Re: [PATCH bpf-next v2] bpf: Change value of MAX_TAIL_CALL_CNT from 32 to 33

On 9/11/21 3:56 AM, Tiezhu Yang wrote:

In the current code, the actual max tail call count is 33 which is greater
than MAX_TAIL_CALL_CNT (defined as 32), the actual limit is not consistent
with the meaning of MAX_TAIL_CALL_CNT, there is some confusion and need to
spend some time to think about the reason at the first glance.

We can see the historical evolution from commit 04fd61ab36ec ("bpf: allow
bpf programs to tail-call other bpf programs") and commit f9dabe016b63
("bpf: Undo off-by-one in interpreter tail call count limit").

In order to avoid changing existing behavior, the actual limit is 33 now,
this is reasonable.

After commit 874be05f525e ("bpf, tests: Add tail call test suite"), we can
see there exists failed testcase.

On all archs when CONFIG_BPF_JIT_ALWAYS_ON is not set:
  # echo 0 > /proc/sys/net/core/bpf_jit_enable
  # modprobe test_bpf
  # dmesg | grep -w FAIL
  Tail call error path, max count reached jited:0 ret 34 != 33 FAIL

On some archs:
  # echo 1 > /proc/sys/net/core/bpf_jit_enable
  # modprobe test_bpf
  # dmesg | grep -w FAIL
  Tail call error path, max count reached jited:1 ret 34 != 33 FAIL

So it is necessary to change the value of MAX_TAIL_CALL_CNT from 32 to 33,
then do some small changes of the related code.

With this patch, it does not change the current limit 33, MAX_TAIL_CALL_CNT
can reflect the actual max tail call count, the tailcall selftests can work
well, and also the above failed testcase in test_bpf can be fixed for the
interpreter (all archs) and the JIT (all archs except for x86).

  # uname -m
  x86_64
  # echo 1 > /proc/sys/net/core/bpf_jit_enable
  # modprobe test_bpf
  # dmesg | grep -w FAIL
  Tail call error path, max count reached jited:1 ret 33 != 34 FAIL

Could you also state in here which archs you have tested with this change? I
presume /every/ arch which has a JIT?

Signed-off-by: Tiezhu Yang <yangtiezhu@loongson.cn>
---

v2:
   -- fix the typos in the commit message and update the commit message.
   -- fix the failed tailcall selftests for x86 jit.
      I am not quite sure the change on x86 is proper, with this change,
      tailcall selftests passed, but tailcall limit test in test_bpf.ko
      failed, I do not know the reason now, I think this is another issue,
      maybe someone more versed in x86 jit could take a look.

There should be a series from Johan coming today with regards to test_bpf.ko
that will fix the "tail call error path, max count reached" test which had an
assumption in that R0 would always be valid for the fall-through and could be
passed to the bpf_exit insn whereas it is not guaranteed and verifier, for
example, forbids a subsequent access to R0 w/o reinit. For your testing, I
would suggested to recheck once this series is out.

  arch/arm/net/bpf_jit_32.c         | 11 ++++++-----
  arch/arm64/net/bpf_jit_comp.c     |  7 ++++---
  arch/mips/net/ebpf_jit.c          |  4 ++--
  arch/powerpc/net/bpf_jit_comp32.c |  4 ++--
  arch/powerpc/net/bpf_jit_comp64.c | 12 ++++++------
  arch/riscv/net/bpf_jit_comp32.c   |  4 ++--
  arch/riscv/net/bpf_jit_comp64.c   |  4 ++--
  arch/sparc/net/bpf_jit_comp_64.c  |  8 ++++----
  arch/x86/net/bpf_jit_comp.c       | 10 +++++-----
  include/linux/bpf.h               |  2 +-
  kernel/bpf/core.c                 |  4 ++--
  11 files changed, 36 insertions(+), 34 deletions(-)

[...]

[...]

Why such inconsistency to e.g. above with arm64 case but also compared to
x86 32 bit which uses JAE? If so, we should cleanly follow the reference
implementation (== interpreter) _everywhere_ and _not_ introduce additional
variants/implementations across JITs.

_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel