llext: add initial support for arc

[laurenmurphyx64]

Adds compiler flag(s) and architecture-specific relocations for ARC.

Notes:

• Tests don't work without CONFIG_ARC_MPU_ENABLE=n
• So userspace doesn't work (will come in with another PR)
• Not all relocations supported (future PRs)

[pillo79]

Nice, one more arch on the list! 👍

(all tests pass with just 2 relocations?)

Perhaps that's due to -mlong-calls. Not sure about the evils that may be unleashed dropping it though 🙂

[mathieuchopstm]

Surface-level comments as I'm not well versed in ARC specifics:

[laurenmurphyx64]

Added R_ARC_B26 and R_ARC_S25W_PCREL to support hs5/6x, still need to fix some build errors on hs5x

Edit: Deferring build errors on two tests to another PR, to be linked with this issue

[abrodkin]

Based on what relocations were generated by our testsuite, which turned out to be 4.

@laurenmurphyx64 what is that testsuite? May we look at it and see what kind of results it produces for ARC?

[laurenmurphyx64]

Based on what relocations were generated by our testsuite, which turned out to be 4.

@laurenmurphyx64 what is that testsuite? May we look at it and see what kind of results it produces for ARC?

@abrodkin

We have 5 emulated ARC boards:

qemu_arc/qemu_arc_em      qemu_arc/qemu_arc_hs5x    qemu_arc/qemu_arc_hs/xip
qemu_arc/qemu_arc_hs      qemu_arc/qemu_arc_hs6x

To run the testsuite on one (keep adding -p [BOARD NAME] to run on multiple boards):

twister -T tests/subsys/llext/simple -p qemu_arc/qemu_arc_em

You can check the output in zephyrproject/zephyr/twister-out/. You can also run directly with West and see the build / run output in the console. I'd recommend doing this first because it's easier to see laid out.

west build -p -b qemu_arc/qemu_arc_em tests/subsys/llext/simple -t run

The tests output a lot of detail about relocations. Example:

D: Counting exported symbols...
D: symbol count 40
D: unhandled symbol 1, name threads_kernel_objects_ext.c, type tag 4, bind 0, sect 65521
D: unhandled symbol 2, name , type tag 3, bind 0, sect 1
D: unhandled symbol 3, name , type tag 3, bind 0, sect 3
D: unhandled symbol 4, name , type tag 3, bind 0, sect 4
D: unhandled symbol 5, name device_get_binding, type tag 2, bind 0, sect 1
D: unhandled symbol 6, name device_is_ready, type tag 2, bind 0, sect 1
D: unhandled symbol 7, name k_thread_create, type tag 2, bind 0, sect 1
D: unhandled symbol 8, name k_thread_join, type tag 2, bind 0, sect 1
...
D: Linking ELF...
D: relocation section .rela.text (2) acting on section 1 has 67 relocations
D: relocation 2:0 info 0x1711 (type 17, sym 23) offset 18 sym_name z_impl_device_get_binding sym_type 0 sym_bind 1 sym_ndx 0
I: found symbol z_impl_device_get_binding at 0x80009540
I: writing relocation symbol z_impl_device_get_binding type 17 sym 23 at addr 0x804032d2 addr 0x80009540
D: relocation 2:1 info 0x1811 (type 17, sym 24) offset 54 sym_name z_impl_device_is_ready sym_type 0 sym_bind 1 sym_ndx 0
I: found symbol z_impl_device_is_ready at 0x80009528
I: writing relocation symbol z_impl_device_is_ready type 17 sym 24 at addr 0x804032f6 addr 0x80009528
D: relocation 2:2 info 0x1911 (type 17, sym 25) offset 164 sym_name z_impl_k_thread_create sym_type 0 sym_bind 1 sym_ndx 0
I: found symbol z_impl_k_thread_create at 0x8000bac0
I: writing relocation symbol z_impl_k_thread_create type 17 sym 25 at addr 0x80403364 addr 0x8000bac0
D: relocation 2:3 info 0x1a11 (type 17, sym 26) offset 218 sym_name z_impl_k_thread_join sym_type 0 sym_bind 1 sym_ndx 0
I: found symbol z_impl_k_thread_join at 0x8000d22c
I: writing relocation symbol z_impl_k_thread_join type 17 sym 26 at addr 0x8040339a addr 0x8000d22c
...

The files in tests/subsys/llext/simple/src with the ending _ext.c are the sources for the test extensions, but if you run with Twister they / Zephyr are built with different CONFIG options specified in tests/subsys/llext/simple/testcase.yaml. tests/subsys/llext/simple/src/test_llext_simple.c is the actual Zephyr test app that loads, invokes, and unloads the compiled extensions.

[pillo79]

Thanks! Spotted 2 minor issues, not blockers but should probably be addressed in a followup PR.