[U-Boot] [BUG] booting on the Tinker Board (tinker-rk3288_defconfig) is broken

[Tom Rini <trini@konsulko.com>]

On Wed, Feb 13, 2019 at 09:13:31PM +0100, Heinrich Schuchardt wrote:
> 
> 
> On 2/13/19 2:24 PM, Tom Rini wrote:
> > On Wed, Feb 13, 2019 at 02:16:12PM +0100, Heinrich Schuchardt wrote:
> >>
> >>
> >> On 2/13/19 1:25 PM, Tom Rini wrote:
> >>> On Wed, Feb 13, 2019 at 11:13:19AM +0100, Philipp Tomsich wrote:
> >>>>
> >>>>
> >>>>> On 13.02.2019, at 11:10, David Wu <david.wu@rock-chips.com> wrote:
> >>>>>
> >>>>> Hi Philipp,
> >>>>>
> >>>>> 在 2019/2/12 下午9:54, Philipp Tomsich 写道:
> >>>>>>> On 12.02.2019, at 13:38, David Wu <david.wu@rock-chips.com> wrote:
> >>>>>>>
> >>>>>>> Hi Philipp,
> >>>>>>>
> >>>>>>> 在 2019/2/10 上午2:24, Philipp Tomsich 写道:
> >>>>>>>> That said, I have been fighting issues from this patchset when trying to read
> >>>>>>>> from devices on the i2c0 bus on a RK3399—which had me add a “simple”
> >>>>>>>> pinctrl device again, as pulling all the dependencies into the SPL is a pain.
> >>>>>>>
> >>>>>>> I think you can also add the simple ops if necessary. Full pinctrl
> >>>>>>> should only be used at the U-boot stage, or the SPL stage where the TPL is available.
> >>>>>>>
> >>>>>>> For most boards about 3288, there is no need to enable pinctrl at SPL
> >>>>>>> stage, such as Tinker board.
> >>>>>> Could you check whether TPL_MAX_SIZE and SPL_MAX_SIZE are correct for
> >>>>>> the RK3288? If not, please submit a patch.
> >>>>>
> >>>>> It is correct, 32KB.
> >>>>
> >>>> Interesting...
> >>>> I wonder why the build did not fail then, if SPL became too big.
> >>>
> >>> This is probably the "we need BOARD_SIZE_CHECK used more widely too"
> >>> issue that's come up in other threads.  The linker-only check does not
> >>> catch the CONFIG_OF_SEPARATE dtb exceeds limit problem as it's
> >>> post-link.
> >>>
> >>
> >> This is the error message I currently get for U-Boot 2019.04 when
> >> building for the Tinker Board:
> >>
> >> mkdir -p foo/usr/lib/u-boot/tinker/
> >> denx/tools/mkimage -n rk3288 -T rksd -d denx/spl/u-boot-spl-dtb.bin \
> >>   foo/usr/lib/u-boot/tinker/u-boot.img
> >> Warning: SPL image is too large (size 0x8800) and will not boot
> >> Error: image verification failed
> >> cat denx/u-boot-dtb.bin >> foo/usr/lib/u-boot/tinker/u-boot.img
> >> cp sd_fusing.sh foo/usr/lib/u-boot/tinker/
> >>
> >> @Tom
> >> Shouldn't the mkimage step be included in Travis CI testing?
> > 
> > No, mkimage fails for almost every aarch64 target as we don't include
> > ATF or the other blobs required to make things bootable.
> > BOARD_SIZE_CHECK is where we can confirm that the U-Boot binary isn't
> > too large.
> > 
> 
> Hello Tom,
> 
> BOARD_SIZE_CHECK seems to be about the U-Boot image not about SPL. I

I was possibly being slightly too literal, sorry.  We need to adapt and
use the "check the resulting final binary is OK size-wise" logic to be
available and used everywhere.  Given that we're now at I think 3 or 4
different SoCs running into this same exact problem, we really do need
to.

> guess we need to copy this line to rk3288_common.h:
> 
> include/configs/rk3188_common.h:31:
> #define CONFIG_SPL_MAX_SIZE          (0x8000 - 0x800)
> 
> But when I insert this line I get no linker error but the size of
> denx/spl/u-boot-spl-dtb.bin is still too big.
> 
> Shouldn't count the DTB size be accounted for when doing the
> CONFIG_SPL_MAX_SIZE check?
> 
> CONFIG_SPL_MAX_SIZE      30720 bytes
> spl/u-boot-spl-nodtb.bin 28932 bytes
> spl/u-boot-spl.dtb        4096 bytes
> spl/u-boot-spl-dtb.bin   32986 bytes

SPL_MAX_SIZE is a linker check and as such will not catch post-linking
things, is the problem.  We should however make it not be a linker check
and instead leverage it with the logic behind BOARD_SIZE_CHECK, thanks!

-- 
Tom
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 819 bytes
Desc: not available
URL: <http://lists.denx.de/pipermail/u-boot/attachments/20190213/278b6c25/attachment.sig>