netdev - Re: [PATCH net] bpf: do not use reciprocal divide

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Date:	Wed, 15 Jan 2014 16:26:23 +0100
From:	Martin Schwidefsky <schwidefsky@...ibm.com>
To:	Eric Dumazet <eric.dumazet@...il.com>
Cc:	Heiko Carstens <heiko.carstens@...ibm.com>,
	Hannes Frederic Sowa <hannes@...essinduktion.org>,
	netdev@...r.kernel.org, dborkman@...hat.com,
	darkjames-ws@...kjames.pl, Mircea Gherzan <mgherzan@...il.com>,
	Russell King <rmk+kernel@....linux.org.uk>,
	Matt Evans <matt@...abs.org>
Subject: Re: [PATCH net] bpf: do not use reciprocal divide

On Wed, 15 Jan 2014 06:21:56 -0800
Eric Dumazet <eric.dumazet@...il.com> wrote:

> On Wed, 2014-01-15 at 11:51 +0100, Heiko Carstens wrote:
> > On Wed, Jan 15, 2014 at 09:13:22AM +0100, Martin Schwidefsky wrote:
> > > On Wed, 15 Jan 2014 09:00:07 +0100
> > > Heiko Carstens <heiko.carstens@...ibm.com> wrote:
> > > 
> > > > On Tue, Jan 14, 2014 at 11:02:41PM -0800, Eric Dumazet wrote:
> > > > > diff --git a/arch/s390/net/bpf_jit_comp.c b/arch/s390/net/bpf_jit_comp.c
> > > > > index 16871da37371..e349dc7d0992 100644
> > > > > --- a/arch/s390/net/bpf_jit_comp.c
> > > > > +++ b/arch/s390/net/bpf_jit_comp.c
> > > > > @@ -371,11 +371,11 @@ static int bpf_jit_insn(struct bpf_jit *jit, struct sock_filter *filter,
> > > > >  		/* dr %r4,%r12 */
> > > > >  		EMIT2(0x1d4c);
> > > > >  		break;
> > > > > -	case BPF_S_ALU_DIV_K: /* A = reciprocal_divide(A, K) */
> > > > > -		/* m %r4,<d(K)>(%r13) */
> > > > > -		EMIT4_DISP(0x5c40d000, EMIT_CONST(K));
> > > > > -		/* lr %r5,%r4 */
> > > > > -		EMIT2(0x1854);
> > > > > +	case BPF_S_ALU_DIV_K: /* A /= K */
> > > > > +		/* lhi %r4,0 */
> > > > > +		EMIT4(0xa7480000);
> > > > > +		/* d %r4,<d(K)>(%r13) */
> > > > > +		EMIT4_DISP(0x5d40d000, EMIT_CONST(K));
> > > > >  		break;
> > > > 
> > > > The s390 part looks good.
> > > 
> > > Does it? The divide instruction is signed, for the special
> > > case of K==1 this can now cause an exception if the quotient
> > > gets too large. We should add a check for K==1 and do nothing
> > > in this case. With a divisor of at least 2 the result will
> > > stay in the limit.
> > 
> > Indeed. That's quite subtle.
> 
> net/core/filter.c does :
> 
> A /= K;
> 
> Why is this working in generic code (if K == 1), not in s390 one ?

The C compiler naturally can to a u32/u32 division, it either uses
the "dlr" instruction which is unsigned, or uses a call to a function
to do the u32/u32 math. See the lovely code in arch/s390/lib/div64.c
for the kernel version of that code.

-- 
blue skies,
   Martin.

"Reality continues to ruin my life." - Calvin.

--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html