the layout does not fit
ast2600's u-boot, so it updates the layout by changing the existing
one.

On g220a system, we have tested the new layout and verifies that it
works fine, with below changes
1. Pick the ast2600's change[2] in kernel;
2. Pick the `image_types_phosphor.bbclass` change[3];
3. Update the u-boot env to boot from the new location, or use the new
u-boot-aspeed-sdk_2019.04.bb[4]
   On g220a, we use the new u-boot.

With the above changes, g220a system works fine with the new u-boot
and the updated flash layout.

To make it consistent for both ast2500 and ast2600, it's better to use
the same flash layout (and u-boot).
So I would suggest merge the changes of [1] and [3], for now g220a and
meta-ampere/meta-jade could be impacted.

[1]: https://lore.kernel.org/openbmc/20210517040036.13667-4-quan@os.amperecomputing.com/
[2]: https://lore.kernel.org/linux-arm-kernel/20210316085932.2601-1-troy_lee@aspeedtech.com/
[3]: https://gerrit.openbmc-project.xyz/c/openbmc/meta-phosphor/+/39343
[4]: https://github.com/openbmc/openbmc/blob/master/meta-aspeed/recipes-bsp/u-boot/u-boot-aspeed-sdk_2019.04.bb

-- 
BRs,
Lei YU

Re: Unify the 64MiB flash layout

[Patrick Williams]

[-- Attachment #1: Type: text/plain, Size: 911 bytes --]

On Thu, May 27, 2021 at 03:13:44PM +0800, Lei Yu wrote:
 
> With the above changes, g220a system works fine with the new u-boot
> and the updated flash layout.

Does it work for the flash0 u-boot to be able to boot flash1's system?
Maybe this doesn't affect g220a but in general we've found issues with
changing these kind of layouts and not updating both flash banks.  Due
to our secureboot implementation, we're not able to update the primary
flash bank (since it is write-protected).

This change doesn't currently affect us right now, but if we're making
a breaking change anyhow should we increase the u-boot partition
sufficiently that it won't become an issue later on as ast2600 systems
go into production?  We've found that over time the size tends to get
bigger with newer (more optimizing?) compilers, so I think we should
ensure there is some buffer there.

-- 
Patrick Williams

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: Unify the 64MiB flash layout

[Joel Stanley]

On Thu, 27 May 2021 at 07:14, Lei Yu <yulei.sh@bytedance.com> wrote:
>
> This email is to share some information about the OpenBMC 64MiB flash layout.
>
> In current Linux kernel, there is openbmc-flash-layout-64.dtsi that is
> being used on meta-bytedance/meta-g220a system, there is on-going
> change[1] for aspeed-bmc-ampere-mtjade.dts to use the layout as well.
>
> However, according to ast2600's change[2], the layout does not fit
> ast2600's u-boot, so it updates the layout by changing the existing
> one.
>
> On g220a system, we have tested the new layout and verifies that it
> works fine, with below changes
> 1. Pick the ast2600's change[2] in kernel;
> 2. Pick the `image_types_phosphor.bbclass` change[3];
> 3. Update the u-boot env to boot from the new location, or use the new
> u-boot-aspeed-sdk_2019.04.bb[4]
>    On g220a, we use the new u-boot.
>
> With the above changes, g220a system works fine with the new u-boot
> and the updated flash layout.
>
> To make it consistent for both ast2500 and ast2600, it's better to use
> the same flash layout (and u-boot).
> So I would suggest merge the changes of [1] and [3], for now g220a and
> meta-ampere/meta-jade could be impacted.

Thanks for the summary Lei. I will merge [2] for v5.14 and backport it
to dev-5.10.

