cpu: cortexm: make newlib a default module

[kaspar030]

Contribution description

Currently, newlib is always selected using USEMODULE for all Cortex-M.
This PR turns it into DEFAULT_MODULE. The advantage is that now it can be disabled on-demand using DISABLE_MODULE.

Testing procedure

CI compiling successfully should be fine.

Issues/PRs references

With #10744, this allows deployments without newlib altogether in some cases.

[jcarrano]

To be more precise, this does really disable newlib, only the bindings to the system calls (which are not "system calls" in RIOT really.)

I'm trying to achieve the same, but in a more granular way, in #9258. Many applications will not be able to remove their dependencies on all syscalls, but they can still benefit from not including certain parts.

[kaspar030]

To be more precise, this does really disable newlib

Yup. Useful for chopping off unused fat, or possibly using a slimmer libc, or ...

[cladmi]

Can you give a testing procedure for the NEW possible usage?

[cladmi]

If it should be used to select a different libc I would not do it with DEFAULT_MODULE.
libc should be in DEFAULT_MODULE, and newlib_nano would be the default implementation.

So I am interested at seeing the usage.

Also right now, when not setting newlib I think it would make the default C library being used asnewlib.inc.mk would not be included. Which feels different from the meaning of disabling newlib.

[kaspar030]

Can you give a testing procedure for the NEW possible usage?

Add DISABLE_MODULE += newlib newlib_nano to tests/minimal/Makefile.
Then try compilation with master and with this branch for e.g., samr21-xpro.

On master, you'll get the usual "EXPECT ERRORS" warning. There even seems to be some effect on code size, but that must be accidental.

Size:

   text    data     bss     dec     hex filename
   3044      36    2604    5684    1634 /home/kaspar/src/riot/tests/minimal/bin/samr21-xpro/tests_minimal.elf

This is the output of BOARD=samr21-xpro make info-modules:

board
core
core_msg
cortexm_common
cortexm_common_periph
cpu
isrpipe
newlib
newlib_nano
newlib_syscalls_default
periph
periph_common
periph_gpio
periph_pm
periph_uart
pm_layered
sam0_common_periph
stdio_uart
sys
tsrb

Note newlib*, stdio_uart and periph_uart.

With this PR (and the same modified Makefile):

Size:

   text    data     bss     dec     hex filename
   2140      16    2524    4680    1248 /home/kaspar/src/riot/tests/minimal/bin/samr21-xpro/tests_minimal.elf

info-modules:

[kaspar@ng minimal (cortexm_make_newlib_default_module)]$ BOARD=samr21-xpro make info-modules
board
core
core_msg
cortexm_common
cortexm_common_periph
cpu
periph
periph_common
periph_gpio
periph_pm
pm_layered
sam0_common_periph
sys
[kaspar@ng minimal (cortexm_make_newlib_default_module)]$

Note the absence of newlib*, stdio_uart and periph_uart.

[kaspar030]

Also right now, when not setting newlib I think it would make the default C library being used asnewlib.inc.mk would not be included. Which feels different from the meaning of disabling newlib.

Actually with our toolchains, newlib is the default. By disabling the newlib module, what actually happens is that RIOT's newlib support code (the syscalls) and their dependency don't get built/linked.
Which is useful for bare-metal stuff (e.g., riotboot).

libc should be in DEFAULT_MODULE, and newlib_nano would be the default implementation.

Agreed, but IMO that would be a next step.

[jcarrano]

To be more precise, this does really disable newlib

Oh, OK, good to know. I think we need a test to show that a functioning application can be built without newlib, and that newlib itself is not being included (i.e. that the change is effective). For the second point, maybe verify that newlib init stuff is not the symbols in the elf file.

[kaspar030]

I think we need a test to show that a functioning application can be built without newlib, and that newlib itself is not being included (i.e. that the change is effective). For the second point, maybe verify that newlib init stuff is not the symbols in the elf file.

I've added a commit disabling newlib for the bootloader.

This is before (on samr21-xpro):

   text    data     bss     dec     hex filename
   2544      24     672    3240     ca8 /home/kaspar/src/riot/bootloaders/riotboot/bin/samr21-xpro/riotboot.elf

This is after:

   text    data     bss     dec     hex filename
   1124       0     512    1636     664 /home/kaspar/src/riot/bootloaders/riotboot/bin/samr21-xpro/riotboot.elf                              

@basilfx can you give this a try on efm32?

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. If you want me to ignore this issue, please mark it with the "State: don't stale" label. Thank you for your contributions.

[fjmolinas]

@basilfx can you give this a try on efm32?

I have 1 efm32, what would you want me to test? just build riotboot? @kaspar030

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. If you want me to ignore this issue, please mark it with the "State: don't stale" label. Thank you for your contributions.

[fjmolinas]

ping @kaspar030 are you still interested in this one?

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. If you want me to ignore this issue, please mark it with the "State: don't stale" label. Thank you for your contributions.

[miri64]

ping @kaspar030 are you still interested in this one?

Obviously not...