• View
• Edit
• History
• Print

MaximPegasusDuktapeProblems

Below are various problems I had building Duktape 2.1 for the Maxim Pegasus.

See Maxim Pegasus for instructions about how to build from the accessors repo.

See below for possibly useful information.

Makefile Creation

I attempted to create a makefile by hand, but it seems easier to use the mbed command to create a makefile and add to that.

1. In /ptII/org/terraswarm/accessor/accessors/web/hosts/duktape:
     mbed new mbed-duktape2
     cd mbed-duktape2/
     mbed export -i GCC_ARM -m MAX32630FTHR

   [$[Get Code]]
2. Add the following to duktape/mbed-duktape2/Makefile:
   CFLAGS += -DDUK_F_32BIT_PTRS -DDUK_USE_DEBUG -DDUK_USE_DEBUG_LEVEL=0
   CXX_FLAGS += -DEDUK_RUN_RAMPJSDISPLAY $(CFLAGS)
   
   #       ../duktape2/examples/cmdline/duk_cmdline.c                                                                                                      
   C_SRCS = \
           ../duktape2/src/duktape.c \
           ../eduk2/duk_stack.c \
           ../eduk2/nofileio.c \
           ../eduk2/c_eventloop.c \
           ../eduk2/modSearch.c \
           ../duktape2/extras/print-alert/duk_print_alert.c \
           ../duktape2/extras/console/duk_console.c \
           ../duktape2/extras/logging/duk_logging.c \
           ../duktape2/extras/module-duktape/duk_module_duktape.c
   
   CPP_SRCS = \
           ../eduk2.cpp
   
   OBJECTS = $(C_SRCS:%.c=%.o)
   OBJECTS += $(CPP_SRCS:%.cpp=%.o)
   
   INCLUDE_PATHS += -I../../duktape2
   INCLUDE_PATHS += -I../../duktape2/src
   INCLUDE_PATHS += -I../../eduk2

   [$[Get Code]]
   FIXME why can't we put the above in a separate duktape.mk file and include it in the Makefile?
3. Copy the eduk2.cpp file:
   cp BUILD/mbed-duktape2.bin /Volumes/DAPLINK/

   [$[Get Code]]
4. The red LED will blink, indicating that the file is being uploaded
5. Wait for the message about "Disk Not Ejected Properly", which indicates that the processor has rebooted.
6. Start up CoolTerm, find the port, set it for 9600-8-None-1
7. Messages should appear.

Duktape Size

The issue is that the Accessor host starts up and echoes to the port, but we never see the numbers. We get:

  eduk: About to run test/auto/RampJSDisplay.js

With the default Duktape2.0.1 settings, the size is:

    text    data     bss     dec     hex filename
  437688   15364    8316  461368   70a38 BUILD/mbed-duktape2.elf

With the following changes to duktape2/duk_config.j:

Sizes:

     text    data     bss     dec     hex filename
 531736   16388    8316  556440   87d98 BUILD/mbed-duktape2.elf

Then I get quite a bit of output in the terminal and it eventually fails with:

Interestingly, it appears that enabling the debugging means that we never get to the first println in eduk.cpp. Presumably we are consuming too much memory with the debugging.

So, it appears there is a stack problem. This could be because we are running with the mbed RTOS, instead we should try to compile without it.

limits problem

Under the Maxim Pegasus (Cortex Arm 4), adding debugging fails:

  D0 duk_heap_alloc.c:621 (duk__dump_type_sizes): duk_re_compiler_ctx=196
  D0 duk_heap_alloc.c:624 (duk__dump_type_limits): limits

See Maxim Pegasus Limits Problem

Size, continued

The Makefile contains

Looking at the max3263x.ld file shows various [Ss]tack symbols.

Running nm shows that Stack_Size is 0x5000 or 20480 decimal

mbed-os/targets/TARGET_Maxim/TARGET_MAX32630/device/TOOLCHAIN_GCC_ARM/startup_max3263x.S contains:

So, let's define __STACK_SIZE! Editing the Makefile:

  ASM_FLAGS += -D__STACK_SIZE=0x00050000

The mbed RTOS presumably defines a stack size per thread. Changing this globally could be bad. There is a web page somewhere about how to change this on a per thread basis.