Cheers,

Joel

>
> [1]: https://lore.kernel.org/openbmc/20210517040036.13667-4-quan@os.amperecomputing.com/
> [2]: https://lore.kernel.org/linux-arm-kernel/20210316085932.2601-1-troy_lee@aspeedtech.com/
> [3]: https://gerrit.openbmc-project.xyz/c/openbmc/meta-phosphor/+/39343
> [4]: https://github.com/openbmc/openbmc/blob/master/meta-aspeed/recipes-bsp/u-boot/u-boot-aspeed-sdk_2019.04.bb
>
> --
> BRs,
> Lei YU

Re: Unify the 64MiB flash layout

[Troy Lee]

Hi Team,

On Thu, Jun 3, 2021 at 2:00 PM Joel Stanley <joel@jms.id.au> wrote:
>
> On Thu, 27 May 2021 at 07:14, Lei Yu <yulei.sh@bytedance.com> wrote:
> >
> > This email is to share some information about the OpenBMC 64MiB flash layout.
> >
> > In current Linux kernel, there is openbmc-flash-layout-64.dtsi that is
> > being used on meta-bytedance/meta-g220a system, there is on-going
> > change[1] for aspeed-bmc-ampere-mtjade.dts to use the layout as well.
> >
> > However, according to ast2600's change[2], the layout does not fit
> > ast2600's u-boot, so it updates the layout by changing the existing
> > one.
> >
> > On g220a system, we have tested the new layout and verifies that it
> > works fine, with below changes
> > 1. Pick the ast2600's change[2] in kernel;
> > 2. Pick the `image_types_phosphor.bbclass` change[3];
> > 3. Update the u-boot env to boot from the new location, or use the new
> > u-boot-aspeed-sdk_2019.04.bb[4]
> >    On g220a, we use the new u-boot.
> >
> > With the above changes, g220a system works fine with the new u-boot
> > and the updated flash layout.
> >
> > To make it consistent for both ast2500 and ast2600, it's better to use
> > the same flash layout (and u-boot).
> > So I would suggest merge the changes of [1] and [3], for now g220a and
> > meta-ampere/meta-jade could be impacted.
>
> Thanks for the summary Lei. I will merge [2] for v5.14 and backport it
> to dev-5.10.
>
> Cheers,
>
> Joel
>
> >
> > [1]: https://lore.kernel.org/openbmc/20210517040036.13667-4-quan@os.amperecomputing.com/
> > [2]: https://lore.kernel.org/linux-arm-kernel/20210316085932.2601-1-troy_lee@aspeedtech.com/
> > [3]: https://gerrit.openbmc-project.xyz/c/openbmc/meta-phosphor/+/39343
Thanks for Lei and Joel's help. For change [3], I have resubmitted it
against openbmc/openbmc repository.
Please refer to: https://gerrit.openbmc-project.xyz/c/openbmc/openbmc/+/43876

And the patch has been verified with evb-ast2600 machine running with
AST2600-A3 EVB. However, to run on A3 EVB, the machine has to enable
SPL for u-boot by the following patch:
```
diff --git a/meta-aspeed/conf/machine/evb-ast2600.conf
b/meta-aspeed/conf/machine/evb-ast2600.conf
index e011045d2..2915b599f 100644
--- a/meta-aspeed/conf/machine/evb-ast2600.conf
+++ b/meta-aspeed/conf/machine/evb-ast2600.conf
@@ -1,6 +1,7 @@
 KERNEL_DEVICETREE = "aspeed-ast2600-evb.dtb"
-UBOOT_MACHINE = "ast2600_openbmc_defconfig"
-UBOOT_DEVICETREE = "ast2600a1-evb"
+UBOOT_MACHINE = "ast2600_openbmc_spl_defconfig"
+UBOOT_DEVICETREE = "ast2600-evb"
+SPL_BINARY = "spl/u-boot-spl.bin"

 require conf/machine/include/ast2600.inc
```

> > [4]: https://github.com/openbmc/openbmc/blob/master/meta-aspeed/recipes-bsp/u-boot/u-boot-aspeed-sdk_2019.04.bb
> >
> > --
> > BRs,
> > Lei YU

Thanks,
Troy Lee

Re: Unify the 64MiB flash layout

[Lei Yu]

