[U-Boot] [RFC/PATCH] sunxi: Populate the default environment for sunxi-based systems

[U-Boot] [RFC/PATCH] sunxi: Populate the default environment for sunxi-based systems

[Karsten Merker]

Hello,

in the following email you can find a patch to populate the
default environment for sunxi-based systems about which I would
apprechiate your opinion.

With the current progress in mainlining sunxi platform support,
mainline u-boot will probably soon be able to replace
u-boot-sunxi (https://github.com/linux-sunxi/u-boot-sunxi) for
many use cases.  Currently the default environment in mainline
u-boot for sunxi-based systems is rather bare; the only
predefined variable is bootm_size (which is required for proper
initrd- and dtb-relocation).  U-boot-sunxi predefines a set of
environment variables which are commonly used by existing boot
scripts and also offers a predefined bootcmd for automatic search
and execution of bootscripts.  I think that to facilitate using
mainline u-boot as a drop-in replacement for u-boot-sunxi,
mainline u-boot should provide similar defaults.

The original environment in u-boot-sunxi was written for an older
u-boot release and does not make use of modern u-boot features
(e.g. it uses hardcoded filesystem-specific load commands), so
just copying the environment 1:1 to mainline u-boot does not make
sense IMHO.  My patch tries to implement the minimum required for
autobooting existing boot scripts with more modern u-boot functions.

In addition to that, the patch defines some standard environment
variables which are commonly defined on other u-boot platforms,
but which have not been defined by u-boot-sunxi, in particular
kernel_addr_r, ramdisk_addr_r and fdt_addr_r.  Debian supports a
wide range of armhf system and tries to unify support for the
different platforms as good as possible.  Having the same
predefined variables with (platform-specific) load addresses on
all platforms would help us a lot with regard to generic boot
scripts and simplified user documentation.

Regarding the actual values for the various load addresses I am
not in all cases perfectly sure about the best variant. 
Following is the reasoning for my choices:

- fdt_addr_r

  U-boot-sunxi loads the FEX binary (Allwinner's proprietary
  hardware description format for their android kernels, i.e.
  a file functionally roughly equivalent to a dtb) to address 
  0x43000000, so I guess using the same for the dtb makes
  sense.

- scriptaddr

  has historically been expected to be 0x44000000; some old boot
  scripts assume this address.

- kernel_addr_r and ramdisk_addr_r

  Allwinner has historically loaded their kernels to 0x48000000
  (system specific kernels with all drivers compiled in, no initrd).

  Debian has been using 0x46000000 for the kernel and 0x48000000
  for the initrd. These values work well in practice, so I have
  used them in the patch, but if there are reasons to change them,
  I am happy to do so.

Comments welcome :-).

Regards,
Karsten

-- 
1.7.10.4

[U-Boot] [RFC/PATCH 1/1] sunxi: Populate the default environment for sunxi-based systems

[Karsten Merker]

Predefine a set of standard environment variables (kernel_addr_r,
ramdisk_addr_r, fdt_addr_r and scriptaddr) to sensible values on
sunxi-based systems and provide a default environment that allows
autobooting of existing boot scripts which were written for
pre-mainlining u-boot-sunxi versions.

Signed-off-by: Karsten Merker <merker@debian.org>
---
 include/configs/sunxi-common.h |   12 +++++++++++-
 1 file changed, 11 insertions(+), 1 deletion(-)

diff --git a/include/configs/sunxi-common.h b/include/configs/sunxi-common.h
index 6a3044f..ac8cb8f 100644
--- a/include/configs/sunxi-common.h
+++ b/include/configs/sunxi-common.h
@@ -124,7 +124,17 @@
 #define CONFIG_ENV_SIZE			(128 << 10)	/* 128 KiB */
 
 #define CONFIG_EXTRA_ENV_SETTINGS \
-	"bootm_size=0x10000000\0"
+	"bootm_size=0x10000000\0" \
+	"fdt_addr_r=0x43000000\0" \
+	"scriptaddr=0x44000000\0" \
+	"kernel_addr_r=0x46000000\0" \
+	"ramdisk_addr_r=0x48000000\0" \
+	"device=mmc\0" \
+	"partition=0:1\0" \
+	"bootdelay=3\0" \
+	"bootcmd=if test -n \"${loadbootscr}\"; then run loadbootscr; fi\0" \
+	"bootscr=boot.scr\0" \
+	"loadbootscr=if test -n \"${device}\" && test -n \"${partition}\"; then for p in /boot/ /; do load ${device} ${partition} ${scriptaddr} ${p}${bootscr} && echo \"Running ${p}${bootscr} from ${device} ${partition}\" && source ${scriptaddr}; done; fi\0"
 
