Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
MIME-Version: 1.0
References: <CAPcxDJ7nriZEJHF6dMPR7tQ+dGuueyRjw1NC+4CbjxiAT_S+ZA@mail.gmail.com>
 <20220208150945.266978-1-juew@google.com> <Ygwka++3eipjQzB2@zn.tnic>
 <YgwnqTc8FGG3orcE@agluck-desk3.sc.intel.com> <YgzRsfWOmqkNiNI7@zn.tnic>
In-Reply-To: <YgzRsfWOmqkNiNI7@zn.tnic>
From:   Jue Wang <juew@google.com>
Date:   Wed, 16 Feb 2022 07:50:24 -0800
Message-ID: <CAPcxDJ4=iknzYbN=b2oQxvJyODkQ_MWEj1wkKEuGwf4HQ3aPZA@mail.gmail.com>
Subject: Re: [PATCH] x86/mce: Add workaround for SKX/CLX/CPX spurious machine checks
To:     Borislav Petkov <bp@alien8.de>
Cc:     "Luck, Tony" <tony.luck@intel.com>, x86@kernel.org,
        linux-kernel@vger.kernel.org, patches@lists.linux.dev
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

On Wed, Feb 16, 2022 at 2:28 AM Borislav Petkov <bp@alien8.de> wrote:
>
> On Tue, Feb 15, 2022 at 02:22:33PM -0800, Luck, Tony wrote:
> > This early in do_machine check we don't know whether this was from
> > a over enthusistic REP;MOVS fetch, or a "normal" machine check.
> > I don't think there is an easy way to tell the difference.
>
> That's what I am wondering: whether we can compare the buffers REP;
> MOVS was accessing and determine whether the access was out of bounds.
> Something ala _ASM_EXTABLE_ as it is done in arch/x86/lib/copy_mc_64.S,
> for example, which will land us in fixup_exception().
>
> Now there we'd need to know the range the thing was copying which should
> be in pt_regs and the address the MCE reported. If latter is not in the
> former range, we say ignore.

This is a great idea.

My slight reservation is that this suggests all use cases of "REP; MOVS*" must
take the _ASM_EXTABLE_ form, which is not possible; considering
"REP; MOVS*" can be exercised from any user space program.

>
> There's even some blurb about "recovering from fast-string exceptions"
> over copy_mc_enhanced_fast_string...
>
> Hmmm?
If there is a way to get all users of "REP; MOVS*" to use
copy_mc_enhanced_fast_string, this could work. I am not sure this is possible.

>
> > The first check:
> >
> >       if ((mcgstatus & MCG_STATUS_LMCES)
> >
> > is for "is this a local machine check"? So no broadcast sync
> > needed. But that needs a comment.
>
> Yap.
Updated in the latest patch sent.
>
> --
> Regards/Gruss,
>     Boris.
>
> https://people.kernel.org/tglx/notes-about-netiquette