STM32F767 freeze on chibios update Topic is solved

[tsm]

Hi all,

I've been searching for related posts but to no avail. If I missed anything please direct me there.

I'm rebooting a project, started by someone else, where chibios is at the following revision:

Code: Select all

commit 1d3bbc72c326e87bb688cbce39ea93b71b4406d3 (origin/master, origin/HEAD)
Author: Giovanni Di Sirio <[email protected]>
Date:   Fri Mar 23 14:46:11 2018 +0000

    Restored launch files.
    
    git-svn-id: svn://svn.code.sf.net/p/chibios/svn/trunk@11850 110e8d01-0319-4d1e-a829-52ad28d1bb01

I updated chibios to ver-2019.1.3 as well as my chconf.h file (so that RT version went from 5 to 6) and found that my board freezes. When pausing with a debugger I see that it is always stopped at the following line in chSysInit:

Code: Select all

#if CH_DBG_STATISTICS == TRUE
  /* Starting measurement for this thread.*/
  chTMStartMeasurementX(&currp->stats);
#endif

but I'm not sure if this is just an artifact of the debugger.
After a bit of searching (and git blame) I seem to conclude that the cause is updating RT to version 6 (_CHIBIOS_RT_CONF_VER_6_0_) on this commit:

Code: Select all

commit 26db32042bd7a2437246765ccf9157dd17ec824b
Author: Giovanni Di Sirio <[email protected]>
Date:   Sat Jun 16 13:55:17 2018 +0000

    Moved RT version to 6.0, updated all chconf.h files.
    
    git-svn-id: svn://svn.code.sf.net/p/chibios/svn/trunk@12106 110e8d01-0319-4d1e-a829-52ad28d1bb01

By testing with it's previous commit (and RT version 5 on my chconf) all works well. I've compared my chconf with the one provided in the demos and the only difference I see is that all definitions are guarded by an #if !defined(...) (which I tried as well) and some CH_DBG_ flags that were setup already.

Do I need to update any other file? mcuconf or halconf?

Thank you

[Giovanni]

Hi,

In general you should take all configuration files AND makefiles from the version you are using.

Your description does not give ma an hint about what the problem could be. That function is the calibration for time measurement, you could disable time measurement in chconf.h but that would be a workaround not a solution.

Could you try the latest officially release one? 19.1.3. Taking the latest subversion trunk code would be fine too.

Giovanni

[tsm]

Hi Giovanni,

thanks for helping.

I forgot to say that on openocd's trace I get the following:

Info : Successfully loaded memory map of ChibiOS/RT target running version 6.0.3
Error: ChibiOS registry integrity check failed, double linked list violation
Info : Only showing current execution because of a broken ChibiOS thread registry

I'll review the makefiles and configuration as suggested and comeback with more information.

Thank You

EDIT:


I've tested with 19.1.3 as well (in the following revision). That was my first test before bisecting the logs.

Code: Select all

commit 313416b8fda90d9973a749a0a35970956852c286 (HEAD, tag: ver19.1.3)
Merge: bbc61c3df 14e10a2e8
Author: Giovanni Di Sirio <[email protected]>
Date:   Sun Jul 14 07:26:31 2019 +0000

    git-svn-id: svn://svn.code.sf.net/p/chibios/svn/tags/ver19.1.3@12914 27425a3e-05d8-49a3-a47f-9c15f0e5edd8

[tsm]

I've updated my chconf and nothing changed. By comparing makefiles I also don't see any changes besides the original ones that were made on my project. I have also reduced my main to a LED blinker but the problem persists.

By disabling all the debug flags and pausing execution with the debugger the thred stops at:

Code: Select all

Reset_Handler:
         b          _crt0_entry

[Giovanni]

It should never stop there unless crt0_xxx.S is missing. I suggest you to review your makefile, in case of doubt just try to build one of the existing demos and verify functionality.

Giovanni

[tsm]

hm I compared makefiles and haven't seen any relevant difference yet. When disassembling the .elf file I can see the crt0_entry:

Code: Select all

00200210 <_crt0_entry>:
  200210:       b672            cpsid   i
  200212:       4839            ldr     r0, [pc, #228]  ; (2002f8 <endfiniloop+0x6>)
  200214:       f380 8808       msr     MSP, r0
  200218:       4838            ldr     r0, [pc, #224]  ; (2002fc <endfiniloop+0xa>)
  20021a:       f380 8809       msr     PSP, r0
(...)

I noticed when pausing and then resuming with the debugger the debugging session stops with:

Code: Select all

Info : The target is not running when halt was requested, stopping GDB.
Warn : target stm32f7x.cpu is not halted (gdb fileio)
Info : Successfully loaded memory map of ChibiOS/RT target running version 6.0.3
Error: ChibiOS registry integrity check failed, double linked list violation
Info : Only showing current execution because of a broken ChibiOS thread registry.
Info : dropped 'gdb' connection

What does the following mean?

Code: Select all

Error: ChibiOS registry integrity check failed, double linked list violation
Info : Only showing current execution because of a broken ChibiOS thread registry.

[Giovanni]

That message about the broken list probably is caused by the OS not yet initialized at that point.

Try compiling using -O0 and disabling LTO, you need to understand exactly where it fails. Also enable assertions and state checker in chconf.h.

Giovanni

[tsm]

That message about the broken list probably is caused by the OS not yet initialized at that point.

Try compiling using -O0 and disabling LTO, you need to understand exactly where it fails. Also enable assertions and state checker in chconf.h.

Giovanni

I've tried with without LTO and O0 but no luck. After updating makefiles, I generated compile_commands.json and noticed the following differences on the non-working branch:

- I'm compilling an extra file chibios/os/hal/ports/STM32/LLD/EXTIv1
- I had -mthumb mode enabled. I've restored this which entailed removing -mthumb and adding -mno-thumb-interwork -DTHUMB_NO_INTERWORKING. But still the same behaviour

I've reduced my main() to simply blinking some LEDS and started removing cpp files until things work. Odd enough it works without two files: one class which is a thin wrapper around SerialDriver and a Syslog class that sends some data through TCP (using lwip). This syslog uses chThdCreateStatic with a stack size of 512, but I'm a bit lost to why it works this way

[Giovanni]

Those things are out of my visibility, can't really help with that.

My generic advice is to remove everything then re-introduce things one at time.

Giovanni

[rew]

- I'm compilling an extra file chibios/os/hal/ports/STM32/LLD/EXTIv1
- I had -mthumb mode enabled. I've restored this which entailed removing -mthumb and adding -mno-thumb-interwork -DTHUMB_NO_INTERWORKING. But still the same behaviour

If you need the "EXTIv1 driver", (or not) you seem to be using and STM32xxxx CPU. These are "cortex-M" arm-variants. These do not have the original ARM instruction set, but only the thumb instruction set.
So... I would expect that you NEED the -mthumb option to the compiler.