Dear Alexei,
Thanks for the feedback. I have two questions:
1) My understanding is that if the "noinstr" attribute is to be
applied to a given function, any functions that are called within it
must be explicitly marked as "noinstr" or "__always_inline". This
means that marking "bpf_common_lru_pop_free" and
"bpf_common_lru_push_free" as "no instr" would involve marking many
other additional functions (we can count about 30 of them). Would this
be acceptable?
2) Spinner has been finding several more deadlock bugs in ebpf helper
functions. And it is likely to find more as we keep analyzing the ebpf
subsystem. We are wondering how you suggest we report the bugs?
Oftentimes, we are not sure how to best patch the bugs and hence we
prefer to just report the bugs alongside PoCs and lockdep splats. If
it helps, we can batch the bugs that look similar in one report.
Please advise.
Thanks,
Priya
On Sat, Sep 28, 2024 at 3:25 AM Alexei Starovoitov
<alexei.starovoitov@xxxxxxxxx> wrote:
>
> On Sat, Sep 28, 2024 at 4:02 AM Priya Bala Govindasamy <pgovind2@xxxxxxx> wrote:
> >
> >
> > SEC("kprobe/bpf_lru_pop_free+0x352")
> > int test_prog2(void *ctx){
>
> Instead of resending old issue and/or claiming new bug
> please send a patch that makes these functions "noinstr".
