[U-Boot] Latest u-boot release on BeagleBone Black for FreeBSD

[U-Boot] Latest u-boot release on BeagleBone Black for FreeBSD

[Xuebing Wang]

Hi u-boot community,

I am trying to port u-boot (release u-boot-2014.04-rc3.tar.bz2) to 
FreeBSD on BeagleBone Black.

In FreeBSD, there is a u-boot loader (named ubldr), which can call 
u-boot API to get fdt (Flat Device Tree) data. Example is:
-- fdt header

I have to comment out below 3 lines, in order to get correct fdt data in 
FreeBSD ubldr from u-boot. Would you please advice what is the best way 
to fix this?

In file common/env_common.c:
const uchar *env_get_addr(int index)
{
//      if (gd->env_valid)
//              return (uchar *)(gd->env_addr + index);
//      else
                 return &default_environment[index];
}

-- 
Thanks,
Xuebing Wang

[U-Boot] Latest u-boot release on BeagleBone Black for FreeBSD

[Ian Lepore]

On Mon, 2014-04-07 at 15:16 +0800, Xuebing Wang wrote:
> Hi u-boot community,
> 
> I am trying to port u-boot (release u-boot-2014.04-rc3.tar.bz2) to 
> FreeBSD on BeagleBone Black.
> 
> In FreeBSD, there is a u-boot loader (named ubldr), which can call 
> u-boot API to get fdt (Flat Device Tree) data. Example is:
> -- fdt header
> 
> I have to comment out below 3 lines, in order to get correct fdt data in 
> FreeBSD ubldr from u-boot. Would you please advice what is the best way 
> to fix this?
> 
> In file common/env_common.c:
> const uchar *env_get_addr(int index)
> {
> //      if (gd->env_valid)
> //              return (uchar *)(gd->env_addr + index);
> //      else
>                  return &default_environment[index];
> }
> 

Are you using a fairly recent ubldr on the freebsd side?  I made some
changes around March 1st on how ubldr finds / loads fdt data...

It first looks for env vars named fdtaddr then fdt_addr in the u-boot
environment.  If the variable exists it tries to validate the data at
that address as a valid dtb header and if it is, that's the data that is
passed to the kernel.  Otherwise it next it looks for env vars named
fdtfile then fdt_file.  If one of those exists it tries to load that
filename from the freebsd filesystem in the usual places where kernel
and modules are located (/boot/kernel, /boot/modules).

It seems strange to me that things work only if you comment out the
dynamic environment in u-boot.  You said "in order to get the correct
fdt data" and I wonder what you mean by that.  Do you mean that the
u-boot script stuff is loading fdt data, but it's not the freebsd data
but rather the linux data?

Unfortunately, freebsd isn't able to use linux or vendor-supplied dtb
files yet.  We're working towards using only documented bindings but
right now we still need freebsd-specific extra properties in the dtb.

I added the support for fdtfile/fdt_file specifically to handle this
situation until we're able to handle the vendor dtb files.  If you
manipulate your u-boot environment so that fdtaddr and fdt_addr are
unset and fdtfile is the name of the appropriate freebsd dtb file (which
you must install on your freebsd image along with the kernel), then
everything should work out properly.

Of course an alternative is to install the freebsd dtb files onto the
msdos partition along with u-boot and let u-boot load them and set
fdtaddr.  If this is what you're doing and it still isn't working, I
suspect the problem is in ubldr, not u-boot itself.  We'll have to turn
on some debugging in sys/boot/fdt/fdt_loader_cmd.c to track that down.

-- Ian

[U-Boot] Latest u-boot release on BeagleBone Black for FreeBSD

[Tim Kientzle]

On May 7, 2014, at 8:02 AM, Tim Kientzle <tim@kientzle.com> wrote:

> 
> On May 6, 2014, at 7:09 PM, Xuebing Wang <xbing6@gmail.com> wrote:
> 
>> 2) Would you please point to me which revision that reserves both r8 and r9?
> 
> ------------------------------------------------------------------------
> r258527 | andrew | 2013-11-24 12:33:38 -0800 (Sun, 24 Nov 2013) | 3 lines
> 
> Recent versions of U-Boot require us to also backup and restore r9 for API
> calls to work.
> 
> ------------------------------------------------------------------------
> 

And MFCed to Stable-10:

------------------------------------------------------------------------
r265064 | ian | 2014-04-28 16:46:04 -0700 (Mon, 28 Apr 2014) | 2 lines

MFC r257210, r258527: No hardfloat in ubldr, save/restore r9 for api calls.

