Hello,
This patch against 2.6.20 makes the NMI watchdog use PERFSEL1/PERFCTR1
instead of PERFSEL0/PERFCTR0 on processors supporting Intel architectural
perfmon, such as Intel Core 2. Although all PMU events can work on
both counters, the Precise Event-Based Sampling (PEBS) requires that the
event be in PERFCTR0 to work correctly (see section 18.14.4.1 in the
IA32 SDM Vol 3b).
A similar patch for x86-64 is to follow.
Changelog:
- make the i386 NMI watchdog use PERFSEL1/PERFCTR1 instead of PERFSEL0/PERFCTR0
on processors supporting the Intel architectural perfmon (e.g. Core 2 Duo).
This allows PEBS to work when the NMI watchdog is active.
signed-off-by: stephane eranian <[email protected]>
diff -urNp --exclude=.git linux-2.6.20.orig/arch/i386/kernel/nmi.c linux-2.6.20.base/arch/i386/kernel/nmi.c
--- linux-2.6.20.orig/arch/i386/kernel/nmi.c 2007-02-04 10:44:54.000000000 -0800
+++ linux-2.6.20.base/arch/i386/kernel/nmi.c 2007-02-17 07:59:41.000000000 -0800
@@ -288,7 +288,7 @@ static int __init check_nmi_watchdog(voi
* 32nd bit should be 1, for 33.. to be 1.
* Find the appropriate nmi_hz
*/
- if (wd->perfctr_msr == MSR_ARCH_PERFMON_PERFCTR0 &&
+ if (wd->perfctr_msr == MSR_ARCH_PERFMON_PERFCTR1 &&
((u64)cpu_khz * 1000) > 0x7fffffffULL) {
u64 count = (u64)cpu_khz * 1000;
do_div(count, 0x7fffffffUL);
@@ -685,8 +685,8 @@ static int setup_intel_arch_watchdog(voi
(ebx & ARCH_PERFMON_UNHALTED_CORE_CYCLES_PRESENT))
goto fail;
- perfctr_msr = MSR_ARCH_PERFMON_PERFCTR0;
- evntsel_msr = MSR_ARCH_PERFMON_EVENTSEL0;
+ perfctr_msr = MSR_ARCH_PERFMON_PERFCTR1;
+ evntsel_msr = MSR_ARCH_PERFMON_EVENTSEL1;
if (!reserve_perfctr_nmi(perfctr_msr))
goto fail;
@@ -958,7 +958,7 @@ __kprobes int nmi_watchdog_tick(struct p
apic_write(APIC_LVTPC, APIC_DM_NMI);
}
else if (wd->perfctr_msr == MSR_P6_PERFCTR0 ||
- wd->perfctr_msr == MSR_ARCH_PERFMON_PERFCTR0) {
+ wd->perfctr_msr == MSR_ARCH_PERFMON_PERFCTR1) {
/* P6 based Pentium M need to re-unmask
* the apic vector but it doesn't hurt
* other P6 variant.
--
-Stephane
On Saturday 17 February 2007 17:06, Stephane Eranian wrote:
> Hello,
>
> This patch against 2.6.20
Please always submit against mainline:
patching file arch/i386/kernel/nmi.c
Hunk #1 FAILED at 288.
Hunk #2 succeeded at 740 (offset 55 lines).
Hunk #3 succeeded at 1017 with fuzz 1 (offset 59 lines).
1 out of 3 hunks FAILED -- rejects in file arch/i386/kernel/nmi.c
Patch patches/i386-make-nmi-use-perfctr1-for-architectural-perfmon does not apply (enforce with -f)
-Andi
Andi,
On Sun, Feb 18, 2007 at 05:56:05PM +0100, Andi Kleen wrote:
> On Saturday 17 February 2007 17:06, Stephane Eranian wrote:
> > Hello,
> >
> > This patch against 2.6.20
>
> Please always submit against mainline:
>
Isn't 2.6.20 the latest mainline?
> patching file arch/i386/kernel/nmi.c
> Hunk #1 FAILED at 288.
