boards: msp430: garbage collect dead code when linking.

[basilfx]

The MSP430-based board are blacklisted from U8g2 package, because the flash file size exceeds the memory size. There was a similar problem with the ATMega targets, which was fixed with #5632.

This PR does the same for MSP430-based boards. However, I don't have any boards to test it with and I don't know what implications it has for other programs. Without this patch the U8g2 test application fails to compile, with this patch it succeeds.

msp430-size /home/basilfx/Desktop/RIOT/tests/pkg_u8g2/bin/chronos/pkg_u8g2.elf
   text    data     bss     dec     hex filename
  14472      90    2644   17206    4336 /home/basilfx/Desktop/RIOT/tests/pkg_u8g2/bin/chronos/pkg_u8g2.elf

msp430-size /home/basilfx/Desktop/RIOT/tests/pkg_u8g2/bin/msb-430/pkg_u8g2.elf
   text    data     bss     dec     hex filename
  14886      18    2702   17606    44c6 /home/basilfx/Desktop/RIOT/tests/pkg_u8g2/bin/msb-430/pkg_u8g2.elf

msp430-size /home/basilfx/Desktop/RIOT/tests/pkg_u8g2/bin/wsn430-v1_3b/pkg_u8g2.elf
   text    data     bss     dec     hex filename
  14838      18    2702   17558    4496 /home/basilfx/Desktop/RIOT/tests/pkg_u8g2/bin/wsn430-v1_3b/pkg_u8g2.elf

[jnohlgard]

Could this not be set directly in the cpu/msp430-common makefile instead?

[basilfx]

You are absolutely right. I just couldn't find that file since I wasn't looking for Makefiles in the cpu/ itself :-)

Fixed it.

[jnohlgard]

ACK if Murdock agrees.

[jnohlgard]

Btw, should #5632 be modified the same way perhaps? (move from boards to cpu dir)

[basilfx]

Murdock agrees.

Yes, I think a similar change must be made for #5632. Maybe this can be part of #5451 and/or #5590 (cc @mali)? I'm not really good with Makefiles and I don't have boards to test this with :-)

[LudwigKnuepfer]

Just for curiosity: could you please post numbers for master and with/without the -ffunction-sections -fdata-sections change?

[basilfx]

Just for curiosity: could you please post numbers for master and with/without the -ffunction-sections -fdata-sections change?

I would love to do, but compilation fails before it can output numbers. However, it spits out numbers that exceed 1 MB due to all the fonts being compiled in.

[OlegHahm]

You could try to modify the linkerscripts to fake more available memory.

[basilfx]

Well, another problem is that the MSP-430 cannot address more than 1 MB (20 bit address if I recall correctly).

/build/buildd/gcc-msp430-4.6.3~mspgcc-20120406/./gcc-4.6.3/gcc/config/msp430/crt0.S:195: relocation truncated to fit: R_MSP430_16_BYTE against symbol `__data_load_start' defined in *ABS* section in /home/basilfx/Bureaublad/RIOT/tests/pkg_u8g2/bin/chronos/pkg_u8g2.elf

Anyway, this is some of the build output for Chronos board (1.67 MB):

/usr/lib/gcc/msp430/4.6.3/../../../../msp430/bin/ld: /home/basilfx/Bureaublad/RIOT/tests/pkg_u8g2/bin/chronos/pkg_u8g2.elf section `.rodata' will not fit in region `rom'
/usr/lib/gcc/msp430/4.6.3/../../../../msp430/bin/ld: section .vectors loaded at [000000000000ff80,000000000000ffff] overlaps section .rodata loaded at [000000000000bb10,00000000001bba6b]
/usr/lib/gcc/msp430/4.6.3/../../../../msp430/bin/ld: region `rom' overflowed by 1751878 bytes

[OlegHahm]

Ah, yes, true. Anyhow, we it's obvious that this improves a lot. Did anyone check if it is still working on hardware?

[kYc0o]

I can give a try on a Z1 node

[kYc0o]

I confirm it still works

[basilfx]

Rebased.

Murdock failed this afternoon, but it already succeeded yesterday and no new commits were added since. Related to #5689?

[kYc0o]

I don't think so. I detected the same problem in 2 different PRs.

[kaspar030]

I don't think so. I detected the same problem in 2 different PRs.

Fixed. A system upgrade made gdb collide with stack randomization seccomp options.

[kYc0o]

Great! Thanks @kaspar030 :D

[basilfx]

Murdock complains, but I don't think it has to do with this PR. Something about not being able to reach Github.com for jsmn package.