From: Konrad Rzeszutek <konradr@redhat.com>
Subject: Re: [RHEL5.1 PATCH] BZ# 229563: LTC32322-HTS Info test causes 4 Node 3950 to lock up
Date: Sat, 24 Mar 2007 09:49:29 -0400
Bugzilla: 229563
Message-Id: <20070324134929.GA23210@localhost.localdomain>
Changelog: [x86] tell sysrq-m to poke the nmi watchdog


> Have you timed sysrq-m before/after the patch?

I did. The original proposed patch is horribly slow. It adds about 156% performance degradation.


> >+			/* this loop can take a while with 256 GB and 4k 
> >pages
> >+			   so update the NMI watchdog */
> >+			touch_nmi_watchdog();
> > 			page = pfn_to_page(pgdat->node_start_pfn + i);
> > 			total++;
> > 			if (PageReserved(page))
> 
> I'm concerned that touch_nmi_watchdog() for every single page might be a 
> bit expensive.  If the difference is negligible, it's fine as it is.  If 
> it's a substantial difference, we might want to wrap it in something 
> like if(unlikely(i & ~(MAX_ORDER_NR_PAGES - 1) == 0)).  If we can't scan 

I am not sure how that would work, as it would still touch every page after
the first 1024. I think you meant something along these lines:

diff --git a/arch/x86_64/mm/init.c b/arch/x86_64/mm/init.c
index ec31534..fd7f8a6 100644
--- a/arch/x86_64/mm/init.c
+++ b/arch/x86_64/mm/init.c
@@ -26,6 +26,7 @@ #include <linux/poison.h>
 #include <linux/dma-mapping.h>
 #include <linux/module.h>
 #include <linux/memory_hotplug.h>
+#include <linux/nmi.h>
 
 #include <asm/processor.h>
 #include <asm/system.h>
@@ -72,6 +73,11 @@ void show_mem(void)
 
 	for_each_online_pgdat(pgdat) {
                for (i = 0; i < pgdat->node_spanned_pages; ++i) {
+			/* this loop can take a while with 256 GB and 4k pages
+			   so update the NMI watchdog */
+			if (unlikely(i % MAX_ORDER_NR_PAGES == 0)) {
+				touch_nmi_watchdog();
+			}
 			page = pfn_to_page(pgdat->node_start_pfn + i);
 			total++;
 			if (PageReserved(page))