> Hunk #2 succeeded at 740 (offset 55 lines).
> Hunk #3 succeeded at 1017 with fuzz 1 (offset 59 lines).
> 1 out of 3 hunks FAILED -- rejects in file arch/i386/kernel/nmi.c
> Patch patches/i386-make-nmi-use-perfctr1-for-architectural-perfmon does not apply (enforce with -f)
>
Maybe I messed up my original tree. Do you want me to resubmit?
Sorry about that.
--
-Stephane
On Monday 19 February 2007 17:36, Stephane Eranian wrote:
> Andi,
>
> On Sun, Feb 18, 2007 at 05:56:05PM +0100, Andi Kleen wrote:
> > On Saturday 17 February 2007 17:06, Stephane Eranian wrote:
> > > Hello,
> > >
> > > This patch against 2.6.20
> >
> > Please always submit against mainline:
> >
> Isn't 2.6.20 the latest mainline?
ftp://ftp.kernel.org/pub/linux/kernel/v2.6/snapshots/patch-2.6.20-git* is
> Do you want me to resubmit?
Yes please.
-Andi
Hello,
This patch against 2.6.20-git14 makes the NMI watchdog use PERFSEL1/PERFCTR1
instead of PERFSEL0/PERFCTR0 on processors supporting Intel architectural
perfmon, such as Intel Core 2. Although all PMU events can work on
both counters, the Precise Event-Based Sampling (PEBS) requires that the
event be in PERFCTR0 to work correctly (see section 18.14.4.1 in the
IA32 SDM Vol 3b).
A similar patch for x86-64 is to follow.
Changelog:
- make the i386 NMI watchdog use PERFSEL1/PERFCTR1 instead of PERFSEL0/PERFCTR0
on processors supporting the Intel architectural perfmon (e.g. Core 2 Duo).
This allows PEBS to work when the NMI watchdog is active.
signed-off-by: stephane eranian <[email protected]>
diff -urNp --exclude=.git linux-2.6.20.orig/arch/i386/kernel/nmi.c linux-2.6.20.base/arch/i386/kernel/nmi.c
--- linux-2.6.20.orig/arch/i386/kernel/nmi.c 2007-02-19 13:37:26.000000000 -0800
+++ linux-2.6.20.base/arch/i386/kernel/nmi.c 2007-02-19 13:54:42.000000000 -0800
@@ -307,7 +307,7 @@ static int __init check_nmi_watchdog(voi
nmi_hz = 1;
if (wd->perfctr_msr == MSR_P6_PERFCTR0 ||
- wd->perfctr_msr == MSR_ARCH_PERFMON_PERFCTR0) {
+ wd->perfctr_msr == MSR_ARCH_PERFMON_PERFCTR1) {
nmi_hz = adjust_for_32bit_ctr(nmi_hz);
}
}
@@ -741,8 +741,8 @@ static int setup_intel_arch_watchdog(voi
(ebx & ARCH_PERFMON_UNHALTED_CORE_CYCLES_PRESENT))
goto fail;
- perfctr_msr = MSR_ARCH_PERFMON_PERFCTR0;
- evntsel_msr = MSR_ARCH_PERFMON_EVENTSEL0;
+ perfctr_msr = MSR_ARCH_PERFMON_PERFCTR1;
+ evntsel_msr = MSR_ARCH_PERFMON_EVENTSEL1;
if (!reserve_perfctr_nmi(perfctr_msr))
goto fail;
@@ -1022,7 +1022,7 @@ __kprobes int nmi_watchdog_tick(struct p
write_watchdog_counter(wd->perfctr_msr, NULL);
}
else if (wd->perfctr_msr == MSR_P6_PERFCTR0 ||
- wd->perfctr_msr == MSR_ARCH_PERFMON_PERFCTR0) {
+ wd->perfctr_msr == MSR_ARCH_PERFMON_PERFCTR1) {
/* P6 based Pentium M need to re-unmask
* the apic vector but it doesn't hurt
* other P6 variant.