[patch] fix NMI watchdog, 2.5.34

[patch] fix NMI watchdog, 2.5.34

[Ingo Molnar]

the attached patch fixes the NMI watchdog to trigger on all CPUs - the
cpu_up() code broke it long time ago. With this patch NMI interrupts get
generated on all CPUs, not just the boot CPU.

	Ingo

--- linux/arch/i386/kernel/io_apic.c.orig	Mon Sep  9 21:34:53 2002
+++ linux/arch/i386/kernel/io_apic.c	Mon Sep  9 21:37:27 2002
@@ -1490,7 +1490,7 @@
 	end_lapic_irq
 };
 
-static void enable_NMI_through_LVT0 (void * dummy)
+void enable_NMI_through_LVT0 (void * dummy)
 {
 	unsigned int v, ver;
 
--- linux/arch/i386/kernel/smpboot.c.orig	Mon Sep  9 21:35:48 2002
+++ linux/arch/i386/kernel/smpboot.c	Mon Sep  9 21:43:00 2002
@@ -452,6 +452,11 @@
 	while (!test_bit(smp_processor_id(), &smp_commenced_mask))
 		rep_nop();
 	setup_secondary_APIC_clock();
+	if (nmi_watchdog == NMI_IO_APIC) {
+		disable_8259A_irq(0);
+		enable_NMI_through_LVT0(NULL);
+		enable_8259A_irq(0);
+	}
 	enable_APIC_timer();
 	/*
 	 * low-memory mappings have been cleared, flush them from
--- linux/include/asm-i386/apic.h.orig	Mon Sep  9 21:37:43 2002
+++ linux/include/asm-i386/apic.h	Mon Sep  9 21:37:51 2002
@@ -89,6 +89,7 @@
 
 extern unsigned int apic_timer_irqs [NR_CPUS];
 extern int check_nmi_watchdog (void);
+extern void enable_NMI_through_LVT0 (void * dummy);
 
 extern unsigned int nmi_watchdog;
 #define NMI_NONE	0

Re: [patch] fix NMI watchdog, 2.5.34

[Rusty Russell]

In message <Pine.LNX.4.44.0209092144140.10544-100000@localhost.localdomain> you
 write:
> 
> the attached patch fixes the NMI watchdog to trigger on all CPUs - the
> cpu_up() code broke it long time ago. With this patch NMI interrupts get
> generated on all CPUs, not just the boot CPU.

Well spotted.  You might want to test the following patch which
catches calls to smp_call_function() before the cpus are actually
online.  I ran a variant on my (crappy, old, SMP) box before I sent
the patch to Linus, and all I saw was the (harmless) tlb_flush.

diff -urNp --exclude TAGS -X /home/rusty/current-dontdiff --minimal linux-2.5.34/arch/i386/kernel/smp.c working-2.5.34-smp_call_cpus/arch/i386/kernel/smp.c
--- linux-2.5.34/arch/i386/kernel/smp.c	Wed Aug 28 09:29:40 2002
+++ working-2.5.34-smp_call_cpus/arch/i386/kernel/smp.c	Tue Sep 10 14:50:15 2002
@@ -561,9 +561,15 @@ int smp_call_function (void (*func) (voi
  * hardware interrupt handler or from a bottom half handler.
  */
 {
+	extern int smp_done;
 	struct call_data_struct data;
 	int cpus = num_online_cpus()-1;
 
+	if (!smp_done) {
+		printk(KERN_ERR "smp_call_function %p called before SMP!\n",
+		       func);
+		show_stack(NULL);
+	}
 	if (!cpus)
 		return 0;
 
diff -urNp --exclude TAGS -X /home/rusty/current-dontdiff --minimal linux-2.5.34/arch/i386/kernel/smpboot.c working-2.5.34-smp_call_cpus/arch/i386/kernel/smpboot.c
--- linux-2.5.34/arch/i386/kernel/smpboot.c	Sun Sep  1 12:22:57 2002
+++ working-2.5.34-smp_call_cpus/arch/i386/kernel/smpboot.c	Tue Sep 10 14:35:07 2002
@@ -1218,7 +1218,10 @@ int __devinit __cpu_up(unsigned int cpu)
 	return 0;
 }
 
+unsigned int smp_done = 0;
+
 void __init smp_cpus_done(unsigned int max_cpus)
 {
 	zap_low_mappings();
+	smp_done = 1;
 }
--
  Anyone who quotes me in their sig is an idiot. -- Rusty Russell.

