Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: quoted-printable
From:   Qian Cai <cai@lca.pw>
Mime-Version: 1.0 (1.0)
Subject: Re: [RFC PATCH] iommu/amd: fix a race in fetch_pte()
Date:   Sat, 18 Apr 2020 09:01:35 -0400
Message-Id: <57CBF6B2-4745-4E36-9AA5-7E0876E3DA8F@lca.pw>
References: <20200418121022.GA6113@8bytes.org>
Cc:     iommu@lists.linux-foundation.org, linux-kernel@vger.kernel.org
In-Reply-To: <20200418121022.GA6113@8bytes.org>
To:     Joerg Roedel <joro@8bytes.org>
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk


> On Apr 18, 2020, at 8:10 AM, Joerg Roedel <joro@8bytes.org> wrote:
>=20
> Yes, your patch still looks racy. You need to atomically read
> domain->pt_root to a stack variable and derive the pt_root pointer and
> the mode from that variable instead of domain->pt_root directly. If you
> read the domain->pt_root twice there could still be an update between
> the two reads.
> Probably the lock in increase_address_space() can also be avoided if
> pt_root is updated using cmpxchg()?

Hard to tell without testing further. I=E2=80=99ll leave that optimization i=
n the future, and focus on fixing those races first.=