See if things build without these ldflags

[alex]

No description provided.

[alex]

So, clearly it doesn't... but now I understand a little bit more about why:

The problem is that https://github.com/torvalds/linux/blob/v5.2/scripts/mod/modpost.c#L697-L698 is never triggered.

I'm not sure what file this is run on, but I wonder if this isn't a .a vs. .o issue?

[alex]

Tracing a bit more, that line of code follows after https://github.com/torvalds/linux/blob/v5.2/scripts/mod/modpost.c#L419-L450

Assuming I'm right that this is getting passed the .a file, it seems that that's our problem -- the .a won't have an ELF header here like a .o would.

I'm not sure what the best way to validate this hypothesis is, but I think fixing this is a pre-req for the mainline build system changes. Do we need to modify modpost in some way? Is there a way to convert our .a to a .o (ar -x && $CC *.o -o final.o)?

[alex]

Ok, confirmed that if I take the .a, do:

$ mkdir z
$ cd z
$ ar -x ../target/x86_64-linux-kernel-module/debug/libhello_world.a
$ ld -relocatable *.o -o final.o

and then replace target/x86_64-linux-kernel-module/debug/libhello_world.a with z/final.o in helloworld-objs in the Makefile then everything builds correctly, without the EXTRA_LDFLAGS.

@geofft do you see a clear way to integrate this into our existing system, or should I just try to make this a part of the new build system in alex/linux#3?

[geofft]

#175 looks like a clean way to do it - namely ld -r -o foo.o --whole-archive foo.a seems to have the same behavior as unpacking the archive and ld -r'ing its constituent files. The rest is Inspired by the approach you're taking at alex/linux#3. (You might want to take the --whole-archive trick though.)

[geofft]

Now that #175 merged, can we close this?