2003-07-14 15:12:33

by John Levon

Subject: [PATCH] OProfile: dynamically allocate MSR struct

Andi pointed out to me that cpu_msrs was a source of bloat. Dynamically
allocate it instead. Tested on my SMP box.

Note that we could eventually use per cpu stuff here, but I'd like to
wait on that for now.

regards
john


diff -Naur -X dontdiff linux-cvs/arch/i386/oprofile/nmi_int.c linux-fixes/arch/i386/oprofile/nmi_int.c
--- linux-cvs/arch/i386/oprofile/nmi_int.c 2003-06-18 15:06:05.000000000 +0100
+++ linux-fixes/arch/i386/oprofile/nmi_int.c 2003-07-14 15:34:29.000000000 +0100
@@ -12,6 +12,7 @@
#include <linux/smp.h>
#include <linux/oprofile.h>
#include <linux/sysdev.h>
+#include <linux/slab.h>
#include <asm/nmi.h>
#include <asm/msr.h>
#include <asm/apic.h>
@@ -20,7 +21,7 @@
#include "op_x86_model.h"

static struct op_x86_model_spec const * model;
-static struct op_msrs cpu_msrs[NR_CPUS];
+static struct op_msrs * cpu_msrs;
static unsigned long saved_lvtpc[NR_CPUS];

static int nmi_start(void);
@@ -125,6 +126,10 @@

static int nmi_setup(void)
{
+ cpu_msrs = kmalloc(sizeof(struct op_msrs) * NR_CPUS, GFP_KERNEL);
+ if (!cpu_msrs)
+ return -ENOMEM;
+
/* We walk a thin line between law and rape here.
* We need to be careful to install our NMI handler
* without actually triggering any NMIs as this will
@@ -185,6 +190,7 @@
on_each_cpu(nmi_cpu_shutdown, NULL, 0, 1);
unset_nmi_callback();
enable_lapic_nmi_watchdog();
+ kfree(cpu_msrs);
}

2003-07-14 15:39:58

by John Levon

Subject: Re: [PATCH] OProfile: dynamically allocate MSR struct

On Mon, Jul 14, 2003 at 04:17:28PM +0100, John Levon wrote:

> Andi pointed out to me that cpu_msrs was a source of bloat. Dynamically
> allocate it instead. Tested on my SMP box.

Gah, bogus patch, please ignore.

thanks
john