makefiles/arch/msp430.inc.mk: Fix compilation issues with GCC 12

[maribu]

Contribution description

This fixes the following compilation issues:

/home/maribu/Repos/software/RIOT/cpu/msp430fxyz/periph/gpio.c: In function 'gpio_periph_mode':
/home/maribu/Repos/software/RIOT/cpu/msp430fxyz/periph/gpio.c:95:15: error: array subscript 0 is outside array bounds of 'msp_port_isr_t[0]' [-Werror=array-bounds]
   95 |         sel = &(isrport->SEL);
      |               ^~~~~~~~~~~~~~~
cc1: all warnings being treated as errors

by adding CFLAGS += --param-min-pagesize=0 for GCC 12 (same issue as already fixed for AVR).

and:

/usr/lib/gcc/msp430-elf/12.2.0/../../../../msp430-elf/bin/ld: warning: /home/maribu/Repos/software/RIOT/cpu/msp430_common/ldscripts/xfa.ld contains output sections; did you forget -T?

by adding the missing -T.

Testing procedure

The following should still work:

• make BOARD=msb-430 -C examples/hello-world
• make BOARD=msb-430 -C tests/xfa flash test

Issues/PRs references

None

[maribu]

@kaspar030: Could you give this a try? If this doesn't cause any regressions, could you test again with

diff --git a/cpu/msp430_common/ldscripts/xfa.ld b/cpu/msp430_common/ldscripts/xfa.ld
index 7b115a66c1..6989376a66 100644
--- a/cpu/msp430_common/ldscripts/xfa.ld
+++ b/cpu/msp430_common/ldscripts/xfa.ld
@@ -6,8 +6,6 @@ SECTIONS
     } > ROM
 }
 
-INSERT AFTER .rodata;
-
 SECTIONS
 {
     .data :
@@ -20,5 +18,3 @@ SECTIONS
     PROVIDE (edata = .);
     PROVIDE (__dataend = .);
 }
-
-INSERT AFTER .data;

Also applied? These INSERT AFTER doesn't seem to work anymore across linker scripts with recent binutils. But I think just by having the XFA linker script being executed after the first, it will place the XFA stuff at the end anyway.

[riot-ci]

Murdock results

✔️ PASSED

b86366e makefiles/toolchain: Fix incorrect assumptions on address space layout

Success	Failures	Total	Runtime
6931	0	6931	09m:40s

Artifacts

• Documentation preview

[maribu]

What a pain. I think that linking now fails with the old binutils in the container is actually a bug in that old binutils version :/

[maribu]

This one is still needed

[benpicco]

But CI is not happy about it

[maribu]

Yeah, seems to be a major issue to get the xfa linker script working across different versions of binutils :-/

[maribu]

OK, brute force solution: Use the wrapper linker scripts for binutils >= 2.40, don't use them with binutils < 2.40.

I have no idea why ld 2.40 is so much disagreeing with its younger self about what a valid linker script should look like and how to come up with one that both agree on being valid :/

I would like to test this on real hardware first. (For BUILD_IN_DOCKER=1 it should still work, as nothing has changed then. But I would like to be sure that this now not only links on newer toolchains, but also boots and behaves fine.)

[maribu]

Now all is green and it does work with both CI and local toolchain (at least for my Olimex MSP430H1611 board).

[maribu]

bors merge

[bors]

Build failed (retrying...):

• tools-build-success

[maribu]

bors cancel
bors merge

[bors]

Canceled.

[bors]

Build failed (retrying...):

• tools-build-success

[bors]

Build failed (retrying...):

• tools-build-success

[bors]

Build succeeded!

The publicly hosted instance of bors-ng is deprecated and will go away soon.

If you want to self-host your own instance, instructions are here.
For more help, visit the forum.

If you want to switch to GitHub's built-in merge queue, visit their help page.

• Murdock
• static-tests
• tools-build-success

[maribu]

Thx :)