On Wed, Jun 9, 2021 at 1:30 PM Troy Lee <leetroy@gmail.com> wrote:
>
> Hi Team,
>
> On Thu, Jun 3, 2021 at 2:00 PM Joel Stanley <joel@jms.id.au> wrote:
> >
> > On Thu, 27 May 2021 at 07:14, Lei Yu <yulei.sh@bytedance.com> wrote:
> > >
> > > This email is to share some information about the OpenBMC 64MiB flash layout.
> > >
> > > In current Linux kernel, there is openbmc-flash-layout-64.dtsi that is
> > > being used on meta-bytedance/meta-g220a system, there is on-going
> > > change[1] for aspeed-bmc-ampere-mtjade.dts to use the layout as well.
> > >
> > > However, according to ast2600's change[2], the layout does not fit
> > > ast2600's u-boot, so it updates the layout by changing the existing
> > > one.
> > >
> > > On g220a system, we have tested the new layout and verifies that it
> > > works fine, with below changes
> > > 1. Pick the ast2600's change[2] in kernel;
> > > 2. Pick the `image_types_phosphor.bbclass` change[3];
> > > 3. Update the u-boot env to boot from the new location, or use the new
> > > u-boot-aspeed-sdk_2019.04.bb[4]
> > >    On g220a, we use the new u-boot.
> > >
> > > With the above changes, g220a system works fine with the new u-boot
> > > and the updated flash layout.
> > >
> > > To make it consistent for both ast2500 and ast2600, it's better to use
> > > the same flash layout (and u-boot).
> > > So I would suggest merge the changes of [1] and [3], for now g220a and
> > > meta-ampere/meta-jade could be impacted.
> >
> > Thanks for the summary Lei. I will merge [2] for v5.14 and backport it
> > to dev-5.10.
> >
> > Cheers,
> >
> > Joel
> >
> > >
> > > [1]: https://lore.kernel.org/openbmc/20210517040036.13667-4-quan@os.amperecomputing.com/
> > > [2]: https://lore.kernel.org/linux-arm-kernel/20210316085932.2601-1-troy_lee@aspeedtech.com/
> > > [3]: https://gerrit.openbmc-project.xyz/c/openbmc/meta-phosphor/+/39343
> Thanks for Lei and Joel's help. For change [3], I have resubmitted it
> against openbmc/openbmc repository.
> Please refer to: https://gerrit.openbmc-project.xyz/c/openbmc/openbmc/+/43876
>
> And the patch has been verified with evb-ast2600 machine running with
> AST2600-A3 EVB. However, to run on A3 EVB, the machine has to enable
> SPL for u-boot by the following patch:
> ```
> diff --git a/meta-aspeed/conf/machine/evb-ast2600.conf
> b/meta-aspeed/conf/machine/evb-ast2600.conf
> index e011045d2..2915b599f 100644
> --- a/meta-aspeed/conf/machine/evb-ast2600.conf
> +++ b/meta-aspeed/conf/machine/evb-ast2600.conf
> @@ -1,6 +1,7 @@
>  KERNEL_DEVICETREE = "aspeed-ast2600-evb.dtb"
> -UBOOT_MACHINE = "ast2600_openbmc_defconfig"
> -UBOOT_DEVICETREE = "ast2600a1-evb"
> +UBOOT_MACHINE = "ast2600_openbmc_spl_defconfig"
> +UBOOT_DEVICETREE = "ast2600-evb"
> +SPL_BINARY = "spl/u-boot-spl.bin"
>
>  require conf/machine/include/ast2600.inc
> ```

Thanks for the information.
@Troy May I ask what the "spl/u-boot-spl.bin" is and do we have source
code of it?

@Joel Does OpenBMC has its own open-sourced u-boot-spl that does the
verification of the u-boot-proper?



-- 
BRs,
Lei YU

Re: Unify the 64MiB flash layout

[Troy Lee]

