llext samples fail with BUS ERROR (zephyr-3.7 stable)

[fulup-bzh]

In the process of trying LLEXT I check samples/subsys/llext/modules and fail to get it working.

• zephyr 3.7 stable
• Linux zephyr-sdk-0.16.8
• board frdm_k64f

do_llext_load seams to work find withall symbols are resolved. But when launching extension function, I get a bus fault.

Dis anyone success in running LLEXT samples ?

thank you

[00:00:00.568,000] <dbg> llext: do_llext_load: Allocating memory for symbol table...
[00:00:00.576,000] <dbg> llext: do_llext_load: Copying symbols...
[00:00:00.583,000] <dbg> llext: llext_copy_symbols: function symbol 0 name hello_world addr 0x20002121
[00:00:00.592,000] <dbg> llext: do_llext_load: Linking ELF...
[00:00:00.599,000] <dbg> llext: llext_link: relocation section .rel.text (2) acting on section 1 has 2 relocations
[00:00:00.609,000] <dbg> llext: llext_link: relocation 2:0 info 0x202 (type 2, sym 2) offset 16 sym_name  sym_type 3 sym_bind 0 sym_ndx 3
[00:00:00.622,000] <inf> llext: writing relocation symbol  type 2 sym 2 at addr 0x20002130 addr 0x20002180
[00:00:00.632,000] <dbg> elf: arch_elf_relocate: 2 20002130 20002180 
[00:00:00.639,000] <dbg> llext: llext_link: relocation 2:1 info 0xe02 (type 2, sym 14) offset 20 sym_name printk sym_type 0 sym_bind 1 sym_ndx 0
[00:00:00.652,000] <inf> llext: found symbol printk at 0x65c9
[00:00:00.659,000] <inf> llext: writing relocation symbol printk type 2 sym 14 at addr 0x20002134 addr 0x65c9
[00:00:00.669,000] <dbg> elf: arch_elf_relocate: 2 20002134 65c9 printk
[00:00:00.676,000] <dbg> llext: llext_link: relocation section .rel.exported_sym (5) acting on section 4 has 2 relocations
[00:00:00.687,000] <dbg> llext: llext_link: relocation 5:0 info 0x202 (type 2, sym 2) offset 0 sym_name  sym_type 3 sym_bind 0 sym_ndx 3
[00:00:00.700,000] <inf> llext: writing relocation symbol  type 2 sym 2 at addr 0x200021e0 addr 0x20002180
[00:00:00.710,000] <dbg> elf: arch_elf_relocate: 2 200021e0 20002180 
[00:00:00.717,000] <dbg> llext: llext_link: relocation 5:1 info 0xf02 (type 2, sym 15) offset 4 sym_name hello_world sym_type 2 sym_bind 1 sym_ndx 1
[00:00:00.731,000] <inf> llext: writing relocation symbol hello_world type 2 sym 15 at addr 0x200021e4 addr 0x20002121
[00:00:00.742,000] <dbg> elf: arch_elf_relocate: 2 200021e4 20002121 hello_world
[00:00:00.750,000] <dbg> llext: llext_export_symbols: sym 0x20002121 name hello_world in 0x200021c4
[00:00:00.759,000] <dbg> llext: do_llext_load: loaded module, .text at 0x20002120, .rodata at 0x20002180
[00:00:00.769,000] <inf> llext: Loaded extension ext
[00:00:00.774,000] <err> os: ***** BUS FAULT *****
[00:00:00.780,000] <err> os:   Instruction bus error
[00:00:00.785,000] <err> os:   NXP MPU error, port 3
[00:00:00.791,000] <err> os:     Mode: Supervisor, Instruction Address: 0x20002120
[00:00:00.799,000] <err> os:     Type: Read, Master: 0, Regions: 0x8400
[00:00:00.807,000] <err> os: r0/a1:  0x20002121  r1/a2:  0x000086e6  r2/a3:  0x00000000
[00:00:00.815,000] <err> os: r3/a4:  0x00000000 r12/ip:  0x00002801 r14/lr:  0x000007ad
[00:00:00.824,000] <err> os:  xpsr:  0x61000000
[00:00:00.829,000] <err> os: Faulting instruction address (r15/pc): 0x20002120
[00:00:00.837,000] <err> os: >>> ZEPHYR FATAL ERROR 27: Unknown error on CPU 0
[00:00:00.845,000] <err> os: Current thread: 0x200005d0 (unknown)
[00:00:00.852,000] <err> os: Halting system

[fulup-bzh]

Find the issue. LLEXT samples prj.conf should include CONFIG_ARM_MPU=n as listed in #72957 (comment)

Note that is fix was not included neither in 3.7 neither in current Master when it should be, isn't it ?

[DerekSnell]

Hi @fulup-bzh ,
Thank you for reporting the solution. Since you found the solution, are you willing to contribute and submit a Pull Request to fix this? It appears to be a simple one to submit. Please follow the Contribution Guidelines.

Thanks

[aaillet]

Hi @fulup-bzh , Thank you for reporting the solution. Since you found the solution, are you willing to contribute and submit a Pull Request to fix this? It appears to be a simple one to submit. Please follow the Contribution Guidelines.

Thanks

@DerekSnell from what I can see from test script for this sample, it's a known case that we need to disable ARM MPU : https://github.com/zephyrproject-rtos/zephyr/blob/main/samples/subsys/llext/modules/sample.yaml#L36

Another way stated here would be to enable CONFIG_USERSPACE for the example to work.

I'm going to try a pull request, but in my opinion, this modification is too situational to be added to the sample prj.conf.

[fulup-bzh]

@DerekSnell I ask @aaillet one of our engineer to create a pull request.

[fulup-bzh]

Without ARM MPU sample does not work at least on ARM architecture, and I fail to make it work with CONFIG_USERSPACE. I lost a full day searching to get a solution, and from what I saw in forums, I'm not the only one who fail on LLEXT samples.

I did not test on Intel architecture, but at least on Arm this feature should be added to default config file.

• either directly in prj.conf
• in arm specific config, when LLEXT_CONFIG is selected.

[pillo79]

This sample shows how to call the same function either directly or via LLEXT. The README in that folder specifies how it is supposed to be built, and specifically mentions looking at sample.yaml for the arch-specific extra configs that are currently required for LLEXT.

I would not add CONFIG_ARM_MPU=n to the common config. The sample is designed precisely to higlight these differences! 🙂 Perhaps we should add a note in the prj.conf that points to the docs and/or sample.yaml instead?

[fulup-bzh]

@pillo79 publishing "non working" samples is really a good idea to promote technology. In order to reduce the initial cost of adoption sample should work out of the box. Which is not the case. CONFIG_ARM_MPU=n flag should be set automatically when selecting "CONFIG_MODULES=y" or even better when selecting CONFIG_LLEXT=y at a global level.