Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
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: [patch 13/16] x86/entry: Move irqflags and context tracking to C for simple idtentries
Date:   Wed, 26 Feb 2020 11:15:31 -0800
Message-Id: <A4E714B9-ECBA-4984-986C-02B4EAF5018C@amacapital.net>
References: <20200226162811.GA18400@hirez.programming.kicks-ass.net>
Cc:     Thomas Gleixner <tglx@linutronix.de>,
        LKML <linux-kernel@vger.kernel.org>, X86 ML <x86@kernel.org>,
        Steven Rostedt <rostedt@goodmis.org>,
        Brian Gerst <brgerst@gmail.com>,
        Juergen Gross <JGross@suse.com>,
        Paolo Bonzini <pbonzini@redhat.com>,
        Arnd Bergmann <arnd@arndb.de>
In-Reply-To: <20200226162811.GA18400@hirez.programming.kicks-ass.net>
To:     Peter Zijlstra <peterz@infradead.org>
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk


> On Feb 26, 2020, at 8:28 AM, Peter Zijlstra <peterz@infradead.org> wrote:
>=20
> =EF=BB=BFOn Wed, Feb 26, 2020 at 07:11:39AM -0800, Andy Lutomirski wrote:
>=20
>> In some sense, this is a weakness of the magic macro approach.  Some
>> of the entries just want to have code that runs before all the entry
>> fixups.  This is an example of it.  So are the cr2 reads.  It can all
>> be made to work, but it's a bit gross.
>=20
> Right. In my current pile (new patche since last posting) I also have
> one that makes #DB save-clear/restore DR7.
>=20
> I got it early enough that only a watchpoint on the task stack can still
> screw us over, since I also included your patch that excludes
> cpu_entry_area.

Hmm. It would be nice to prevent watchpoints on the task stack, but that wou=
ld need some trickery.  It could be done.

>=20
> Pushing it earlier still would require calling into C from the entry
> stack, which I know is on your todo list, but we're not quite there yet.

Indeed.

This is my main objection to the DEFINE_IDTENTRY stuff. It=E2=80=99s *great*=
 for the easy cases, but it=E2=80=99s not so great for the nasty cases. Mayb=
e we should open code PF, MC, DB, etc.  (And kill the kvm special case for P=
F.  I have a working patch for that and I can send it.)

Anyway, this isn=E2=80=99t actually what I was concerned about. I meant DR6,=
 not DR7.  Specifically, if we get traced too early in do_debug / exc_debug,=
 we can recursively debug and clobber DR6.  The result will be incorrect beh=
avior in the outer do_debug. We can fix this the same way we handle CR2.  I j=
ust haven=E2=80=99t done it on the existing entry code because it=E2=80=99s t=
oo messy.=