Viktor Malík

Hm, my output on latest master (unless I did some mistake): ``` # src/bpftrace test.bt -v [...] BPF program is too large. Processed 1000001 insn processed 1000001 insns (limit 1000000)...

Strange, I'm using the same LLVM with pretty much the same features: ``` # bpftrace --info System OS: Linux 6.8.8-200.fc39.x86_64 #1 SMP PREEMPT_DYNAMIC Sat Apr 27 17:42:13 UTC 2024 Arch:...

I had a closer look at the issue that I'm observing. From verifier log: ``` [...] 31: (85) call bpf_for_each_map_elem#164 ; R0_w=scalar() R7=0 R10=fp0 fp-24=0 fp-32=mmmmmmmm [...] from 31 to...

> Nah, there's just some confusion. The BPF headers require a [TARGET_ARCH_xyz](https://docs.ebpf.io/ebpf-library/libbpf/ebpf/PT_REGS_IP/) definition, e.g. `TARGET_ARCH_x86`. > > This is giving the `Arch` API a way to declare per-arch definitions that...

I generally like this for the already stated purposes (readability, chaining, etc.) but we must realize that allowing this for all functions will allow things like: ``` "pid: %d\n".printf(pid); 4.bswap();...

> What if instead of calling release automatically in the relevant code paths, we just stick all these pointers into a hidden map? This is an excellent idea and very...

> I'm not sure I understand. If these are in an LRU map for example, do they get automatically dec-refed on eviction? Regardless, if the thing ends up in two...

Summarizing the discussion from today's office hours: At the moment, we;re putting this feature on hold since we need to clarify some things and think more about the implementation course...

> It looks like this kfunc has a simple interface, so it might be a good candidate for our first kfunc. @viktormalik Hm, I already looked at it but I...

@amscanne you brought this up in the office hours, would you mind if I took it (unless you already started working on it)?