Hi Lei,

On Wed, Jun 9, 2021 at 2:43 PM Lei Yu <yulei.sh@bytedance.com> wrote:
>
> On Wed, Jun 9, 2021 at 1:30 PM Troy Lee <leetroy@gmail.com> wrote:
> >
> > Hi Team,
> >
> > On Thu, Jun 3, 2021 at 2:00 PM Joel Stanley <joel@jms.id.au> wrote:
> > >
> > > On Thu, 27 May 2021 at 07:14, Lei Yu <yulei.sh@bytedance.com> wrote:
> > > >
> > > > This email is to share some information about the OpenBMC 64MiB flash layout.
> > > >
> > > > In current Linux kernel, there is openbmc-flash-layout-64.dtsi that is
> > > > being used on meta-bytedance/meta-g220a system, there is on-going
> > > > change[1] for aspeed-bmc-ampere-mtjade.dts to use the layout as well.
> > > >
> > > > However, according to ast2600's change[2], the layout does not fit
> > > > ast2600's u-boot, so it updates the layout by changing the existing
> > > > one.
> > > >
> > > > On g220a system, we have tested the new layout and verifies that it
> > > > works fine, with below changes
> > > > 1. Pick the ast2600's change[2] in kernel;
> > > > 2. Pick the `image_types_phosphor.bbclass` change[3];
> > > > 3. Update the u-boot env to boot from the new location, or use the new
> > > > u-boot-aspeed-sdk_2019.04.bb[4]
> > > >    On g220a, we use the new u-boot.
> > > >
> > > > With the above changes, g220a system works fine with the new u-boot
> > > > and the updated flash layout.
> > > >
> > > > To make it consistent for both ast2500 and ast2600, it's better to use
> > > > the same flash layout (and u-boot).
> > > > So I would suggest merge the changes of [1] and [3], for now g220a and
> > > > meta-ampere/meta-jade could be impacted.
> > >
> > > Thanks for the summary Lei. I will merge [2] for v5.14 and backport it
> > > to dev-5.10.
> > >
> > > Cheers,
> > >
> > > Joel
> > >
> > > >
> > > > [1]: https://lore.kernel.org/openbmc/20210517040036.13667-4-quan@os.amperecomputing.com/
> > > > [2]: https://lore.kernel.org/linux-arm-kernel/20210316085932.2601-1-troy_lee@aspeedtech.com/
> > > > [3]: https://gerrit.openbmc-project.xyz/c/openbmc/meta-phosphor/+/39343
> > Thanks for Lei and Joel's help. For change [3], I have resubmitted it
> > against openbmc/openbmc repository.
> > Please refer to: https://gerrit.openbmc-project.xyz/c/openbmc/openbmc/+/43876
> >
> > And the patch has been verified with evb-ast2600 machine running with
> > AST2600-A3 EVB. However, to run on A3 EVB, the machine has to enable
> > SPL for u-boot by the following patch:
> > ```
> > diff --git a/meta-aspeed/conf/machine/evb-ast2600.conf
> > b/meta-aspeed/conf/machine/evb-ast2600.conf
> > index e011045d2..2915b599f 100644
> > --- a/meta-aspeed/conf/machine/evb-ast2600.conf
> > +++ b/meta-aspeed/conf/machine/evb-ast2600.conf
> > @@ -1,6 +1,7 @@
> >  KERNEL_DEVICETREE = "aspeed-ast2600-evb.dtb"
> > -UBOOT_MACHINE = "ast2600_openbmc_defconfig"
> > -UBOOT_DEVICETREE = "ast2600a1-evb"
> > +UBOOT_MACHINE = "ast2600_openbmc_spl_defconfig"
> > +UBOOT_DEVICETREE = "ast2600-evb"
> > +SPL_BINARY = "spl/u-boot-spl.bin"
> >
> >  require conf/machine/include/ast2600.inc
> > ```
>
> Thanks for the information.
> @Troy May I ask what the "spl/u-boot-spl.bin" is and do we have source
> code of it?

SPL stands for Secondary Program Loader in u-boot, it can be seem as a
small boot loader which runs in SRAM and inits DRAM, then it loads
u-boot.bin from SPI Flash/eMMC into DRAM.
You could turn enable it by setting "CONFIG_SPL=y" in defconfig.
The source code can be found in
- arch/arm/mach-aspeed/ast2600/spl
- common/spl
The u-boot.bin can be built as a FIT image, so that the SPL can verify
u-boot to achieve chain of trust.

After enables the CONFIG_SPL, you also have to tell yocto to copy
u-boot-spl.bin to the very beginning of image-u-boot and u-boot.bin at
64KB offset.
By setting the SPL_BINARY variable in machine config,
obmc-phosphor-image will handle the feature.
https://github.com/openbmc/openbmc/blob/77a1904b1833cd3b70b0dfd68671745eaacc9b19/meta-phosphor/classes/image_types_phosphor.bbclass#L239-L244

Troy
>
> @Joel Does OpenBMC has its own open-sourced u-boot-spl that does the
> verification of the u-boot-proper?
>
>
>
> --
> BRs,
> Lei YU