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;
Date:   Mon, 9 Jan 2023 08:59:42 -0800
From:   "Paul E. McKenney" <paulmck@kernel.org>
To:     Peter Zijlstra <peterz@infradead.org>
Cc:     linux-kernel@vger.kernel.org, kernel-team@meta.com,
        Thomas Gleixner <tglx@linutronix.de>,
        Ingo Molnar <mingo@redhat.com>, Borislav Petkov <bp@alien8.de>,
        Dave Hansen <dave.hansen@linux.intel.com>,
        "H. Peter Anvin" <hpa@zytor.com>, x86@kernel.org
Subject: Re: [PATCH x86/nmi 1/2] x86/nmi: Accumulate NMI-progress evidence in
 exc_nmi()
Message-ID: <20230109165942.GQ4028633@paulmck-ThinkPad-P17-Gen-1>
Reply-To: paulmck@kernel.org
References: <20230105011537.1774941-1-paulmck@kernel.org>
 <Y7w+r3Nbubin8sNN@hirez.programming.kicks-ass.net>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <Y7w+r3Nbubin8sNN@hirez.programming.kicks-ass.net>
Precedence: bulk

On Mon, Jan 09, 2023 at 05:19:59PM +0100, Peter Zijlstra wrote:
> On Wed, Jan 04, 2023 at 05:15:36PM -0800, Paul E. McKenney wrote:
> 
> > diff --git a/arch/x86/kernel/nmi.c b/arch/x86/kernel/nmi.c
> > index cec0bfa3bc04f..4f1651dc65b3a 100644
> > --- a/arch/x86/kernel/nmi.c
> > +++ b/arch/x86/kernel/nmi.c
> > @@ -69,6 +69,15 @@ struct nmi_stats {
> >  	unsigned int unknown;
> >  	unsigned int external;
> >  	unsigned int swallow;
> > +	unsigned long recv_jiffies;
> > +	unsigned long idt_seq;
> > +	unsigned long idt_nmi_seq;
> > +	unsigned long idt_ignored;
> > +	atomic_long_t idt_calls;
> > +	unsigned long idt_seq_snap;
> > +	unsigned long idt_nmi_seq_snap;
> > +	unsigned long idt_ignored_snap;
> > +	long idt_calls_snap;
> >  };
> 
> Urgh, this is more than a whole cacheline of extra data :/ Can't we make
> this #ifdef CONFIG_NMI_CHECK_CPU ?

We can.  However, the new data is at the end of the structure.  Plus doing
that will also require converting the IS_ENABLED() checks to #ifdef or
to lots of tiny functions, neither of which will be at all pretty.

Another approach would be to have a #else that created a union of all
of these fields, thus reducing this unused space so that of a single
unsigned long, while still permitting IS_ENABLED() checks.  Thoughts?

> > @@ -509,8 +526,15 @@ DEFINE_IDTENTRY_RAW(exc_nmi)
> >  
> >  	inc_irq_stat(__nmi_count);
> >  
> > -	if (!ignore_nmis)
> > +	if (IS_ENABLED(CONFIG_NMI_CHECK_CPU) && ignore_nmis) {
> > +		WRITE_ONCE(nsp->idt_ignored, nsp->idt_ignored + 1);
> > +	} else if (!ignore_nmis) {
> > +		WRITE_ONCE(nsp->idt_nmi_seq, nsp->idt_nmi_seq + 1);
> > +		WARN_ON_ONCE(!(nsp->idt_nmi_seq & 0x1));
> >  		default_do_nmi(regs);
> > +		WRITE_ONCE(nsp->idt_nmi_seq, nsp->idt_nmi_seq + 1);
> > +		WARN_ON_ONCE(nsp->idt_nmi_seq & 0x1);
> > +	}
> >  
> >  	irqentry_nmi_exit(regs, irq_state);
> 
> That is not a NO-OP when !CONFIG_NMI_CHECK_CPU :/

Good catch, will fix!

							Thanx, Paul