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;
User-Agent: Cyrus-JMAP/3.5.0-alpha0-448-gae190416c7-fm-20210505.004-gae190416
Mime-Version: 1.0
Message-Id: <861e2ac3-b56b-4879-9a78-e20bac4906f1@www.fastmail.com>
In-Reply-To: <CAJvTdKkTSv7zc2eT1Ni6+k=V_faCCWmMo5oNunQQt-P7FkON-Q@mail.gmail.com>
References: <20210523193259.26200-1-chang.seok.bae@intel.com>
 <20210523193259.26200-22-chang.seok.bae@intel.com>
 <8c8d91ae-5a3b-9523-725d-134840102df7@kernel.org>
 <CAJvTdKnU6g2H-R67G8LCs4cs0_NqnOzoKbC3J+iNGayFE5RGMQ@mail.gmail.com>
 <5a645958-10ab-46e7-9650-e5746a8d2276@www.fastmail.com>
 <CAJvTdKkTSv7zc2eT1Ni6+k=V_faCCWmMo5oNunQQt-P7FkON-Q@mail.gmail.com>
Date:   Mon, 24 May 2021 20:44:36 -0700
From:   "Andy Lutomirski" <luto@kernel.org>
To:     "Len Brown" <lenb@kernel.org>
Cc:     "Bae, Chang Seok" <chang.seok.bae@intel.com>,
        "Borislav Petkov" <bp@suse.de>,
        "Thomas Gleixner" <tglx@linutronix.de>,
        "Ingo Molnar" <mingo@kernel.org>,
        "the arch/x86 maintainers" <x86@kernel.org>,
        "Brown, Len" <len.brown@intel.com>,
        "Dave Hansen" <dave.hansen@intel.com>,
        "Liu, Jing2" <jing2.liu@intel.com>,
        "Shankar, Ravi V" <ravi.v.shankar@intel.com>,
        "Linux Kernel Mailing List" <linux-kernel@vger.kernel.org>
Subject: Re: [PATCH v5 21/28] x86/fpu/amx: Initialize child's AMX state
Content-Type: text/plain;charset=utf-8
Content-Transfer-Encoding: quoted-printable
Precedence: bulk


On Mon, May 24, 2021, at 11:21 AM, Len Brown wrote:
> On Mon, May 24, 2021 at 2:14 PM Andy Lutomirski <luto@kernel.org> wrot=
e:
>=20
> > What does VOLATILE mean in this context?
>=20
> Volatile means caller-saved.

Just like every other extended math register except some XMMs on Windows=
. (Thanks you so, so much, Microsoft, for screwing this up, and thank yo=
u Intel for indulging Microsoft.)

>=20
> Volatile registers can not be used for globals, static, or for
> parameter passing.
>=20
> ie. By the time the callee is running, they must be assumed to be inva=
lid.

Callees can=E2=80=99t assume anything about any of the registers unless =
explicitly specified. TILE is no different from RBP or XMM in this regar=
d.

>=20
> This means that any routine, including the target of a system call,
> such as fork/clone, can't assume that any data exists in these
> registers.
>=20

If we actually believe this, then we should clear xmm, ymm, zmm, etc on =
every system call.  Barring that, let=E2=80=99s come up with reasonable =
semantics, document it, and implement it.