ESP8266 Community Forum

Hi,

I received a couple of ESP-01 boards last week and started playing around with them. I installed the esp-open-sdk and compiled the toolchain. I have since compiled a few pure C examples and wrote a few on my own as well with success.
Lately I decided to up my game and try to get something up and running based on C++. The example I am trying to compile right now uses std::string and std::function. Whenever I try to add libstdc++ as an additional library, the linker complains that section `.text' will not fit in region `iram1_0_seg'.
Looking at the map file, all libstdc++ object do indeed seem to end up in the iram1_0 segment. I am not completely familiar with the memory space layout of the ESP8266, but from what I get from reading this forum, one can choose to place its code directly in iram1_0_seg or in irom0_0_seg where it will be loaded and cached in ram on demand. Is there a way to have the libstdc++ objects place in irom0_0_seg?
Is there a working example somewhere of a project leveraging the c++ standard library that I could start from?

I am attaching my Makefile for completeness.

Thanks in advance.

bouvrais wrote:Is there a way to have the libstdc++ objects place in irom0_0_seg?

You can use a custom linker script to put certain pieces of code in the irom0_0_seg. The linker scripts that the Arduino platform uses employ this technique. Alternately, you can rename all of the text segments in the stdc++ library to irom0_0_seg and then use the standard linker scripts.

By the way, this issue isn't specific to C++ or the stdc++ libraries. Any source code or library written by another party that isn't specifically for the ESP8266 will have the same issue.

Thanks dkinzer.
The Arduino ESP8266 GitHub was indeed a good source for inspiration. Creating a custom linker script and moving libstdc++. (as well as libgcc.a) to the irom0_0_seg did indeed hush most of the linker errors. For the remaining ones, I had to stub some libc functions on which libstdc++ rely and are not provided with the toolchain.
So now my app does compile and link correctly. And of course it crashes on load with a cycling Fatal Exception over and over again :



According to my map file, 0x4024129c is the address of the __udivsi3 libgcc function.
This happens with an empty user_init() function and as far as I know static constructors are not called by default (one needs to call them explicitely) so I would like to rule that out as a possible cause. That leaves me with very few ideas other than the bootloader doesn't like the code I am providing.

Any suggestion on how I could investigate this crash further?

So! After some trial and error I found that if I remove libgcc.a from irom and let it back into iram, my empty user_init() behaves correctly (no crash). So I am pretty much back to square one as doing this not only fulls the iram (which becomes too small) but now code in the text section of libstdc++ complains that calls made to symbols in libgcc are out of range:



So that leaves me with two questions:

• Why does the app crash when libgcc is in irom? I guess understanding this would be easier with information on the bootloader.
• Would libstdc++ need to be recompiled with -mlongcalls? Why is that not the default (or is it and something else explains this)?