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;
Date:   Fri, 8 May 2020 10:48:09 +0200
From:   Peter Zijlstra <peterz@infradead.org>
To:     Andy Lutomirski <luto@kernel.org>
Cc:     Thomas Gleixner <tglx@linutronix.de>,
        LKML <linux-kernel@vger.kernel.org>, X86 ML <x86@kernel.org>,
        "Paul E. McKenney" <paulmck@kernel.org>,
        Alexandre Chartre <alexandre.chartre@oracle.com>,
        Frederic Weisbecker <frederic@kernel.org>,
        Paolo Bonzini <pbonzini@redhat.com>,
        Sean Christopherson <sean.j.christopherson@intel.com>,
        Masami Hiramatsu <mhiramat@kernel.org>,
        Petr Mladek <pmladek@suse.com>,
        Steven Rostedt <rostedt@goodmis.org>,
        Joel Fernandes <joel@joelfernandes.org>,
        Boris Ostrovsky <boris.ostrovsky@oracle.com>,
        Juergen Gross <jgross@suse.com>,
        Brian Gerst <brgerst@gmail.com>,
        Mathieu Desnoyers <mathieu.desnoyers@efficios.com>,
        Josh Poimboeuf <jpoimboe@redhat.com>,
        Will Deacon <will@kernel.org>
Subject: Re: [patch V4 part 1 29/36] x86/mce: Send #MC singal from task work
Message-ID: <20200508084809.GS5298@hirez.programming.kicks-ass.net>
References: <20200505131602.633487962@linutronix.de>
 <20200505134100.957390899@linutronix.de>
 <CALCETrX=Obqn2ms5EYs7HPxTE_UgnVkmt-HoAoGzB4BajuMwLA@mail.gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <CALCETrX=Obqn2ms5EYs7HPxTE_UgnVkmt-HoAoGzB4BajuMwLA@mail.gmail.com>
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Thu, May 07, 2020 at 11:02:09AM -0700, Andy Lutomirski wrote:
> On Tue, May 5, 2020 at 7:13 AM Thomas Gleixner <tglx@linutronix.de> wrote:
> >
> > From: Peter Zijlstra <peterz@infradead.org>
> >
> > Convert #MC over to using task_work_add(); it will run the same code
> > slightly later, on the return to user path of the same exception.
> 
> I think this patch is correct, but I think it's only one small and not
> that obviously wrong step away from being broken:
> 
> >         if ((m.cs & 3) == 3) {
> >                 /* If this triggers there is no way to recover. Die hard. */
> >                 BUG_ON(!on_thread_stack() || !user_mode(regs));
> > -               local_irq_enable();
> > -               preempt_enable();
> >
> > -               if (kill_it || do_memory_failure(&m))
> > -                       force_sig(SIGBUS);
> > -               preempt_disable();
> > -               local_irq_disable();
> > +               current->mce_addr = m.addr;
> > +               current->mce_status = m.mcgstatus;
> > +               current->mce_kill_me.func = kill_me_maybe;
> > +               if (kill_it)
> > +                       current->mce_kill_me.func = kill_me_now;
> > +               task_work_add(current, &current->mce_kill_me, true);
> 
> This is fine if the source was CPL3, but it's not going to work if CPL
> was 0.  We don't *currently* do this from CPL0, but people keep
> wanting to.  So perhaps there should be a comment like:
> 
> /*
>  * The #MC originated at CPL3, so we know that we will go execute the
> task_work before returning to the offending user code.
>  */
> 
> IOW, if we want to recover from CPL0 #MC, we will need a different mechanism.

See part4-18's IDTRENTRY_NOIST. That will get us a clear CPL3/CPL0
separation.

> I also confess a certain amount of sadness that my beautiful
> haha-not-really-atomic-here mechanism isn't being used anymore. :(

I think we have a subtely different interpretation of 'beautiful' here.