Re: net/bpfjit/t_mbuf tests failing on amd64

[Alexander Nasonov <alnsn%yandex.ru@localhost>]

Alexander Nasonov wrote:
> Hi Andreas,
> 
> Andreas Gustafsson wrote:
> > Hi alnsn,
> > 
> > The bpfjit_mbuf_ldb_ind, bpfjit_mbuf_ldh_ind, bpfjit_mbuf_ldw_ind test
> > cases of the net/bpfjit/t_mbuf test are failing on amd64 since the
> > following commits:
> > 
> >   2014.07.22.08.20.08 alnsn src/sys/net/bpfjit.c 1.29
> >   2014.07.22.08.29.51 alnsn src/sys/net/bpfjit.c 1.30
> 
> I noticed it too this morning but I got a chance to look at the
> failures only now. The bpfjit_mbuf_ldw_ind test generates a strange
> code, I'm not sure sljit is supposed to generate such code. I'll
> make a standalone test and send it to Zoltan.

It looks like sljit generates bad code when k=0 here:

        sljit_emit_op2(compiler,
            SLJIT_ADD,
            SLJIT_SCRATCH_REG2, 0,
            BJ_XREG, 0,
            SLJIT_IMM, k);

If k is zero, the new bpfjit version emits the mov instruction:

        sljit_emit_op1(compiler,
            SLJIT_MOV,
            SLJIT_SCRATCH_REG2, 0,
            BJ_XREG, 0);

The tests should be fixed now but I asked Zoltan to take a look at
the problem.

Alex