[maemo-leste] Broken u-boot on n900 hw (possibly CONFIG_POSITION_INDEPENDENT)

[Pali Rohár]

On Sunday 20 November 2022 10:33:35 Tom Rini wrote:
> On Sat, Nov 19, 2022 at 10:13:01PM +0100, Pali Rohár wrote:
> > On Wednesday 09 November 2022 21:33:28 Pali Rohár wrote:
> > > On Wednesday 02 November 2022 01:08:26 Pali Rohár wrote:
> > > > On Friday 21 October 2022 21:09:55 Pali Rohár wrote:
> > > > > On Thursday 20 October 2022 22:06:58 Pali Rohár wrote:
> > > > > > On Monday 10 October 2022 23:30:21 Sicelo wrote:
> > > > > > > On Sun, Oct 09, 2022 at 09:33:17PM +0200, Pali Rohár wrote:
> > > > > > > > Hello!
> > > > > > > > 
> > > > > > > > Current u-boot from master branch does not work on n900 hw. It does not
> > > > > > > > show anything on display, there just nokia logo.
> > > > > > > > 
> > > > > > > > In qemu it is working fine and all tests are passing. Also in qemu is
> > > > > > > > display working without any problem.
> > > > > > > > 
> > > > > > > > Could you help me to debug it? Maybe something is printed to serial
> > > > > > > > console? Or maybe serial console could be used for printf-debugging?
> > > > > > > > 
> > > > > > > > I just figured out that disabling CONFIG_POSITION_INDEPENDENT u-boot
> > > > > > > > option make it working again. But it requires to restoring older version
> > > > > > > > of lowlevel_init.S file which does not depend on POSITION_INDEPENDENT:
> > > > > > > > 
> > > > > > > >     git checkout 7c4ad9821758f98db127e9d6864671c906c02d3b -- board/nokia/rx51/lowlevel_init.S
> > > > > > > > 
> > > > > > > > And also setting CONFIG_SYS_TEXT_BASE=0x80008000 (to some value in RAM).
> > > > > > > 
> > > > > > > Hi
> > > > > > > 
> > > > > > > I have a really crude serial console setup in place now, but very little
> > > > > > > time, and even less experience with u-boot. I should be able to do the
> > > > > > > tests for you on Saturday night (UTC+2). Perhaps over IRC might be a
> > > > > > > good idea, in case I need some hand-holding, and if you still need the
> > > > > > > help by then.
> > > > > > > 
> > > > > > > Regards
> > > > > > > Sicelo
> > > > > > 
> > > > > > Well, I do not know where to start or how to instruct somebody who is
> > > > > > not experienced with u-boot debugging. I think that the first step
> > > > > > should be to look if U-Boot prints something on the serial console or
> > > > > > not. If it prints then try to find via printf-driven-debugging place
> > > > > > where it stuck. If it does not print anything on the console then try to
> > > > > > setup debug early console, check if it prints something now and do same
> > > > > > type of debugging. And if even early debug console does not work then
> > > > > > probably initialize serial console directly in lowlevel_init.S file,
> > > > > > prints something to serial, then move code for printing to later stage
> > > > > > and try to figure out what is broken...
> > > > > 
> > > > > Early debug uart console can be enabled by configs/nokia_rx51_defconfig options:
> > > > > 
> > > > > CONFIG_DEBUG_UART=y
> > > > > CONFIG_DEBUG_UART_OMAP=y
> > > > > CONFIG_DEBUG_UART_SHIFT=2
> > > > > CONFIG_DEBUG_UART_BASE=0x49020000
> > > > > CONFIG_DEBUG_UART_CLOCK=48000000
> > > > > 
> > > > > And all debug verbose logs can be enabled by configs/nokia_rx51_defconfig options:
> > > > > 
> > > > > CONFIG_DM_DEBUG=y
> > > > > CONFIG_LOG=y
> > > > > CONFIG_LOGLEVEL=9
> > > > > CONFIG_LOG_MAX_LEVEL=9
> > > > > CONFIG_LOG_DEFAULT_LEVEL=9
> > > > > 
> > > > > By compiling via 'make u-boot.bin KCPPFLAGS=-DLOG_DEBUG'
> > > > 
> > > > Can somebody with serial console provide this debug output if there is
> > > > something useful?
> > > 
> > > Can somebody help me with this?
> > 
> > Merlijn Wajer now sent me output from serial console:
> > 
> >   ...
> >   initcall: 0001ebd4 (relocated to 8fe0abd4)
> >   MMC:   0
> >      - 0 'omap_hsmmc'
> >      - found
> >   1
> >      - 0 'omap_hsmmc'
> >      - 1 'omap_hsmmc'
> >      - found
> >   2
> >      - 0 'omap_hsmmc'
> >      - 1 'omap_hsmmc'
> >      - not found
> >   OMAP SD/MMC: 0, OMAP SD/MMC: 1
> >   initcall: 0001ea8c (relocated to 8fe0aa8c)
> >   Loading Environment from <NULL>... Using default environment
> >   Destroy Hash Table: 8fe25a98 table = 00000000
> >   Create Hash Table: N=387
> > 
> > After the "Create Hash Table: N=387" line there is nothing more.
> > Repeated tests have same result, always stop at that line.
> > 
> > In qemu after that line I see following output:
> > 
> >   INSERT: table 8fe25a98, filled 1/389 rv 8fd0e2f4 ==> name="bootcmd" value="run sdboot;run emmcboot;run attachboot;echo"
> >   INSERT: table 8fe25a98, filled 2/389 rv 8fd0e344 ==> name="bootdelay" value="30"
> >   INSERT: table 8fe25a98, filled 3/389 rv 8fd0f960 ==> name="baudrate" value="115200"
> >   DELETE CANDIDATE: "preboot"
> >   hdelete: DELETE key "preboot"
> >   DELETE ERROR ##############################
> >   INSERT: table 8fe25a98, filled 4/389 rv 8fd0e574 ==> name="loadaddr" value="0x80008000"
> >   ...
> > 
> > In qemu is u-boot working fine.
> > 
> > So it looks like that there is u-boot crash in environment code and only
> > when running on real HW.
> > 
> > Any idea? Could there be some relocation issue in u-boot env code?
> 
> My first thought here is that something like 534f0fbd6520 ("arm64: Fix
> relocation of env_addr if POSITION_INDEPENDENT=y") might be needed on
> 32bit as well?
> 
> -- 
> Tom

Thank you Tom! This is really the issue. After I applied same fix for
32-bit ARM, U-Boot on real N900 HW booted. I will send a patch for it.

Interesting is that same u-boot binary is working fine in qemu N900. And
it works with and also without fix for env_addr.