Le 2022-09-21 08:18, Stefan Kalkowski a écrit :
Hello Jean-Adrien,
On Tue, Sep 20, 2022 at 05:41:50PM +0200, Jean-Adrien Domage wrote:
/tmp/ccOLpMJW.s: Assembler messages: /tmp/ccOLpMJW.s:987: Error: section name '__irqchip_of_table' already defined as another symbol
This error message indicates a rather obvious error: It seems the symbol with that name is defined twice with different 'type' in different compilation units. However, when using grep I can find the symbol within `repos/dde_linux/src/lib/lx_emul/spec/arm/irqchip.c` but it is not that clear within the Linux source where it is defined. I assume in a linker script but I'm not sure at all. I couldn't find it using grep (not event a declaration, maybe some macro magic happen here). Which let me think that the error might not be due to what I actually think...In general, you might find the compilation unit that already contains the symbol of the Linux kernel by applying `find` to all object files of your target, and by combining `nm` and `grep` to find the locations. You can use the flag `--print-file-name` when calling `nm` to include the corresponding filename in every match.
In the concrete case, I assume it is a combination of macros, like `IRQCHIP_DECLARE`, and the ones it depends on. Probably you've included a Linux kernel source file with interrupt controller definitions, which is in most cases not a good idea.
Best regards Stefan
Dear Genodians,
Thank you, everyone! I appreciate all of your answers! I'm now able to compile everything with dummies, without assembler errors. Here is my story:
I tried with nm and grep on each generated object in the build directory but couldn't find the culprit either.
I couldn't find the symbol with nm & grep before, as I followed the article from Stefan [1]. I shot myself in the foot because I forgot to remove `...| grep -i " t " |...` from the command line. Then I proceeded by eliminating every file referencing `__irqchip_of_table` which as led me to `repos/dde_linux/src/lib/lx_emul/spec/arm/irqchip.c` again.
[1] https://genodians.org/skalk/2021-04-08-dde-linux-experiments-1
With your help, I finally figured it out:
In the concrete case, I assume it is a combination of macros, like `IRQCHIP_DECLARE`, and the ones it depends on. Probably you've included a Linux kernel source file with interrupt controller definitions, which is in most cases not a good idea.
1. As suggested by Stefan, I paid attention not to do so. Only `lx_emul/../irqchip.c` made uses of `IRQCHIP_DECLARE`. To ensure that, commenting out this file from the `target.mk` made the error go away. But I do need it eventually.
2. I suspected a macro expansion from a Linux header to be the culprit. I added `CC_OPT += -S -save-temps=obj` to both `target.mk` of my mmc and sd_card drivers. The build may fail with lots of errors, but I can now compare the `irqchip.i` from mmc/ and sd_card/
The result for mmc is: ``` static const struct of_device_id __of_table_dde_gic_v3 ... __attribute__((__section__("__" "irqchip" "_of_table"))) __attribute__((__aligned__(__alignof__(struct of_device_id)))) = ... ``` Which is probably not what I want. Note how the attribute section is set `__section__("__" "irqchip" "_of_table")` which makes it not friendly to grep for.
For the sd_card driver, I obtain: ``` static void __of_declare_initcall_dde_irqchip_initdde_gic_v3(void)__attribute__((constructor)); static void __of_declare_initcall_dde_irqchip_initdde_gic_v3() { lx_emul_register_of_irqchip_initcall( ... ```
3. With our Linux source tree, something goes wrong when `IRQCHIP_DECLARE` expands. It does not result in an init_call registration, It turns out that `of.h` was missing from the shadow include path xD ...
Thank you very much!
Cheers, Jean-Adrien
--- Jean-Adrien Domage gapfruit AG Operating System Engineer