Received-SPF: pass (google.com: domain of linux-kernel+bounces-3558-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:45e3:2400::1 as permitted sender) client-ip=2604:1380:45e3:2400::1;
Message-ID: <59621e88c5c29bdff8bc06f68b02b2c7a420a09a.camel@linux.ibm.com>
Subject: Re: [PATCH] KVM: s390: selftest: memop: Fix undefined behavior
From: Nina Schoetterl-Glausch <nsg@linux.ibm.com>
To: Claudio Imbrenda <imbrenda@linux.ibm.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>, Shuah Khan <shuah@kernel.org>,
        Janosch Frank <frankja@linux.ibm.com>,
        Christian Borntraeger
 <borntraeger@linux.ibm.com>,
        kvm@vger.kernel.org, linux-kselftest@vger.kernel.org,
        David Hildenbrand <david@redhat.com>, linux-kernel@vger.kernel.org
Date: Mon, 18 Dec 2023 13:18:14 +0100
In-Reply-To: <20231215180206.740df738@p-imbrenda>
References: <20231215161125.943551-1-nsg@linux.ibm.com>
	 <20231215180206.740df738@p-imbrenda>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
User-Agent: Evolution 3.48.4 (3.48.4-1.fc38) 
Precedence: bulk
MIME-Version: 1.0

On Fri, 2023-12-15 at 18:02 +0100, Claudio Imbrenda wrote:
> On Fri, 15 Dec 2023 17:11:25 +0100
> Nina Schoetterl-Glausch <nsg@linux.ibm.com> wrote:
>=20
> > If an integer's type has x bits, shifting the integer left by x or more
> > is undefined behavior.
> > This can happen in the rotate function when attempting to do a rotation
> > of the whole value by 0.
>=20
> is 0 the only problematic value? because in that case...=20
>=20
> >=20
> > Fixes: 0dd714bfd200 ("KVM: s390: selftest: memop: Add cmpxchg tests")
> > Signed-off-by: Nina Schoetterl-Glausch <nsg@linux.ibm.com>
> > ---
> >  tools/testing/selftests/kvm/s390x/memop.c | 8 +++++---
> >  1 file changed, 5 insertions(+), 3 deletions(-)
> >=20
> > diff --git a/tools/testing/selftests/kvm/s390x/memop.c b/tools/testing/=
selftests/kvm/s390x/memop.c
> > index bb3ca9a5d731..2eba9575828e 100644
> > --- a/tools/testing/selftests/kvm/s390x/memop.c
> > +++ b/tools/testing/selftests/kvm/s390x/memop.c
> > @@ -485,11 +485,13 @@ static bool popcount_eq(__uint128_t a, __uint128_=
t b)
> > =20
> >  static __uint128_t rotate(int size, __uint128_t val, int amount)
> >  {
> > -	unsigned int bits =3D size * 8;
> > +	unsigned int left, right, bits =3D size * 8;
> > =20
>=20
> ...why not just:
>=20
> if (!amount)
> 	return val;
>=20
> ?

That works if you move it one statement down (128 would also trigger UB).
% 128 does the trick, is branchless and there is a bit of a symmetry going
on between right and left.
But I can use an early return if you want.

>=20
> > -	amount =3D (amount + bits) % bits;
> > +	right =3D (amount + bits) % bits;
> > +	/* % 128 prevents left shift UB if size =3D=3D 16 && right =3D=3D 0 *=
/
> > +	left =3D (bits - right) % 128;
> >  	val =3D cut_to_size(size, val);
> > -	return (val << (bits - amount)) | (val >> amount);
> > +	return (val << left) | (val >> right);
> >  }
> > =20
> >  const unsigned int max_block =3D 16;
> >=20
> > base-commit: 305230142ae0637213bf6e04f6d9f10bbcb74af8
>=20