`LINUX_KERNEL_VERSION` has no register bounds about ebpf

[eiffel-fl]

I am investigating on this bug since some time ago, I think the problem comes from how we relocate the instruction in cilium/ebpf.
I will investigate more and post my findings here, as I need to understand more how the thing is handled in libbpf, until I finally come up with a solution.

[eiffel-fl]

I just tested 73acad5 and everything is fine with this commit 😅:

; if (LINUX_KERNEL_VERSION < KERNEL_VERSION(5, 11, 0))
10: (18) r1 = 0xffffacb380988000      ; R1_w=map_value(off=0,ks=4,vs=4,imm=0)
12: (61) r1 = *(u32 *)(r1 +0)         ; R1_w=332561
13: (67) r1 <<= 32                    ; R1_w=1428338618925056
14: (c7) r1 s>>= 32                   ; R1_w=332561

[lmb]

Very strange. That commit started copying MapSpec correctly, maybe that was it? No idea what else it could be.

[eiffel-fl]

I sadly do not know the root cause of the fix but I would say this is because we now associated a map with the load:

73acad5#diff-981ef293a9c93614e843135eb5b207f951babe5604aee33c72f66567ccaa01deR331