Include standard filesystem symlinks like /lib -> /usr/lib in the image

[ishitatsuyuki]

Kernel firmware for example requires this because it only searches /lib/firmware/. Including these is in general a good idea for compatibility.

If you put modules_force_load: amdgpu in the config you get:

Sep 20 20:57:44 ishitatsuyuki-ryzen kernel: amdgpu 0000:0a:00.0: loading /lib/firmware/amdgpu/navi10_sos.bin failed with error -4
Sep 20 20:57:44 ishitatsuyuki-ryzen kernel: amdgpu 0000:0a:00.0: Direct firmware load for amdgpu/navi10_sos.bin failed with error -4
Sep 20 20:57:44 ishitatsuyuki-ryzen kernel: amdgpu 0000:0a:00.0: amdgpu: failed to init sos firmware
Sep 20 20:57:44 ishitatsuyuki-ryzen kernel: [drm:psp_sw_init [amdgpu]] *ERROR* Failed to load psp firmware!

and the graphics doesn't initialize. (It does boot with default config because it's loaded after rootfs is mounted.)

mkinitcpio symlinks for the reference.

[ishitatsuyuki]

Actually this error message says -4 which is EINTR, it doesn't even make sense. Guess I'll need to dig deeper into this...

[ishitatsuyuki]

Also #45 seems to be affected by the same issue.

[anatol]

Actually this error message says -4 which is EINTR, it doesn't even make sense.

It is interesting... I do not have an AMD video card and it is hard for me to understand what is going on there. But it would be great to debug and fix this issue.

Do you use strip: true config option by the chance?

[ishitatsuyuki]

Do you use strip: true config option by the chance?

I don't, I only have the modules_force_load line.

[anatol]

I just pushed a commit to wip branch that recreates the same symlink tree as mkinitcpio does. Please try it and let me know if you see any difference.

[anatol]

Now I would like to get to the EINTR error you see:

Sep 20 20:57:44 ishitatsuyuki-ryzen kernel: amdgpu 0000:0a:00.0: loading /lib/firmware/amdgpu/navi10_sos.bin failed with error -4
Sep 20 20:57:44 ishitatsuyuki-ryzen kernel: amdgpu 0000:0a:00.0: Direct firmware load for amdgpu/navi10_sos.bin failed with error -4
Sep 20 20:57:44 ishitatsuyuki-ryzen kernel: amdgpu 0000:0a:00.0: amdgpu: failed to init sos firmware
Sep 20 20:57:44 ishitatsuyuki-ryzen kernel: [drm:psp_sw_init [amdgpu]] *ERROR* Failed to load psp firmware!

Are there any log entries from booster before and after these lines. Any logs that include finit word?

[anatol]

I am looking at the sources of insmod here https://git.kernel.org/pub/scm/utils/kernel/kmod/kmod.git/tree/shared/util.c and the only place where EINTR from kernel is handled is when module content is read from the disk.

I wonder if FINIT_MODULE kernel implementation is simply unable to read the module content in one step and returns the error back to user. I can try to come up with a patch that reads module upfront and then calls INIT_MODULE instead of FINIT_MODULE.

[anatol]

Also here is a manpage for the syscall booster uses to load the modules https://linux.die.net/man/2/finit_module

[anatol]

@ishitatsuyuki btw to enable booster debug logs please add booster.debug kernel parameter.

[ishitatsuyuki]

With 408e628 (and the new wip branch) I'm magically able to boot. Still unsure about the -4 error code, but I think this can be considered resolve once the commit is merged to master.

[ishitatsuyuki]

FWIW I don't think it's finit_module that is failing but rather just the firmware loading call. It's done entirely inside kernel space which shouldn't give EINTR. The relevant code is here.

[anatol]

These are great news @ishitatsuyuki. If you have a chance could you please also check that 90dd03a (parent of 408e628) still has the problem? This way it will prove that symlinks were the root of the issue.

[ishitatsuyuki]

I can confirm that 90dd03a does not work as you mentioned.

[anatol]

Awesome! Thank you for the confirmation Tatsuyuki-san.

I am going to test wip changes for a week or two and if everything is OK I'll push it to master. If you have time please test it as well and let me know if you see any other issues.