 #define CONFIG_SYS_BOOT_GET_CMDLINE
 
-- 
1.7.10.4

[U-Boot] [RFC/PATCH] sunxi: Populate the default environment for sunxi-based systems

[Hans de Goede]

Hi,

On 08/10/2014 11:00 PM, Karsten Merker wrote:
> Hello,
> 
> in the following email you can find a patch to populate the
> default environment for sunxi-based systems about which I would
> apprechiate your opinion.
> 
> With the current progress in mainlining sunxi platform support,
> mainline u-boot will probably soon be able to replace
> u-boot-sunxi (https://github.com/linux-sunxi/u-boot-sunxi) for
> many use cases.  Currently the default environment in mainline
> u-boot for sunxi-based systems is rather bare; the only
> predefined variable is bootm_size (which is required for proper
> initrd- and dtb-relocation).  U-boot-sunxi predefines a set of
> environment variables which are commonly used by existing boot
> scripts and also offers a predefined bootcmd for automatic search
> and execution of bootscripts.  I think that to facilitate using
> mainline u-boot as a drop-in replacement for u-boot-sunxi,
> mainline u-boot should provide similar defaults.
> 
> The original environment in u-boot-sunxi was written for an older
> u-boot release and does not make use of modern u-boot features
> (e.g. it uses hardcoded filesystem-specific load commands), so
> just copying the environment 1:1 to mainline u-boot does not make
> sense IMHO.  My patch tries to implement the minimum required for
> autobooting existing boot scripts with more modern u-boot functions.
> 
> In addition to that, the patch defines some standard environment
> variables which are commonly defined on other u-boot platforms,
> but which have not been defined by u-boot-sunxi, in particular
> kernel_addr_r, ramdisk_addr_r and fdt_addr_r.  Debian supports a
> wide range of armhf system and tries to unify support for the
> different platforms as good as possible.  Having the same
> predefined variables with (platform-specific) load addresses on
> all platforms would help us a lot with regard to generic boot
> scripts and simplified user documentation.
> 
> Regarding the actual values for the various load addresses I am
> not in all cases perfectly sure about the best variant. 
> Following is the reasoning for my choices:
> 
> - fdt_addr_r
> 
>   U-boot-sunxi loads the FEX binary (Allwinner's proprietary
>   hardware description format for their android kernels, i.e.
>   a file functionally roughly equivalent to a dtb) to address 
>   0x43000000, so I guess using the same for the dtb makes
>   sense.
> 
> - scriptaddr
> 
>   has historically been expected to be 0x44000000; some old boot
>   scripts assume this address.
> 
> - kernel_addr_r and ramdisk_addr_r
> 
>   Allwinner has historically loaded their kernels to 0x48000000
>   (system specific kernels with all drivers compiled in, no initrd).
> 
>   Debian has been using 0x46000000 for the kernel and 0x48000000
>   for the initrd. These values work well in practice, so I have
>   used them in the patch, but if there are reasons to change them,
>   I am happy to do so.
> 
> Comments welcome :-).

Thanks for this patch, but this does the same as a series
I've already pending, see:

https://github.com/jwrdegoede/u-boot-sunxi/commits/next

I've not yet send a pull-request for this because it uses the new
generic $bootcmd support:
https://github.com/jwrdegoede/u-boot-sunxi/commit/34e421a155c83c52f6e6edb2594458af27ef7517

This commit has just been merged by Tom Rini, so right now
these patches are good to go. I'll send a pull-request for them
soon.

Let me know if somehow these patches won't work for Debian. One of the
goals of the generic $bootcmd support is that distributions can just
drop an extlinux.conf in the boot partition, and then u-boot will
automatically find it and do the right thing.

