Beaglebone white: Kernel doesn't start with barebox-2014.03.0

[Lucas Stach]

Am Mittwoch, den 16.04.2014, 22:02 +0200 schrieb Uwe Kleine-König:
> On Wed, Apr 16, 2014 at 05:35:52PM +0200, Lucas Stach wrote:
> > Hi Rolf,
> > 
> > Am Dienstag, den 15.04.2014, 13:28 +0200 schrieb Rolf Evers-Fischer:
> > > Dear all,
> > > after update to barebox-2014.03.0, the kernel on my beaglebone-white doesn't
> > > start anymore.
> > > The last line, which appears, is "booting kernel with devicetree".
> > > I discovered that the problem is caused by the commit [1]. When I revert this
> > > commit, everything works fine.
> > > 
> > > Here is some more info, which I gathered with additional printfs:
> > > Without this commit, the '__do_bootm_linux()' function is called with
> > >  data->initrd_address = 0xffffffff
> > >  data->os_res->start = 0x80008000
> > >  data->os_entry = 0
> > >  swap = 0
> > > Finally it invokes the 'start_linux()' function with
> > >  kernel = 0x80008000
> > >  swap = 0
> > >  initrd_start = 0x80808000
> > >  initrd_size = 0
> > >  data->oftree = 0x8b1e0000
> > > 
> > > With this commit (i.e. when the kernel doesn't start), the '__do_bootm_linux()'
> > > function is called with the same parameters as above plus
> > >  free_mem = 0x8041f000
> > > It invokes the 'start_linux()' function with
> > >  kernel = 0x80008000
> > >  swap = 0
> > >  initrd_start = 0x8041f000
> > >  initrd_size = 0
> > >  data->oftree = 0x8041f000
> > > 
> > > Do you have any idea, how to fix this problem? For now I will continue using
> > > barebox-2014.03.0 without the problematic commit.
> > > 
> > > [1] http://www.spinics.net/lists/u-boot-v2/msg17345.html
> > >     "ARM: bootm: pass free memory to __do_bootm_linux"
> > 
> > I think you are trying to boot a uImage. Is this right?
> > 
> > Could you please test with a zImage? If it still fails, can you try the
> > following patch?
> > 
> > --------------------------------->8------------------------------------------
> > From 6e51388bda025508b28d723960055656fa50784d Mon Sep 17 00:00:00 2001
> > From: Lucas Stach <l.stach at pengutronix.de>
> > Date: Wed, 16 Apr 2014 17:23:33 +0200
> > Subject: [PATCH] ARM: bootm: be more clever while deciding where to put zImage
> > 
> > For small systems we would put the zImage at 8MiB after
> > the start of memory, and put the DT a bit after the zImage.
> > When we encounter an image which is bigger than 8MiB
> > uncompressed, the kernel would try to relocate itself
> > and overwrite the DT.
> > 
> > Try to be more clever at zImage placement to avoid
> > triggering the kernel relocation.
> > 
> > Signed-off-by: Lucas Stach <l.stach at pengutronix.de>
> > ---
> >  arch/arm/lib/bootm.c | 31 ++++++++++++++-----------------
> >  1 file changed, 14 insertions(+), 17 deletions(-)
> > 
> > diff --git a/arch/arm/lib/bootm.c b/arch/arm/lib/bootm.c
> > index c63cbc02e35f..2d4b18b7f794 100644
> > --- a/arch/arm/lib/bootm.c
> > +++ b/arch/arm/lib/bootm.c
> > @@ -245,23 +245,6 @@ static int do_bootz_linux(struct image_data *data)
> >  	if (ret)
> >  		return ret;
> >  
> > -	if (load_address == UIMAGE_INVALID_ADDRESS) {
> > -		/*
> > -		 * The kernel should stay in the first 128MiB of RAM, recommended
> > -		 * is 32MiB into RAM so that relocation prior to decompression
> > -		 * can be avoided.
> > -		 */
> > -		if (mem_size > SZ_64M)
> > -			data->os_address = mem_start + SZ_32M;
> > -		else
> > -			data->os_address = mem_start + SZ_8M;
> > -
> > -		load_address = data->os_address;
> > -		if (bootm_verbose(data))
> > -			printf("no os load address, defaulting to 0x%08lx\n",
> > -				load_address);
> > -	}
> > -
> >  	fd = open(data->os_file, O_RDONLY);
> >  	if (fd < 0) {
> >  		perror("open");
> > @@ -292,6 +275,20 @@ static int do_bootz_linux(struct image_data *data)
> >  	if (swap)
> >  		end = swab32(end);
> >  
> > +	if (load_address == UIMAGE_INVALID_ADDRESS) {
> > +		/*
> > +		 * Just use a conservative default of 4 times the size of the
> > +		 * compressed image, to avoid the need for the kernel to reocate itself
> s/reocate/relocate/
> 
Thanks, will fix this.

> > +		 * before decompression.
> > +		 */
> > +		data->os_address = mem_start + PAGE_ALIGN(end * 4);
> I'm not sure if that really matters, but end isn't necessarily[1] holding
> the size of the compressed image. Even if it doesn't matter, using
> 
> +		data->os_address = mem_start + PAGE_ALIGN((end - start) * 4);
> 
> is more clear here.
> 
Right, I've also noticed this. But this error is is spread in a few more
places, so I'll cook another patch for this.

Regards,
Lucas