[DTrace-devel] [PATCH] bpf: work around BPF verifier issues with AND and 2-reg branch conditions
Kris Van Hees
kris.van.hees at oracle.com
Tue Apr 30 19:46:33 UTC 2024
On Tue, Apr 30, 2024 at 12:54:12PM -0400, Eugene Loh wrote:
> Reviewed-by: Eugene Loh <eugene.loh at oracle.com>
> two nits...
>
> On 4/30/24 10:27, Kris Van Hees wrote:
> > Signed-off-by: Kris Van Hees <kris.van.hees at oracle.com>
> > ---
> > bpf/strchr.S | 22 ++++++++++++----------
> > 1 file changed, 12 insertions(+), 10 deletions(-)
> >
> > diff --git a/bpf/strchr.S b/bpf/strchr.S
> > index 8365d9ce..42c6430f 100644
> > --- a/bpf/strchr.S
> > +++ b/bpf/strchr.S
> > @@ -26,8 +26,10 @@
> > *
> > * // xor the char with every byte; a match results in NULL byte
> > * r4 = roundup(r6, 8);
> > - * for (r3 = 0; r3 < r4; r3 += 8)
> > - * ((uint64_t *)dst)[r3] ^= c;
> > + * do {
> > + * r4 -= 8;
> > + * ((uint64_t *)dst)[r4] ^= c;
> > + * } while (r4 > 8);
>
> I think that is wrong, but in any case out of step with the actual code.
> How about either s/8/7/ or (imho better but then needs change in
> corresponding code) s/>/>=/.
Sure, we can use >= 8. I'll apply the patch with that change.
> > *
> > * // put a safeguard in place, then look for that NULL byte
> > * dst[r6] = '\0';
> > @@ -72,24 +74,24 @@ dt_strchr :
> > call BPF_FUNC_probe_read_str /* r6 = bpf_probe_read_str(dst, STRSZ, src) */
> > mov %r6, %r0
> > - jsle %r6, 0, .Lerror
> > + jsle %r6, 1, .Lerror
> > sub %r6, 1 /* r6-- */
> > mov %r4, %r6 /* r4 = roundup(r6, 8) */
> > add %r4, 7
> > - and %r4, -8
> > + rsh %r4, 3
> > + lsh %r4, 3
>
> How about a comment on the << >> 3 thing so that some future maintainer more
> easily understands why this looks so funny.
I think that the >>= followed by <<= is a common enough idiom to mask out the
lower 3 bits that it does not warrant an explicit commit, especially as part
of a roundup() implementation.
> > ldxdw %r1, [%fp+-16]
> > - mov %r3, 0
> > -.Lloop: /* for (r3 = 0; r3 < r4; r3 += 8) */
> > +.Lloop: /* do { */
> > + sub %r4, 8 /* r4 -= 8; */
> > ldxdw %r5, [%fp+-24]
> > - add %r5, %r3
> > + add %r5, %r4
> > ldxdw %r0, [%r5+0]
> > - xor %r0, %r1 /* ((uint64_t *)dst)[r3] ^= c; */
> > + xor %r0, %r1 /* ((uint64_t *)dst)[r4] ^= c; */
> > stxdw [%r5+0], %r0
> > - add %r3, 8
> > - jlt %r3, %r4, .Lloop
> > + jgt %r4, 7, .Lloop /* } while (r4 > 7); */
> > ldxdw %r2, [%fp+-24]
> > add %r2, %r6
More information about the DTrace-devel
mailing list