Remediation for verifier regression in kernels ~6.8 and greater

[javierhonduco]

Some days ago we put a fix for arm64 kernels as a verifier regression was introduced causing loading errors
commit.

Turns out this is not just affecting arm64, as it just happened it's that my arm64 VM was running a more up to date kernel. This regression affects x86 as well.

This commit changes the remediation to unrolloing the binary search loop and adds a kernel regression test. I haven't measured the performance implication of doing this vs the previous remediation. This is out of the scope of this PR and it's something I will work on later on.

When I have spare cycles I should report this upstream.

The error

292: (bf) r6 = r3                     ; R3_w=scalar(id=52209,smin=umin=smin32=umin32=2,smax=umax=smax32=umax32=0x1869e,var_off=(0x0; 0x1ffff)) R6_w=scalar(id=52209,smin=umin=smin32=umin32=2,smax=umax=smax32=umax32=0x1869e,var_off=(0x0; 0x1ffff))
293: (bf) r3 = r2                     ; R2_w=scalar(id=52208,smin=umin=smin32=umin32=1,smax=umax=smax32=umax32=0x1869d,var_off=(0x0; 0x1ffff)) R3_w=scalar(id=52208,smin=umin=smin32=umin32=1,smax=umax=smax32=umax32=0x1869d,var_off=(0x0; 0x1ffff))
294: (2d) if r4 > r8 goto pc+1 296: R0_w=scalar() R1=11 R2_w=scalar(id=52208,smin=umin=smin32=umin32=1,smax=umax=smax32=umax32=0x1869d,var_off=(0x0; 0x1ffff)) R3_w=scalar(id=52208,smin=umin=smin32=umin32=1,smax=umax=smax32=umax32=0x1869d,var_off=(0x0; 0x1ffff)) R4_w=scalar(umin=9,umax=0xfffffffffffffffa) R5_w=scalar() R6_w=scalar(id=52209,smin=umin=smin32=umin32=2,smax=umax=smax32=umax32=0x1869e,var_off=(0x0; 0x1ffff)) R7=scalar(id=52027,smin=umin=smin32=umin32=2,smax=umax=smax32=umax32=0x1869e,var_off=(0x0; 0x1ffff)) R8=scalar(id=24520,umin=8,umax=0xfffffffffffffff9) R9_w=scalar(smin=smin32=0,smax=umax=smax32=umax32=0xff0000,var_off=(0x0; 0xff0000)) R10=fp0 fp-16=mmmmmmmm fp-24=mmmmmmmm fp-32=mmmmmmmm fp-40=mmmmmmmm fp-48=map_value(map=unwind_tables,ks=8,vs=1400000) fp-56=6 fp-64=scalar(id=52027,smin=umin=smin32=umin32=2,smax=umax=smax32=umax32=0x1869e,var_off=(0x0; 0x1ffff)) fp-72=map_value(map=heap,ks=4,vs=1080) fp-80=map_value(map=heap,ks=4,vs=1080) fp-88=map_value(map=exec_mappings,ks=16,vs=32) fp-96=mmmmmmmm fp-104=ctx() fp-112=mmmmmmmm fp-120=scalar(smin=0,smax=umax=0xffffffff,var_off=(0x0; 0xffffffff)) fp-128=scalar(smin=smin32=0,smax=umax=smax32=umax32=255,var_off=(0x0; 0xff)) fp-136=scalar(smin=smin32=0,smax=umax=smax32=umax32=255,var_off=(0x0; 0xff))
; if (table->rows[mid].pc <= pc) {
296: (2d) if r4 > r8 goto pc-83 214: R0_w=scalar() R1=11 R2_w=scalar(id=52208,smin=umin=smin32=umin32=1,smax=umax=smax32=umax32=0x1869d,var_off=(0x0; 0x1ffff)) R3_w=scalar(id=52208,smin=umin=smin32=umin32=1,smax=umax=smax32=umax32=0x1869d,var_off=(0x0; 0x1ffff)) R4_w=scalar(umin=9,umax=0xfffffffffffffffa) R5_w=scalar() R6_w=scalar(id=52209,smin=umin=smin32=umin32=2,smax=umax=smax32=umax32=0x1869e,var_off=(0x0; 0x1ffff)) R7=scalar(id=52027,smin=umin=smin32=umin32=2,smax=umax=smax32=umax32=0x1869e,var_off=(0x0; 0x1ffff)) R8=scalar(id=24520,umin=8,umax=0xfffffffffffffff9) R9_w=scalar(smin=smin32=0,smax=umax=smax32=umax32=0xff0000,var_off=(0x0; 0xff0000)) R10=fp0 fp-16=mmmmmmmm fp-24=mmmmmmmm fp-32=mmmmmmmm fp-40=mmmmmmmm fp-48=map_value(map=unwind_tables,ks=8,vs=1400000) fp-56=6 fp-64=scalar(id=52027,smin=umin=smin32=umin32=2,smax=umax=smax32=umax32=0x1869e,var_off=(0x0; 0x1ffff)) fp-72=map_value(map=heap,ks=4,vs=1080) fp-80=map_value(map=heap,ks=4,vs=1080) fp-88=map_value(map=exec_mappings,ks=16,vs=32) fp-96=mmmmmmmm fp-104=ctx() fp-112=mmmmmmmm fp-120=scalar(smin=0,smax=umax=0xffffffff,var_off=(0x0; 0xffffffff)) fp-128=scalar(smin=smin32=0,smax=umax=smax32=umax32=255,var_off=(0x0; 0xff)) fp-136=scalar(smin=smin32=0,smax=umax=smax32=umax32=255,var_off=(0x0; 0xff))
; LOG("========== left %llu right %llu (shard index %d)", left, right,
214: (18) r9 = 0xbadfadbadfadbad      ; R9_w=0xbadfadbadfadbad
; for (int i = 0; i < MAX_BINARY_SEARCH_DEPTH; i++) {
216: (07) r1 += -1                    ; R1_w=10
217: (bf) r2 = r1                     ; R1_w=10 R2_w=10
218: (67) r2 <<= 32
BPF program is too large. Processed 1000001 insn
processed 1000001 insns (limit 1000000) max_states_per_insn 29 total_states 18306 peak_states 2761 mark_read 106
-- END PROG LOAD LOG --
libbpf: prog 'dwarf_unwind': failed to load: -7
libbpf: failed to load object 'profiler_bpf'
libbpf: failed to load BPF skeleton 'profiler_bpf': -7
thread 'main' panicked at src/profiler.rs:159:36:
load skel: Error: Argument list too long (os error 7)
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace

Test Plan

Added a regression test.