Porting barebox to Novena: misc questions

[Sascha Hauer]

On Fri, Mar 14, 2014 at 11:35:12AM +0800, Sean Cross wrote:
> On 14/3/14 4:27 AM, Sascha Hauer wrote:
> > On Thu, Mar 13, 2014 at 06:18:44PM +0800, Sean Cross wrote:
> >> On 13/3/14 3:38 PM, Sascha Hauer wrote:
> >>> Hi Sean,
> >>>
> >>> On Thu, Mar 13, 2014 at 10:04:35AM +0800, Sean Cross wrote:
> >>>> The "usb" command just hangs the system.  It should at least detect the
> >>>> other ASIX Ethernet port.  I have &usbh1 configured identically to
> >>>> sabrelite.  Is there something else I need to configure?
> >>>
> >>> This usually means the phys are not configured correctly. Are you
> >>> probing from devicetree?
> >>
> >> I am probing from devicetree.  I don't see phys mentioned anywhere
> >> except in the imx6qdl.dtsi file, which means that in theory there
> >> shouldn't be anything for me to modify.
> > 
> > I can try to reproduce this tomorrow on a sabrelite board.
> > 
> >> Board: Kosagi i.MX6DL Novena Board
> >> detected i.MX6 DualLite revision 1.1
> >> ERROR: out of memory
> >>
> >> [<5062e9ed>] (unwind_backtrace+0x1/0x74) from [<5061c525>] (panic+0x1d/0x34)
> >> [<5061c525>] (panic+0x1d/0x34) from [<5061cc27>] (xmemalign+0xf/0x14)
> >> [<5061cc27>] (xmemalign+0xf/0x14) from [<5062f687>] (mmu_init+0x16b/0x1f8)
> >> [<5062f687>] (mmu_init+0x16b/0x1f8) from [<50600797>]
> >> (start_barebox+0x1b/0xd0)
> >> [<50600797>] (start_barebox+0x1b/0xd0) from [<5062efcd>] (__start+0x91/0xa4)
> >> [<5062efcd>] (__start+0x91/0xa4) from [<50600005>]
> >> (__bare_init_start+0x1/0xc)
> >> ### ERROR ### Please RESET the board ###
> >>
> >> If I also enable early MMU, it hangs much much earlier.  When I dug into
> >> it, it looked like it was trying to place a TTB just outside of the
> >> allocated memory region, for some reason.
> >>
> >> This is with me passing SZ_1G to barebox_arm_entry().  If I pass
> >> something small like SZ_64M, it hangs completely.  If I instead pass it
> >> SZ_128M, it works just fine, but of course the MMU still doesn't work.
> > 
> > I suspect you have some memory setting that doesn't work in your
> > .config. Could you post it? Particularly the addresses in your backtrace
> > are outside the reachable memory (0x10000000 + 1GiB = 0x50000000). This
> > should not happen. Do you have CONFIG_RELOCATABLE enabled? If not, make
> > sure CONFIG_ARCH_TEXT_BASE is inside the memory you pass to
> > barebox_arm_entry(). Otherwise the autodetection for usable malloc area
> > space might produce wrong results.
> 
> You're on the right track.  Earlier on I started calling
> barebox_arm_entry(0x10000000 + 0x800000, SZ_1G, fdt), because otherwise
> it would fail in mysterious ways.  Now that I've got ram passing working
> correctly, I've set it back to 0x10000000, and I'm able to get further.
>  I've also increased the MALLOC_SIZE from 8MB to 64MB, which has made
> some out-of-memory errors go away.
> 
> Now, it gets much further, even with CONFIG_MMU_EARLY enabled.  Now it
> crashes with an error that I see SabreSD was getting at one point, but
> I'm not sure how they resolved it:
> 
> Board: Kosagi i.MX6DL Novena Board
> detected i.MX6 DualLite revision 1.1
> BUG: failure at arch/arm/cpu/mmu.c:122/find_pte()!
> BUG!

This can mean that dma_alloc_coherent() is called too early, before
mmu_init(). That's quite unlikely since the chipidea probe is in
device_initcall, way after mmu_initcall. Other possibility is that
mmu_init failed due to missing memory banks.

Please apply "ARM: MMU: Add some debugging aids and hints" I just sent
to the list. It adds more sanity checks and hopefully gives more clue
what is wrong.

> CONFIG_PBL_IMAGE=y
> CONFIG_PBL_MULTI_IMAGES=y
> CONFIG_PBL_RELOCATABLE=y
> CONFIG_IMAGE_COMPRESSION=y
> # CONFIG_IMAGE_COMPRESSION_LZ4 is not set
> CONFIG_IMAGE_COMPRESSION_LZO=y
> # CONFIG_IMAGE_COMPRESSION_GZIP is not set
> # CONFIG_IMAGE_COMPRESSION_NONE is not set
> CONFIG_MMU=y
> CONFIG_MMU_EARLY=y
> CONFIG_HAVE_CONFIGURABLE_TEXT_BASE=y
> CONFIG_TEXT_BASE=0x17800000
> CONFIG_BAREBOX_MAX_PBL_SIZE=0xffffffff
> CONFIG_BAREBOX_MAX_BARE_INIT_SIZE=0xffffffff
> CONFIG_STACK_SIZE=0x400000

This is too much 0x8000 should be enough.

> CONFIG_MALLOC_SIZE=0x4000000

You can also set this to 0x0 in which case barebox will pick a memory
region inside available SDRAM.

Your stack size / malloc size settings shouldn't make problems though.

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 |