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:   Andy Lutomirski <luto@amacapital.net>
Mime-Version: 1.0 (1.0)
Subject: Re: [RFC][PATCH 0/4] x86/entry: disallow #DB more
Date:   Fri, 22 May 2020 15:43:35 -0700
Message-Id: <692612FC-4599-4774-857C-7E20CD99A832@amacapital.net>
References: <20200522222055.GE25128@linux.intel.com>
Cc:     Andy Lutomirski <luto@kernel.org>,
        Peter Zijlstra <peterz@infradead.org>,
        Thomas Gleixner <tglx@linutronix.de>,
        LKML <linux-kernel@vger.kernel.org>, X86 ML <x86@kernel.org>
In-Reply-To: <20200522222055.GE25128@linux.intel.com>
To:     Sean Christopherson <sean.j.christopherson@intel.com>
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk


> On May 22, 2020, at 3:20 PM, Sean Christopherson <sean.j.christopherson@in=
tel.com> wrote:
>=20
> =EF=BB=BFOn Fri, May 22, 2020 at 03:13:57PM -0700, Andy Lutomirski wrote:
>>> On Fri, May 22, 2020 at 1:49 PM Peter Zijlstra <peterz@infradead.org> wr=
ote:
>>>=20
>>> Hai, this kills #DB during NMI/#MC and with that allows removing all the=
 nasty
>>> IST rewrite crud.
>>>=20
>>=20
>> This is great, except that the unconditional DR7 write is going to
>> seriously hurt perf performance.  Fortunately, no one cares about
>> perf, right? :)  Even just reading first won't help enough because DR7
>> reads are likely to be VM exits.  Can we have a percpu dr7 shadow
>> (with careful ordering) or even just a percpu count of dr7 users so we
>> can skip this if there are no breakpoints?
>=20
> Hmm, I believe hw_breakpoint_active() is what you're looking for, KVM uses=

> it to avoid unnecessary restoration of host DR7 after VM-Exit.
>=20
> Amusingly, checking that in the NMI handler could give a false positive if=

> an NMI occurs in guest as DR7 is cleared on exit and KVM invokes the NMI
> handler prior to restoring host DR7.  I doubt that's common enough to care=

> about though.

False positives are unavoidable: there=E2=80=99s no way we can set a percpu v=
ariable and set DR7 without risking an NMI in between.=