Re: [patch] fix NMI watchdog, 2.5.34

[Zwane Mwaikambo]

On Tue, 10 Sep 2002, Rusty Russell wrote:

> Well spotted.  You might want to test the following patch which
> catches calls to smp_call_function() before the cpus are actually
> online.  I ran a variant on my (crappy, old, SMP) box before I sent
> the patch to Linus, and all I saw was the (harmless) tlb_flush.

hmm...

> diff -urNp --exclude TAGS -X /home/rusty/current-dontdiff --minimal linux-2.5.34/arch/i386/kernel/smpboot.c working-2.5.34-smp_call_cpus/arch/i386/kernel/smpboot.c
> --- linux-2.5.34/arch/i386/kernel/smpboot.c	Sun Sep  1 12:22:57 2002
> +++ working-2.5.34-smp_call_cpus/arch/i386/kernel/smpboot.c	Tue Sep 10 14:35:07 2002
> @@ -1218,7 +1218,10 @@ int __devinit __cpu_up(unsigned int cpu)
>  	return 0;
>  }
>  
> +unsigned int smp_done = 0;
> +
>  void __init smp_cpus_done(unsigned int max_cpus)
>  {
>  	zap_low_mappings();
> +	smp_done = 1;

I've got an SMP box which dies reliably at zap_low_mappings, i wonder if 
this could be the same problem. My BSP sits spinning on the completion 
check.

	Zwane

-- 
function.linuxpower.ca

Re: [patch] fix NMI watchdog, 2.5.34

[Rusty Russell]

On Tue, 10 Sep 2002 20:11:49 +0200 (SAST)
Zwane Mwaikambo <zwane@mwaikambo.name> wrote:

> On Tue, 10 Sep 2002, Rusty Russell wrote:
> 
> > Well spotted.  You might want to test the following patch which
> > catches calls to smp_call_function() before the cpus are actually
> > online.  I ran a variant on my (crappy, old, SMP) box before I sent
> > the patch to Linus, and all I saw was the (harmless) tlb_flush.
> 
> hmm...
> 
> > diff -urNp --exclude TAGS -X /home/rusty/current-dontdiff --minimal linux-2.5.34/arch/i386/kernel/smpboot.c working-2.5.34-smp_call_cpus/arch/i386/kernel/smpboot.c
> > --- linux-2.5.34/arch/i386/kernel/smpboot.c	Sun Sep  1 12:22:57 2002
> > +++ working-2.5.34-smp_call_cpus/arch/i386/kernel/smpboot.c	Tue Sep 10 14:35:07 2002
> > @@ -1218,7 +1218,10 @@ int __devinit __cpu_up(unsigned int cpu)
> >  	return 0;
> >  }
> >  
> > +unsigned int smp_done = 0;
> > +
> >  void __init smp_cpus_done(unsigned int max_cpus)
> >  {
> >  	zap_low_mappings();
> > +	smp_done = 1;
> 
> I've got an SMP box which dies reliably at zap_low_mappings, i wonder if 
> this could be the same problem. My BSP sits spinning on the completion 
> check.

Hmmm, I can't see how: you mean it hangs in flush_tlb_all() (waiting for the
ack in smp_call_function()?).  If so, that seems really wierd.  You could add
a printk("here: %u\n", smp_processor_id()) in flush_tlb_all_ipi() to see which
CPU isn't getting it...

Strange,
Rusty.
-- 
   there are those who do and those who hang on and you don't see too
   many doers quoting their contemporaries.  -- Larry McVoy