[PATCH] Possible fix for u-boot on wandboard

[Steven Falco]

I've been looking into the reason why my wandboard would
not boot Fedora reliably.  I would typically get one
good boot every 3 or 4 tries.

It is late, and I need to stop now, but the following
proposed patch to uboot allows my board to boot without
errors, at least a dozen times in a row, which it has
never done before.

Basically, cmdtp->name is not valid for some reason,
so uboot goes off into the weeds.  I have not tracked
down the explanation, but if anyone else would like to
give this a try, I'd very much appreciate the feedback.

Signed-off-by: Steven A. Falco

--

diff -ur BUILD.original/u-boot-2013.07/common/cmd_bootm.c BUILD/u-boot-2013.07/common/cmd_bootm.c
--- BUILD.original/u-boot-2013.07/common/cmd_bootm.c	2013-07-23 07:58:13.000000000 -0400
+++ BUILD/u-boot-2013.07/common/cmd_bootm.c	2013-09-21 21:08:10.623374338 -0400
@@ -997,7 +998,7 @@
    	break;
  #endif
    default:
-		printf("Wrong Image Format for %s command\n", cmdtp->name);
+		printf("Wrong Image Format for command\n");
    	bootstage_error(BOOTSTAGE_ID_FIT_KERNEL_INFO);
    	return NULL;
    }

[Jon]

I can see how you zero in on that area of code as that is where is has
"Wrong Image Format".
(final words when the kernel does not boot with wandboard  quad)

Looking at the 2013.10-rc3 source from the current SRPM, I'm not sure
this patch would apply, probably not.

The cmdtp is the so-called "bootm command table" as noted in the
comments of cmd_bootm.c
So I would imagine cmdtp->cmd would be the string 'bootm' , but could be wrong.
Maybe the "cmdtp" is getting trashed somewhere along the line, and
poking the variable reveals problem.

My only problem with the patch is the area of code might be helpful on
other (working) boards.
Then again, I'm inclined to just hardcode the string 'bootm' right
there and be done.

It would be interesting to turn on/off CONFIG_FIT or whatever, the
ifdef's all around there seems to matter.

-Jon Disnard

On Sat, Sep 21, 2013 at 8:18 PM, Steven Falco  wrote:
> I've been looking into the reason why my wandboard would
> not boot Fedora reliably.  I would typically get one
> good boot every 3 or 4 tries.
>
> It is late, and I need to stop now, but the following
> proposed patch to uboot allows my board to boot without
> errors, at least a dozen times in a row, which it has
> never done before.
>
> Basically, cmdtp->name is not valid for some reason,
> so uboot goes off into the weeds.  I have not tracked
> down the explanation, but if anyone else would like to
> give this a try, I'd very much appreciate the feedback.
>
> Signed-off-by: Steven A. Falco
>
> --
>
> diff -ur BUILD.original/u-boot-2013.07/common/cmd_bootm.c
> BUILD/u-boot-2013.07/common/cmd_bootm.c
> --- BUILD.original/u-boot-2013.07/common/cmd_bootm.c    2013-07-23
> 07:58:13.000000000 -0400
> +++ BUILD/u-boot-2013.07/common/cmd_bootm.c     2013-09-21
> 21:08:10.623374338 -0400
> @@ -997,7 +998,7 @@
>                 break;
>  #endif
>         default:
>           printf("Wrong Image Format for %s command\n", cmdtp->name);
>           printf("Wrong Image Format for command\n");
>           bootstage_error(BOOTSTAGE_ID_FIT_KERNEL_INFO);
>           return NULL;
>   }
>
>
> arm mailing list
> [e-mail]
> https://admin.fedoraproject.org/mailman/listinfo/arm

[Niels de Vos]

On Sat, Sep 21, 2013 at 09:18:35PM -0400, Steven Falco wrote:
> I've been looking into the reason why my wandboard would
> not boot Fedora reliably.  I would typically get one
> good boot every 3 or 4 tries.
>
> It is late, and I need to stop now, but the following
> proposed patch to uboot allows my board to boot without
> errors, at least a dozen times in a row, which it has
> never done before.
>
> Basically, cmdtp->name is not valid for some reason,
> so uboot goes off into the weeds.  I have not tracked
> down the explanation, but if anyone else would like to
> give this a try, I'd very much appreciate the feedback.

Oh, nice catch. That would explain why the garbage gets printed. I don't 
know yet when I have time to test this, but I'll add it on my TODO.

This patch would be a workaround and not a final solution. I guess 
->name should be set correctly somewhere, or other members of *cmdtp 
should not overwrite the pointer.

Anyway, it looks like you narrowed down the cause pretty much. Thanks, 
for sharing!

Niels
> Signed-off-by: Steven A. Falco
>
> --
>
> diff -ur BUILD.original/u-boot-2013.07/common/cmd_bootm.c BUILD/u-boot-2013.07/common/cmd_bootm.c
> --- BUILD.original/u-boot-2013.07/common/cmd_bootm.c	2013-07-23 07:58:13.000000000 -0400
> +++ BUILD/u-boot-2013.07/common/cmd_bootm.c	2013-09-21 21:08:10.623374338 -0400
> @@ -997,7 +998,7 @@
>    	break;
>  #endif
>    default:
> printf("Wrong Image Format for %s command\n", cmdtp->name);
> printf("Wrong Image Format for command\n");
> bootstage_error(BOOTSTAGE_ID_FIT_KERNEL_INFO);
> return NULL;
> }
>
> arm mailing list
> [e-mail]
> https://admin.fedoraproject.org/mailman/listinfo/arm

