Re: Beaglebone black issue loading booting kernel

[Sascha Hauer <s.hauer@pengutronix.de>]

On Mon, Oct 12, 2015 at 12:27:37PM +0300, Antony Pavlov wrote:
> On Mon, 12 Oct 2015 10:24:17 +0200
> Sascha Hauer <s.hauer@pengutronix.de> wrote:
> 
> > Hi Mathieu,
> > 
> > On Sun, Oct 11, 2015 at 06:51:06AM -0400, Mathieu Tétreault wrote:
> > > I compiled barebox to run on the beaglebone black using the
> > > am335x_mlo_defconfig and am335x_defonfig which gaves me the
> > > barebox-am33xx-beaglebone.img(barebox.bin) and
> > > barebox-am33xx-beaglebone-mlo.img(MLO).
> > > 
> > > This part works fine since I'm able to get in barebox and the board is
> > > detected as beaglebone black
> > > 
> > > === log ===
> > > barebox 2015.10.0-00084-g54bf386 #2 Fri Oct 9 15:18:17 EDT 2015
> > > 
> > > 
> > > Board: TI AM335x BeagleBone black
> > > detected 'BeagleBone Black'
> > > =========
> > > 
> > > If I let the autoboot start it always ends up looking for a NFS server
> > > which I don't have right now.
> > > 
> > > What I have right now is an SD card with a rootfs generated by Yocto.
> > > This rootfs is able to boot using the u-boot binary that yocto
> > > compiled.
> > > 
> > > When I use:
> > > ===
> > > barebox@TI AM335x BeagleBone black:/ global.bootm.image=/boot/uImage
> > > barebox@TI AM335x BeagleBone black:/ global.bootm.oftree=/boot/boneblack.dtb
> > > barebox@TI AM335x BeagleBone black:/
> > > global.linux.bootargs.dyn.root="root=/dev/mmcblk0p2 rootfstype=ext4
> > > rootwait"
> > > ===
> > > 
> > > I got the following log
> > > ===
> > > barebox@TI AM335x BeagleBone black:/ bootm
> > >    Image Name:   Linux-3.14.19-yocto-standard
> > >    OS:           Linux
> > >    Architecture: ARM
> > >    Type:         Kernel Image
> > >    Compression:  uncompressed
> > >    Data Size:    5068000 Bytes = 4.8 MiB
> > >    Load Address: 80008000
> > >    Entry Point:  80008000
> > 
> > It seems we stumbled upon another bad image placement. What happens here
> > is:
> > 
> > - The uImage forces the load address to 0x80008000.
> > - We put the device tree binary somewhere behind that
> > - The Kernel uncompresses itself and overwrites the device tree
> > 
> > You have some options to work around this issue. You can use a zImage
> > rather than a uImage. That's my recommended approach. barebox will put
> > the zImage to a suitable place. If you insist on uImage you could also
> > specify the LOADADDR to 0xffffffff. If barebox finds this address it
> > picks a suitable address itself. The image won't work with U-Boot
> > though. The LOADADDR must be passed to the kernel during build. I don't
> > know though how Yocto handles this and how it could be overwritten.
> > 
> > Note that 0x80008000 is a bad choice anyway. This is exactly the place
> > where the kernel self extractor will place the final kernel. To do so it
> > must first move the compressed binary to another place, because
> > otherwise it would overwrite itself.
> > 
> > I would love to fix these load address issues, but unfortunately this
> > area is full of historic cruft where fixing one issue just raises
> > another one :(
> 
> I see that ARM barebox suffering from kexec support absence.
> IMHO I have to revive my 'kexec support for MIPS' series.
> This series contain special post-barebox relocator
> so kernel load memory range can harmlessly overlap barebox memory ranges.

This is not a real issue on ARM. I mean it's currently not supported
that the kernel overlaps barebox memory regions, but we can work around
it quite well.

The issue on ARM is the various constraints where to place images some
of which stated explicitly somewhere, others documented nowhere. Some of
the constraints are:

- With CONFIG_AUTO_ZRELADDR=y the zImage has to be placed in the first
  128MiB of lowmem. How do we know we are about to load a kernel with
  this option enabled?

- Documentation/arm/Booting states that:

 	The dtb must be placed in a region of memory where the kernel
	decompressor will not overwrite it, whilst remaining within the
	region which will be covered by the kernel's low-memory mapping.

  Aha. How do we know a place where the kernel decompressor will not
  overwrite it? We can only guess that, so at some point we used to move
  the dtb up in RAM, sometimes so high that we trapped into the "within
  the region which be covered by the kernel's low-memory mapping"
  constraint. The recommendation is:

	A safe location is just above the 128MiB boundary from start of
	RAM.

  Apparently systems with less than 256MiB of memory are no longer
  interesting.

- With CONFIG_AUTO_ZRELADDR=y the kernel could be placed anywhere in the
  128MiB in RAM, but unfortunately the u-boot uImage format does not
  allow to leave the load address unspecified. So while the kernel
  supports runtime determined SDRAM addresses, the uImage format does
  not and makes this feature unusable with uImages.

IMO the Kernel decompressor overwriting the dtb is a bug that should be
fixed. In the bootloader we have no way to know where the kernel will
decompress itself and I think it should not have to know. The kernel
should just move the dtb somewhere else if it's not happy with the
current place. While this is really annoying I haven't been annoyed
enough to come up with a patch :(

Sascha

-- 
Pengutronix e.K.                           |                             |
Industrial Linux Solutions                 | http://www.pengutronix.de/  |
Peiner Str. 6-8, 31137 Hildesheim, Germany | Phone: +49-5121-206917-0    |
Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-5555 |

_______________________________________________
barebox mailing list
barebox@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/barebox