Alexei Starovoitov <alexei.starovoitov@xxxxxxxxx> writes: > On Tue, Dec 17, 2024 at 3:49 PM Alexei Starovoitov > <alexei.starovoitov@xxxxxxxxx> wrote: >> >> On Mon, Dec 2, 2024 at 4:42 AM Puranjay Mohan <puranjay@xxxxxxxxxx> wrote: >> > >> > Alexei Starovoitov <alexei.starovoitov@xxxxxxxxx> writes: >> > >> > > Puranjay, Andrii and All, >> > > >> > > looks like if (irqs_disabled()) is not enough. >> > > Should we change it to preemptible() ? >> > > >> > > It will likely make it async all the time, >> > > but in this it's an ok trade off? >> > > >> > >> > Yes, as BPF programs can run in all kinds of contexts. >> > >> > We should replace 'if (irqs_disabled())' with 'if (!preemptible())' >> > >> > because the definition is: >> > >> > #define preemptible() (preempt_count() == 0 && !irqs_disabled()) >> > >> > and we need if ((preempt_count() != 0) || irqs_disabled()), in both >> > these cases we want to make it async. >> > >> > I will try to test the fix as Syzbot has now found a reproducer. >> >> Puranjay, >> >> Any progress on a patch ? > > ping. Hi Alexei, Sorry for being AWOL. I was on a long vacation in India and just got back. Here is the patch to fix this: https://lore.kernel.org/all/20250115103647.38487-1-puranjay@xxxxxxxxxx/ Thanks, Puranjay #syz test: https://github.com/puranjaymohan/bpf.git bpf_preemt_fix
Attachment:
signature.asc
Description: PGP signature
