s390/bpf: Emit a barrier for BPF_FETCH instructions
authorIlya Leoshkevich <iii@linux.ibm.com>
Tue, 7 May 2024 00:02:49 +0000 (02:02 +0200)
committerAlexei Starovoitov <ast@kernel.org>
Sun, 12 May 2024 23:55:57 +0000 (16:55 -0700)
commit68378982f0b21de02ac3c6a11e2420badefcb4bc
tree21098308f0f3465e5e38a20e7bd06c2735b7dad5
parent55302bc1ca64160fec4dfa25e52142691ecb5dcd
s390/bpf: Emit a barrier for BPF_FETCH instructions

BPF_ATOMIC_OP() macro documentation states that "BPF_ADD | BPF_FETCH"
should be the same as atomic_fetch_add(), which is currently not the
case on s390x: the serialization instruction "bcr 14,0" is missing.
This applies to "and", "or" and "xor" variants too.

s390x is allowed to reorder stores with subsequent fetches from
different addresses, so code relying on BPF_FETCH acting as a barrier,
for example:

  stw [%r0], 1
  afadd [%r1], %r2
  ldxw %r3, [%r4]

may be broken. Fix it by emitting "bcr 14,0".

Note that a separate serialization instruction is not needed for
BPF_XCHG and BPF_CMPXCHG, because COMPARE AND SWAP performs
serialization itself.

Fixes: ba3b86b9cef0 ("s390/bpf: Implement new atomic ops")
Reported-by: Puranjay Mohan <puranjay12@gmail.com>
Closes: https://lore.kernel.org/bpf/mb61p34qvq3wf.fsf@kernel.org/
Signed-off-by: Ilya Leoshkevich <iii@linux.ibm.com>
Reviewed-by: Puranjay Mohan <puranjay@kernel.org>
Link: https://lore.kernel.org/r/20240507000557.12048-1-iii@linux.ibm.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
arch/s390/net/bpf_jit_comp.c