------------------------------------------------------------------------

[U-Boot] Latest u-boot release on BeagleBone Black for FreeBSD

[Tim Kientzle]

On May 6, 2014, at 7:09 PM, Xuebing Wang <xbing6@gmail.com> wrote:

> 2) Would you please point to me which revision that reserves both r8 and r9?

------------------------------------------------------------------------
r258527 | andrew | 2013-11-24 12:33:38 -0800 (Sun, 24 Nov 2013) | 3 lines

Recent versions of U-Boot require us to also backup and restore r9 for API
calls to work.

------------------------------------------------------------------------

[U-Boot] Latest u-boot release on BeagleBone Black for FreeBSD

[Ian Lepore]

On Wed, 2014-05-07 at 10:09 +0800, Xuebing Wang wrote:
> On 05/04/2014 07:33 PM, Jeroen Hofstee wrote:
> > Hello Xuebing, (freebsd-arm added on cc),
> >
> > On di, 2014-04-08 at 16:52 +0800, Xuebing Wang wrote:
> >> Hi u-boot community,
> >>
> >> I am trying to port u-boot (release u-boot-2014.04-rc3.tar.bz2) to
> >> FreeBSD on BeagleBone Black.
> >>
> >> In FreeBSD, there is a u-boot loader (named ubldr), which can call
> >> u-boot API to get fdt (Flat Device Tree) data.
> >>
> >> I have to comment out below 3 lines, in order to get correct fdt data in
> >> FreeBSD ubldr from u-boot. Would you please advice what is the best way
> >> to fix this?
> >>
> >> In file common/env_common.c:
> >> const uchar *env_get_addr(int index)
> >> {
> >> //      if (gd->env_valid)
> >> //              return (uchar *)(gd->env_addr + index);
> >> //      else
> >>                   return &default_environment[index];
> >> }
> >>
> > Assuming that you checked that your environment is valid you might be
> > facing the fact that the gd pointer is corrupted. gd is a pointer to the
> > "global data" and used for storing globals which are available before
> > and after relocation. On (32bit) ARM this value used to be stored in
> > register r8 but moved to r9 (llvm cannot reserve an arbitrary register,
> > but can reserve r9 for platform specific usage). If ubldr uses r9 you
> > end up with a invalid gd pointer when calling back into u-boot. ubldr
> > now reserves r8 and r9 so a recent version should work fine on an older
> > U-boot as well as current master.
> >
> > Can you check the latest ubldr?
> 
> Hi Jeroen,
> 
> Thanks for your response.
> 
> 1) Today, I tested ubldr in the snapshot build 
> FreeBSD-11.0-CURRENT-arm-armv6-BEAGLEBONE-20140428-r265054.img.bz2
> without commenting out those 3 lines, I still can not get "fdt ls" in 
> ubldr command line.
> 
> After commenting out those 3 lines and rebuild u-boot, I can get "fdt 
> ls" in ubldr command line.
> 
> Note: All my previous test was based on 
> FreeBSD-11.0-CURRENT-arm-armv6-BEAGLEBONE-20140323-r263665.img.bz2
> 
> 2) Would you please point to me which revision that reserves both r8 and r9?
> 
> 
> Thanks.
> 
> 
> > Regards,
> > Jeroen
> >
> >
> >
> 

I think you should be debugging this from the ubldr side.  If 'fdt ls'
fails to load the dtb when the u-boot global env is present, then what
we need to know is what result is being returned when ubldr looks up the
fdtaddr and/or fdt_addr variables.  Does it get an address for the dtb
at all?  Is the address invalid?  Is it valid but it points to something
that doesn't look like a valid dtb header?

-- Ian

[U-Boot] Latest u-boot release on BeagleBone Black for FreeBSD

[Xuebing Wang]

On 05/04/2014 07:33 PM, Jeroen Hofstee wrote:
> Hello Xuebing, (freebsd-arm added on cc),
>
> On di, 2014-04-08 at 16:52 +0800, Xuebing Wang wrote:
>> Hi u-boot community,
>>
>> I am trying to port u-boot (release u-boot-2014.04-rc3.tar.bz2) to
>> FreeBSD on BeagleBone Black.
>>
>> In FreeBSD, there is a u-boot loader (named ubldr), which can call
>> u-boot API to get fdt (Flat Device Tree) data.
>>
>> I have to comment out below 3 lines, in order to get correct fdt data in
>> FreeBSD ubldr from u-boot. Would you please advice what is the best way
>> to fix this?
>>
>> In file common/env_common.c:
>> const uchar *env_get_addr(int index)
>> {
>> //      if (gd->env_valid)
>> //              return (uchar *)(gd->env_addr + index);
>> //      else
>>                   return &default_environment[index];
>> }
>>
> Assuming that you checked that your environment is valid you might be
> facing the fact that the gd pointer is corrupted. gd is a pointer to the
> "global data" and used for storing globals which are available before
> and after relocation. On (32bit) ARM this value used to be stored in
> register r8 but moved to r9 (llvm cannot reserve an arbitrary register,
> but can reserve r9 for platform specific usage). If ubldr uses r9 you
> end up with a invalid gd pointer when calling back into u-boot. ubldr
> now reserves r8 and r9 so a recent version should work fine on an older
> U-boot as well as current master.
>
> Can you check the latest ubldr?

Hi Jeroen,

Thanks for your response.

1) Today, I tested ubldr in the snapshot build 
FreeBSD-11.0-CURRENT-arm-armv6-BEAGLEBONE-20140428-r265054.img.bz2
without commenting out those 3 lines, I still can not get "fdt ls" in 
ubldr command line.

After commenting out those 3 lines and rebuild u-boot, I can get "fdt 
ls" in ubldr command line.

Note: All my previous test was based on 
FreeBSD-11.0-CURRENT-arm-armv6-BEAGLEBONE-20140323-r263665.img.bz2

2) Would you please point to me which revision that reserves both r8 and r9?


Thanks.


> Regards,
> Jeroen
>
>
>

-- 
Thanks,
Xuebing Wang

[U-Boot] Latest u-boot release on BeagleBone Black for FreeBSD

[Jeroen Hofstee]

Hello Xuebing, (freebsd-arm added on cc),

On di, 2014-04-08 at 16:52 +0800, Xuebing Wang wrote:
> Hi u-boot community,
> 
> I am trying to port u-boot (release u-boot-2014.04-rc3.tar.bz2) to 
> FreeBSD on BeagleBone Black.
> 
> In FreeBSD, there is a u-boot loader (named ubldr), which can call 
> u-boot API to get fdt (Flat Device Tree) data.
> 
> I have to comment out below 3 lines, in order to get correct fdt data in 
> FreeBSD ubldr from u-boot. Would you please advice what is the best way 
> to fix this?
> 
> In file common/env_common.c:
> const uchar *env_get_addr(int index)
> {
> //      if (gd->env_valid)
> //              return (uchar *)(gd->env_addr + index);
> //      else
>                  return &default_environment[index];
> }
> 

Assuming that you checked that your environment is valid you might be
facing the fact that the gd pointer is corrupted. gd is a pointer to the
"global data" and used for storing globals which are available before
and after relocation. On (32bit) ARM this value used to be stored in
register r8 but moved to r9 (llvm cannot reserve an arbitrary register,
but can reserve r9 for platform specific usage). If ubldr uses r9 you
end up with a invalid gd pointer when calling back into u-boot. ubldr
now reserves r8 and r9 so a recent version should work fine on an older
U-boot as well as current master.

Can you check the latest ubldr?

Regards,
Jeroen

[U-Boot] Latest u-boot release on BeagleBone Black for FreeBSD

[Xuebing Wang]

Hi u-boot community,

I am trying to port u-boot (release u-boot-2014.04-rc3.tar.bz2) to 
FreeBSD on BeagleBone Black.

In FreeBSD, there is a u-boot loader (named ubldr), which can call 
u-boot API to get fdt (Flat Device Tree) data.

I have to comment out below 3 lines, in order to get correct fdt data in 
FreeBSD ubldr from u-boot. Would you please advice what is the best way 
to fix this?

In file common/env_common.c:
const uchar *env_get_addr(int index)
{
//      if (gd->env_valid)
//              return (uchar *)(gd->env_addr + index);
//      else
                 return &default_environment[index];
}

-- 
Thanks,
Xuebing Wang