Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1339354 > unrolled thread

[patch V2 28/28] x86/perf/intel/rapl: Make it modular

Back to article view | Back to linux.kernel

This discussion starts older than the indexed window; earlier articles aren't shown. The article labeled Started by below is the oldest one visible, not the original post.

Contents

  [patch V2 28/28] x86/perf/intel/rapl: Make it modular Thomas Gleixner <tglx@linutronix.de> - 2016-02-22 12:10 +0100
    Re: [patch V2 28/28] x86/perf/intel/rapl: Make it modular Andi Kleen <andi@firstfloor.org> - 2016-02-22 19:50 +0100
      Re: [patch V2 28/28] x86/perf/intel/rapl: Make it modular Thomas Gleixner <tglx@linutronix.de> - 2016-02-22 20:20 +0100

#1339354 — [patch V2 28/28] x86/perf/intel/rapl: Make it modular

Add the necessary exit functions so it can be built as a module.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
---
 arch/x86/Kconfig.perf                       |   10 ++++++++++
 arch/x86/kernel/cpu/Makefile                |    4 +++-
 arch/x86/kernel/cpu/perf_event_intel_rapl.c |   22 +++++++++++++++++++---
 3 files changed, 32 insertions(+), 4 deletions(-)

--- a/arch/x86/Kconfig.perf
+++ b/arch/x86/Kconfig.perf
@@ -10,4 +10,14 @@ config PERF_EVENTS_INTEL_UNCORE
 
 	  If unsure say y.
 
+config PERF_EVENTS_INTEL_RAPL
+	tristate "Intel rapl performance events"
+	depends on PERF_EVENTS && CPU_SUP_INTEL && PCI
+	default y
+	---help---
+	  Include support for Intel rapl performance events for power
+	  monitoring on modern processors.
+
+	  If unsure say y.
+
 endmenu
--- a/arch/x86/kernel/cpu/Makefile
+++ b/arch/x86/kernel/cpu/Makefile
@@ -39,10 +39,12 @@ obj-$(CONFIG_CPU_SUP_AMD)		+= perf_event
 endif
 obj-$(CONFIG_CPU_SUP_INTEL)		+= perf_event_p6.o perf_event_knc.o perf_event_p4.o
 obj-$(CONFIG_CPU_SUP_INTEL)		+= perf_event_intel_lbr.o perf_event_intel_ds.o perf_event_intel.o
-obj-$(CONFIG_CPU_SUP_INTEL)		+= perf_event_intel_rapl.o perf_event_intel_cqm.o
+obj-$(CONFIG_CPU_SUP_INTEL)		+= perf_event_intel_cqm.o
 obj-$(CONFIG_CPU_SUP_INTEL)		+= perf_event_intel_pt.o perf_event_intel_bts.o
 obj-$(CONFIG_CPU_SUP_INTEL)		+= perf_event_intel_cstate.o
 
+obj-$(CONFIG_PERF_EVENTS_INTEL_RAPL)	+= perf_event_intel_rapl.o
+
 obj-$(CONFIG_PERF_EVENTS_INTEL_UNCORE)	+= perf_event_intel_uncores.o
 perf_event_intel_uncores-objs		:= perf_event_intel_uncore.o \
 					   perf_event_intel_uncore_snb.o \
--- a/arch/x86/kernel/cpu/perf_event_intel_rapl.c
+++ b/arch/x86/kernel/cpu/perf_event_intel_rapl.c
@@ -53,6 +53,8 @@
 #include <asm/cpu_device_id.h>
 #include "perf_event.h"
 
+MODULE_LICENSE("GPL");
+
 /*
  * RAPL energy status counters
  */
@@ -592,6 +594,10 @@ static int rapl_cpu_notifier(struct noti
 	return NOTIFY_OK;
 }
 
+static struct notifier_block rapl_cpu_nb = {
+	.notifier_call	= rapl_cpu_notifier,
+};
+
 static __init void rapl_hsw_server_quirk(void)
 {
 	/*
@@ -666,7 +672,7 @@ static int __init rapl_prepare_cpus(void
 	return 0;
 }
 
-static void __init cleanup_rapl_pmus(void)
+static void cleanup_rapl_pmus(void)
 {
 	int i;
 
@@ -759,7 +765,7 @@ static int __init rapl_pmu_init(void)
 	if (ret)
 		goto out;
 
-	__perf_cpu_notifier(rapl_cpu_notifier);
+	__register_cpu_notifier(&rapl_cpu_nb);
 	cpu_notifier_register_done();
 	rapl_advertise();
 	return 0;
@@ -770,4 +776,14 @@ static int __init rapl_pmu_init(void)
 	cpu_notifier_register_done();
 	return ret;
 }
-device_initcall(rapl_pmu_init);
+module_init(rapl_pmu_init);
+
+static void __exit intel_rapl_exit(void)
+{
+	cpu_notifier_register_done();
+	__unregister_cpu_notifier(&rapl_cpu_nb);
+	perf_pmu_unregister(&rapl_pmus->pmu);
+	cleanup_rapl_pmus();
+	cpu_notifier_register_done();
+}
+module_exit(intel_rapl_exit);

[toc] | [next] | [standalone]

#1339812

Thomas Gleixner <tglx@linutronix.de> writes:

> Add the necessary exit functions so it can be built as a module.

If you make it a module you also need to add MODULE_DEVICE_TABLE
for the PCI IDs and also add x86_cpu_id tables/annotations
for the model numbers.

Otherwise the module would always need to be loaded manually
which would break existing setups.

Of course it's a bit of a waste because in many cases the uncore
driver will not be needed. One alternative would be to only
add some module aliases and then add code in the perf core
to probe the alias when that PMU is accessed. That would
likely work for most of perf, except for perf list (which does
not know what pmu to probe)

Other than that it's very useful, it was long overdue to make
these drivers modular.

-Andi

-- 
ak@linux.intel.com -- Speaking for myself only

[toc] | [prev] | [next] | [standalone]

#1339851

Andi,

On Mon, 22 Feb 2016, Andi Kleen wrote:

> Thomas Gleixner <tglx@linutronix.de> writes:
> 
> > Add the necessary exit functions so it can be built as a module.
> 
> If you make it a module you also need to add MODULE_DEVICE_TABLE
> for the PCI IDs and also add x86_cpu_id tables/annotations
> for the model numbers.
> 
> Otherwise the module would always need to be loaded manually
> which would break existing setups.

I know. That's why I made the default Yes. We can drop that patch or add some
explanation to the help text that 'M' is for manual module loading until
someone adds the necessary magic for auto probing.

Thanks,

	tglx

[toc] | [prev] | [standalone]

Back to top | Article view | linux.kernel

csiph-web