| Message ID | 162388400488.151936.1658153981415911010.stgit@john-XPS-13-9370 (mailing list archive) |
|---|---|
| Headers | show |
| Series | BPF fixes mixed tail and bpf2bpf calls | expand |
On Wed, Jun 16, 2021 at 7:34 PM John Fastabend <john.fastabend@gmail.com> wrote: > > We recently tried to use mixed programs that have both tail calls and > subprograms, but it needs the attached fixes. > > Also added a new test case tailcall_bpf2bpf_5 that simply runs the > previous test case tailcall_bpf2bpf_4 and adds some "noise". The > noise here is just a bunch of map calls to get the verifier to insert > instructions and cause code movement plus it forces used_maps logic > to be used. Originally, I just extended bpf2bpf_4 directly, but if I > got the feedback correct it seems the preference is to have another > test case for this specifically. > > With attached patches our programs are happily running with mixed > subprograms and tailcalls. > > Thanks, > John > > --- Would be nice to include bpf@vger.kernel.org as well. I bet not everyone interested in BPF follows netdev@vger closely. > > John Fastabend (4): > bpf: Fix null ptr deref with mixed tail calls and subprogs > bpf: map_poke_descriptor is being called with an unstable poke_tab[] > bpf: track subprog poke correctly > bpf: selftest to verify mixing bpf2bpf calls and tailcalls with insn patch > > > include/linux/bpf.h | 1 + > kernel/bpf/core.c | 6 ++-- > kernel/bpf/verifier.c | 36 ++++++++++++++----- > .../selftests/bpf/prog_tests/tailcalls.c | 36 +++++++++++++------ > .../selftests/bpf/progs/tailcall_bpf2bpf4.c | 20 ++++++++++- > 5 files changed, 77 insertions(+), 22 deletions(-) > > -- >
We recently tried to use mixed programs that have both tail calls and subprograms, but it needs the attached fixes. Also added a new test case tailcall_bpf2bpf_5 that simply runs the previous test case tailcall_bpf2bpf_4 and adds some "noise". The noise here is just a bunch of map calls to get the verifier to insert instructions and cause code movement plus it forces used_maps logic to be used. Originally, I just extended bpf2bpf_4 directly, but if I got the feedback correct it seems the preference is to have another test case for this specifically. With attached patches our programs are happily running with mixed subprograms and tailcalls. Thanks, John --- John Fastabend (4): bpf: Fix null ptr deref with mixed tail calls and subprogs bpf: map_poke_descriptor is being called with an unstable poke_tab[] bpf: track subprog poke correctly bpf: selftest to verify mixing bpf2bpf calls and tailcalls with insn patch include/linux/bpf.h | 1 + kernel/bpf/core.c | 6 ++-- kernel/bpf/verifier.c | 36 ++++++++++++++----- .../selftests/bpf/prog_tests/tailcalls.c | 36 +++++++++++++------ .../selftests/bpf/progs/tailcall_bpf2bpf4.c | 20 ++++++++++- 5 files changed, 77 insertions(+), 22 deletions(-) --