If you change the value of ASM_FLAGS in the Makefile, then do

  rm ./BUILD/mbed-os/targets/TARGET_Maxim/TARGET_MAX32630/device/TOOLCHAIN_GCC_ARM/startup_max3263x.o

With

  ASM_FLAGS += -D__STACK_SIZE=0x00068000

and

  #define DUK_USE_DEBUG_LEVEL 1

The failure occurs later:

Memory Usage

https://github.com/nuket/mbed-memory-status looks promising

1. Added two lines to mbed-os/targets/T>ARGET_Maxim/TARGET_MAX32630/device/TOOLCHAIN_GCC_ARM/startup_max3263x.S
   .SKIPRAMINIT:
   
   /* Perform system initialization after RAM initialization */
   ldr r0, =SystemInit
   blx r0
   
   /* https://github.com/nuket/mbed-memory-status */
   LDR R0, =fill_isr_stack_with_canary
   BLX R0
   
   /* This must be called to walk the constructor array for static C++ objects */
   /* Note: The linker file must have .data symbols for __X_array_start and __X_array_end */
   /* where X is {preinit, init, fini} */
   ldr r0, =_start
   blx r0
   

   [$[Get Code]]

Editing eduk2.cpp:

Create A Thread

https://developer.mbed.org/blog/entry/Tracing-stack-and-heap-overflow-errors/ indicates that we can create a thread and specify the stack size.

1. In eduk2.cpp, move the bulk of main() into a method called inner_main(). Remove the return statements.
2. In main():
   #include "mbed.h"
   #include "rtos.h"
   
   
   DigitalOut led1(LED1);
   DigitalOut led2(LED2);
   Thread thread;
   
   void led2_thread() {
     while (true) {
       led2 = !led2;
       Thread::wait(1000);
     }
   }
   
   int main(int argc, char *argv[]) {
     fprintf(stderr, "eduk2.cpp main() start\n");
     Thread t(osPriorityNormal, 200 * 1024);
   
     thread.start(led2_thread);
   
   
     fprintf(stderr, "eduk2.cpp main() Thread t: max_stack:%ld\n", t.max_stack());
     print_all_thread_info();
     print_heap_and_isr_stack_info();
     fprintf(stderr, "eduk2.cpp main() after print thread info\n");
   
     t.start(&inner_main);
     fprintf(stderr, "eduk2.cpp main() started duktape thread\n");
   
     while (true) {
       led1 = !led1;
       Thread::wait(500);
     }
   
   }

   [$[Get Code]]

When this runs, it creates two threads, one for Duktape, one to blink the LEDs.

It fails while loading duktape.js:

BTW - to identify the threads, subtract 1 from the entry and use nm.

1. The first thread is entry: 0001ED71
2. Use nm on the *.elf binary:
   bash-3.2$ nm ./BUILD/mbed-duktape2.elf | grep -i 1ED70
   0001ed70 T osTimerThread
   bash-3.2$

   [$[Get Code]]

Here are the other threads.

I'm not sure why entry: 0001DEB9 appears twice in the output.

BTW - Here are the sizes:

   text    data     bss     dec     hex filename
 286720   15184    8316  310220   4bbcc BUILD/mbed-duktape2.elf

In the output, the used values are 00000 because we did not update mbed-os/targets/TARGET_Maxim/TARGET_MAX32630/device/TOOLCHAIN_GCC_ARM/startup_max3263x.S (see above)

ReferenceError: 4

To deal with

  ReferenceError: 4

Edit duk_config.h:

  /* #undef DUK_USE_VERBOSE_ERRORS */

Run make, cp and press the reset. Now:

  ReferenceError: identifier 'require' undefined

See Duktape 1.x compatible module loading framework

• Makefile
  • Added ../duktape2/extras/module-duktape/duk_module_duktape.c to the C_SRCS
  • Added INCLUDE_PATHS += -I../../duktape2/extras/module-duktape
• eduk.cpp
  • Added
    // See Duktape 1.x compatible module loading framework at                                                                
    // https://github.com/svaarala/duktape/tree/master/extras/module-duktape                                                
    extern "C" {
    #include "duk_module_duktape.h"
    }

    [$[Get Code]]
  • After setting ctx, added
      // Duktape 1.x compatible module loading framework.                                                                    
      // https://github.com/svaarala/duktape/tree/master/extras/module-duktape                                              
      duk_module_duktape_init(ctx);
    (:source:)
    
    Got further!
    