Regards,

Hans

[U-Boot] [RFC/PATCH] sunxi: Populate the default environment for sunxi-based systems

[Karsten Merker]

On Mon, Aug 11, 2014 at 06:22:09PM +0200, Hans de Goede wrote:

> On 08/10/2014 11:00 PM, Karsten Merker wrote:
> > Hello,
> > 
> > in the following email you can find a patch to populate the
> > default environment for sunxi-based systems about which I would
> > apprechiate your opinion.
> > 
> > With the current progress in mainlining sunxi platform support,
> > mainline u-boot will probably soon be able to replace
> > u-boot-sunxi (https://github.com/linux-sunxi/u-boot-sunxi) for
> > many use cases.  Currently the default environment in mainline
> > u-boot for sunxi-based systems is rather bare; the only
> > predefined variable is bootm_size (which is required for proper
> > initrd- and dtb-relocation).  U-boot-sunxi predefines a set of
> > environment variables which are commonly used by existing boot
> > scripts and also offers a predefined bootcmd for automatic search
> > and execution of bootscripts.  I think that to facilitate using
> > mainline u-boot as a drop-in replacement for u-boot-sunxi,
> > mainline u-boot should provide similar defaults.

[...]

> Thanks for this patch, but this does the same as a series
> I've already pending, see:
> 
> https://github.com/jwrdegoede/u-boot-sunxi/commits/next

Sorry, I had not seen that - in this case please disregard
my patch.

> I've not yet send a pull-request for this because it uses the new
> generic $bootcmd support:
> https://github.com/jwrdegoede/u-boot-sunxi/commit/34e421a155c83c52f6e6edb2594458af27ef7517
> 
> This commit has just been merged by Tom Rini, so right now
> these patches are good to go. I'll send a pull-request for them
> soon.
> 
> Let me know if somehow these patches won't work for Debian. One of the
> goals of the generic $bootcmd support is that distributions can just
> drop an extlinux.conf in the boot partition, and then u-boot will
> automatically find it and do the right thing.

I have just run a few short tests with them - looks good. Using
the new infrastructure requires adjustments to existing
bootscripts written for u-boot-sunxi due to different variable
names (${devtype} vs. ${device}, ${devnum}:${bootpart} vs. 
${partition}), so it is not a real drop-in replacement for
u-boot-sunxi, but if it gets adopted by all platforms, I think
having a common and versatile infrastructure for all of them is
worth breaking compatibility to u-boot-sunxi in this regard.

Regards,
Karsten
-- 
Gem. Par. 28 Abs. 4 Bundesdatenschutzgesetz widerspreche ich der Nutzung
sowie der Weitergabe meiner personenbezogenen Daten f?r Zwecke der
Werbung sowie der Markt- oder Meinungsforschung.

[U-Boot] [RFC/PATCH] sunxi: Populate the default environment for sunxi-based systems

[Ian Campbell]

On Mon, 2014-08-11 at 22:13 +0200, Karsten Merker wrote:
(sorry for the delay, just getting back from vacation)

(sorry again if this comes through twice, my MUA crashed the first time
I hit send, I don't think the mail made it out though...)

> > I've not yet send a pull-request for this because it uses the new
> > generic $bootcmd support:
> > https://github.com/jwrdegoede/u-boot-sunxi/commit/34e421a155c83c52f6e6edb2594458af27ef7517
> > 
> > This commit has just been merged by Tom Rini, so right now
> > these patches are good to go. I'll send a pull-request for them
> > soon.
> > 
> > Let me know if somehow these patches won't work for Debian. One of the
> > goals of the generic $bootcmd support is that distributions can just
> > drop an extlinux.conf in the boot partition, and then u-boot will
> > automatically find it and do the right thing.
> 
> I have just run a few short tests with them - looks good. Using
> the new infrastructure requires adjustments to existing
> bootscripts written for u-boot-sunxi due to different variable
> names (${devtype} vs. ${device}, ${devnum}:${bootpart} vs. 
> ${partition}), so it is not a real drop-in replacement for
> u-boot-sunxi, but if it gets adopted by all platforms, I think
> having a common and versatile infrastructure for all of them is
> worth breaking compatibility to u-boot-sunxi in this regard.

I agree, the consistency is well worth it.

With Debian hat on: we should have a conversation with Vagrant WRT the
planned u-boot version for Jessie. v2014.10 should be available before
the freeze in November so hopefully we can go with that for Jessie, but
if not then perhaps we can backport some minimal set of patches so that
at least the env names match even if we don't manage to get the full
updated default env.

Ian