[Steven Falco]

On 09/22/2013 06:36 AM, Niels de Vos wrote:
> On Sat, Sep 21, 2013 at 09:18:35PM -0400, Steven Falco wrote:
> > I've been looking into the reason why my wandboard would
> > not boot Fedora reliably.  I would typically get one
> > good boot every 3 or 4 tries.
> >
> > It is late, and I need to stop now, but the following
> > proposed patch to uboot allows my board to boot without
> > errors, at least a dozen times in a row, which it has
> > never done before.
> >
> > Basically, cmdtp->name is not valid for some reason,
> > so uboot goes off into the weeds.  I have not tracked
> > down the explanation, but if anyone else would like to
> > give this a try, I'd very much appreciate the feedback.
>
> Oh, nice catch. That would explain why the garbage gets printed. I don't
> know yet when I have time to test this, but I'll add it on my TODO.
>
> This patch would be a workaround and not a final solution. I guess
> ->name should be set correctly somewhere, or other members of *cmdtp
> should not overwrite the pointer.

Definitely not the final patch - it is just a hack to let
the board boot.  I have not had time yet to find the root
cause. Maybe later today.

But if it works for other people, at least I will know
I am on the right track.

There are other problems.  I added some debug, and the
magic number for the kernel is invalid:

Retrieving file: /vmlinuz-3.11.0-300.fc20.armv7hl
4748528 bytes read in 299 ms (15.1 MiB/s)
append: console=ttymxc0,115200 root=UUID=04e54c55-4d1e-4db7-b8df-5bd9defdc4ee ro rhgb quiet LANG=en_US.UTF-8
Retrieving file: /dtb-3.11.0-300.fc20.armv7hl/imx6dl-wandboard.dtb
21000 bytes read in 235 ms (86.9 KiB/s)
*  kernel: cmdline image address = 0x13000000
genimg_get_format: header at 13000000 magic 0000a0e1
genimg_get_format: fit header at 13000000 magic 0000a0e1
Wrong Image Format for command
ERROR: can't get kernel image!

I guess this is because of the vmlinuz format rather than
uImage format.  I really need to look at how the whole
extlinux mechanism works, and what it expects.

I've used uboot on various projects for many years, but the
extlinux mechanism is new to me.

Steve
> Anyway, it looks like you narrowed down the cause pretty much. Thanks,
> for sharing!
>
> Niels
> > Signed-off-by: Steven A. Falco
> >
> > --
> >
> > diff -ur BUILD.original/u-boot-2013.07/common/cmd_bootm.c BUILD/u-boot-2013.07/common/cmd_bootm.c
> > --- BUILD.original/u-boot-2013.07/common/cmd_bootm.c	2013-07-23 07:58:13.000000000 -0400
> > +++ BUILD/u-boot-2013.07/common/cmd_bootm.c	2013-09-21 21:08:10.623374338 -0400
> > @@ -997,7 +998,7 @@
> >    	break;
> >   #endif
> >    default:
> > printf("Wrong Image Format for %s command\n", cmdtp->name);
> > printf("Wrong Image Format for command\n");
> > bootstage_error(BOOTSTAGE_ID_FIT_KERNEL_INFO);
> > return NULL;
> > }
> >
> > arm mailing list
> > [e-mail]
> > https://admin.fedoraproject.org/mailman/listinfo/arm

[Clive Messer]

On Sat, 2013-09-21 at 21:18 -0400, Steven Falco wrote:
> diff -ur BUILD.original/u-boot-2013.07/common/cmd_bootm.c BUILD/u-boot-2013.07/common/cmd_bootm.c
> --- BUILD.original/u-boot-2013.07/common/cmd_bootm.c	2013-07-23 07:58:13.000000000 -0400
> +++ BUILD/u-boot-2013.07/common/cmd_bootm.c	2013-09-21 21:08:10.623374338 -0400
> @@ -997,7 +998,7 @@
>    	break;
>   #endif
>    default:
> printf("Wrong Image Format for %s command\n", cmdtp->name);
> printf("Wrong Image Format for command\n");
> bootstage_error(BOOTSTAGE_ID_FIT_KERNEL_INFO);
> return NULL;
> }

I'm sure you don't need me to confirm you're on the right track, but
yes, with the patch, able to boot every time on the Wandboard, rather
than a one in five chance.

Clive

[Steven Falco]

On 09/22/2013 09:03 AM, Clive Messer wrote:
> On Sat, 2013-09-21 at 21:18 -0400, Steven Falco wrote:
> > diff -ur BUILD.original/u-boot-2013.07/common/cmd_bootm.c BUILD/u-boot-2013.07/common/cmd_bootm.c
> > --- BUILD.original/u-boot-2013.07/common/cmd_bootm.c	2013-07-23 07:58:13.000000000 -0400
> > +++ BUILD/u-boot-2013.07/common/cmd_bootm.c	2013-09-21 21:08:10.623374338 -0400
> > @@ -997,7 +998,7 @@
> >    		break;
> >    #endif
> >    	default:
> > printf("Wrong Image Format for %s command\n", cmdtp->name);
> > printf("Wrong Image Format for command\n");
> > bootstage_error(BOOTSTAGE_ID_FIT_KERNEL_INFO);
> > return NULL;
> > }
>
> I'm sure you don't need me to confirm you're on the right track, but
> yes, with the patch, able to boot every time on the Wandboard, rather
> than a one in five chance.
>
> Clive

Excellent!  Thanks for testing.  I'll continue to sort this out, and
come up with a proper fix.

Steve