    !![[#deterministicTemporalSemantics]] deterministicTemporalSemantics.js missing
    (:source lang=javascript:)
    duk2.cpp main() done with duk_create_heap_default()
    eduk2.cpp main() about to call eventloop_register()
    eduk2.cpp main() done with eventloop_register()
    eduk: About to run test/auto/RampJSDisplay.js
    ../eduk2.cpp:85: About to load C version of c_eventloop.
    ../eduk2.cpp: Loading C version of c_eventloop worked
    ../../eduk2/nofileio.c:190 Could not find common/modules/deterministicTemporalSemantics.js
    ../eduk2.cpp:106: Loading C version of duktapeHost failed.  Error was:
    ReferenceError: identifier 'FileIo' undefined

    [$[Get Code]]

The solution to the above was to add detrministicTemporalSemantics.js to the list of files for which we generate .h files and modify eduk2/nofileio.c to use it.

gettimeofday

The problem now is that the swarmlet loads but fails to produce output.

Compiling the code under x86_64 on the mac works properly. See the instructions at the top of eduk2.cpp for how to do this. The makefile is too messed up to easily add this there, so the commands are run by hand.

Adding fprintfs to c_eventloop.c shows that we get output like:

In particular, it looks like gettimeofday() always returns 0!

The binary has three definitions of gettimeofday

eduk2.cpp defines a _gettimeofday, but that is now ifdef'd out for mbed.

Running

  nm $(filter %.o, $^) | grep gettimeofday

returns

       U gettimeofday

which indicates that we are looking for the gettimeofday() definition and that it is not defined in the .o files

Adding -Wl,-v,-t to the $(LD) line shows that gettimeofday is being found here:

  (/usr/local/Cellar/arm-none-eabi-gcc/5-2016-q3-update/bin/../lib/gcc/arm-none-eabi/5.4.1/../../../../arm-none-eabi/lib/armv7e-m/softfp/libc.a)lib_a-gettimeofdayr.o

Running nm on /usr/local/Cellar/arm-none-eabi-gcc/5-2016-q3-update/bin/../lib/gcc/arm-none-eabi/5.4.1/../../../../arm-none-eabi/lib/armv7e-m/softfp/libc.a shows the following:

So, it appears that lib_a-sysgettod.o defines a gettimeofday() that calls _gettimeofday_r(), which is also defined.

It is not clear why the addresses are 00000000.

It looks like gettimeofday_r is a reentrant version of gettimeofday.

It could be that the gettimeofday() that is defined in the library returns 0, but there is no documentation about that.

See

• DuktapeEclipseARMCortexM3 -> TimerSetup
• DuktapeEclipseARMCortexM4 -> TimerSetup
• TockOS Compilation -> gettimeofday

https://github.com/ARMmbed/mbed-time defines a version, we could try that.

Whoops. gettimeofday() returns 0 if everything is ok. Correcting c_eventloop.h and eduk2.cpp results in:

Added perror():

results in

It looks like gcc-arm-none-eabi-5_4-2016q3-20160926/src/newlib/newlib/libc/sys/arm/syscalls.c contains the definition:

Interestingly at the end of the verbose output of ld, we have

(/usr/local/Cellar/arm-none-eabi-gcc/5-2016-q3-update/bin/../lib/gcc/arm-none-eabi/5.4.1/../../../../arm-none-eabi/lib/armv7e-m/softfp/libnosys.a)gettod.o

Going to https://launchpad.net/gcc-arm-embedded/+download and downloading https://launchpad.net/gcc-arm-embedded/5.0/5-2016-q3-update/+download/gcc-arm-none-eabi-5_4-2016q3-20160926-src.tar.bz2 and looking for gettod finds ./src/newlib/libgloss/libnosys/gettod.c:

Revisiting the code and printing errno:

Now, we get

/src/newlib/newlib/libc/include/sys/errno.h has

So, what is happening here is that we get the errno = 88 and then the fprintf fails with a Bad file number, which is why perror() displays that.

So, gettimeofday is not implemented!

See MaximPegasus gettimeofday