The scheduler reads the IPCC scores when balancing load. These reads can
be quite frequent. Hardware can also update the HFI table frequently.
Concurrent access may cause a lot of lock contention. It gets worse as the
number of CPUs increases.
Instead, create separate per-CPU IPCC scores that the scheduler can read
without the HFI table lock.
Cc: Ben Segall <bsegall@google.com>
Cc: Daniel Bristot de Oliveira <bristot@redhat.com>
Cc: Dietmar Eggemann <dietmar.eggemann@arm.com>
Cc: Ionela Voinescu <ionela.voinescu@arm.com>
Cc: Joel Fernandes (Google) <joel@joelfernandes.org>
Cc: Len Brown <len.brown@intel.com>
Cc: Lukasz Luba <lukasz.luba@arm.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Cc: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Tim C. Chen <tim.c.chen@intel.com>
Cc: Valentin Schneider <vschneid@redhat.com>
Cc: x86@kernel.org
Cc: linux-pm@vger.kernel.org
Cc: linux-kernel@vger.kernel.org
Suggested-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ricardo Neri <ricardo.neri-calderon@linux.intel.com>
---
Changes since v2:
* Only create these per-CPU variables when Intel Thread Director is
supported.
Changes since v1:
* Added this patch.
---
drivers/thermal/intel/intel_hfi.c | 46 +++++++++++++++++++++++++++++++
1 file changed, 46 insertions(+)
diff --git a/drivers/thermal/intel/intel_hfi.c b/drivers/thermal/intel/intel_hfi.c
index 2527ae3836c7..b06021828892 100644
--- a/drivers/thermal/intel/intel_hfi.c
+++ b/drivers/thermal/intel/intel_hfi.c
@@ -29,6 +29,7 @@
#include <linux/kernel.h>
#include <linux/math.h>
#include <linux/mutex.h>
+#include <linux/percpu.h>
#include <linux/percpu-defs.h>
#include <linux/printk.h>
#include <linux/processor.h>
@@ -170,6 +171,43 @@ static struct workqueue_struct *hfi_updates_wq;
#define HFI_UPDATE_INTERVAL HZ
#define HFI_MAX_THERM_NOTIFY_COUNT 16
+#ifdef CONFIG_IPC_CLASSES
+static int __percpu *hfi_ipcc_scores;
+
+static int alloc_hfi_ipcc_scores(void)
+{
+ if (!cpu_feature_enabled(X86_FEATURE_ITD))
+ return 0;
+
+ hfi_ipcc_scores = __alloc_percpu(sizeof(*hfi_ipcc_scores) *
+ hfi_features.nr_classes,
+ sizeof(*hfi_ipcc_scores));
+
+ return !hfi_ipcc_scores;
+}
+
+static void set_hfi_ipcc_score(void *caps, int cpu)
+{
+ int i, *hfi_class;
+
+ if (!cpu_feature_enabled(X86_FEATURE_ITD))
+ return;
+
+ hfi_class = per_cpu_ptr(hfi_ipcc_scores, cpu);
+
+ for (i = 0; i < hfi_features.nr_classes; i++) {
+ struct hfi_cpu_data *class_caps;
+
+ class_caps = caps + i * hfi_features.class_stride;
+ WRITE_ONCE(hfi_class[i], class_caps->perf_cap);
+ }
+}
+
+#else
+static int alloc_hfi_ipcc_scores(void) { return 0; }
+static void set_hfi_ipcc_score(void *caps, int cpu) { }
+#endif /* CONFIG_IPC_CLASSES */
+
static void get_hfi_caps(struct hfi_instance *hfi_instance,
struct thermal_genl_cpu_caps *cpu_caps)
{
@@ -192,6 +230,8 @@ static void get_hfi_caps(struct hfi_instance *hfi_instance,
cpu_caps[i].efficiency = caps->ee_cap << 2;
++i;
+
+ set_hfi_ipcc_score(caps, cpu);
}
raw_spin_unlock_irq(&hfi_instance->table_lock);
}
@@ -580,8 +620,14 @@ void __init intel_hfi_init(void)
if (!hfi_updates_wq)
goto err_nomem;
+ if (alloc_hfi_ipcc_scores())
+ goto err_ipcc;
+
return;
+err_ipcc:
+ destroy_workqueue(hfi_updates_wq);
+
err_nomem:
for (j = 0; j < i; ++j) {
hfi_instance = &hfi_instances[j];
--
2.25.1On Tue, Feb 7, 2023 at 6:02 AM Ricardo Neri
<ricardo.neri-calderon@linux.intel.com> wrote:
>
> The scheduler reads the IPCC scores when balancing load. These reads can
> be quite frequent. Hardware can also update the HFI table frequently.
> Concurrent access may cause a lot of lock contention. It gets worse as the
> number of CPUs increases.
>
> Instead, create separate per-CPU IPCC scores that the scheduler can read
> without the HFI table lock.
>
> Cc: Ben Segall <bsegall@google.com>
> Cc: Daniel Bristot de Oliveira <bristot@redhat.com>
> Cc: Dietmar Eggemann <dietmar.eggemann@arm.com>
> Cc: Ionela Voinescu <ionela.voinescu@arm.com>
> Cc: Joel Fernandes (Google) <joel@joelfernandes.org>
> Cc: Len Brown <len.brown@intel.com>
> Cc: Lukasz Luba <lukasz.luba@arm.com>
> Cc: Mel Gorman <mgorman@suse.de>
> Cc: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
> Cc: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
> Cc: Steven Rostedt <rostedt@goodmis.org>
> Cc: Tim C. Chen <tim.c.chen@intel.com>
> Cc: Valentin Schneider <vschneid@redhat.com>
> Cc: x86@kernel.org
> Cc: linux-pm@vger.kernel.org
> Cc: linux-kernel@vger.kernel.org
> Suggested-by: Peter Zijlstra (Intel) <peterz@infradead.org>
> Signed-off-by: Ricardo Neri <ricardo.neri-calderon@linux.intel.com>
> ---
> Changes since v2:
> * Only create these per-CPU variables when Intel Thread Director is
> supported.
>
> Changes since v1:
> * Added this patch.
> ---
> drivers/thermal/intel/intel_hfi.c | 46 +++++++++++++++++++++++++++++++
> 1 file changed, 46 insertions(+)
>
> diff --git a/drivers/thermal/intel/intel_hfi.c b/drivers/thermal/intel/intel_hfi.c
> index 2527ae3836c7..b06021828892 100644
> --- a/drivers/thermal/intel/intel_hfi.c
> +++ b/drivers/thermal/intel/intel_hfi.c
> @@ -29,6 +29,7 @@
> #include <linux/kernel.h>
> #include <linux/math.h>
> #include <linux/mutex.h>
> +#include <linux/percpu.h>
> #include <linux/percpu-defs.h>
> #include <linux/printk.h>
> #include <linux/processor.h>
> @@ -170,6 +171,43 @@ static struct workqueue_struct *hfi_updates_wq;
> #define HFI_UPDATE_INTERVAL HZ
> #define HFI_MAX_THERM_NOTIFY_COUNT 16
>
> +#ifdef CONFIG_IPC_CLASSES
It would be good to provide a (concise) description of this variable.
> +static int __percpu *hfi_ipcc_scores;
> +
> +static int alloc_hfi_ipcc_scores(void)
> +{
> + if (!cpu_feature_enabled(X86_FEATURE_ITD))
> + return 0;
> +
> + hfi_ipcc_scores = __alloc_percpu(sizeof(*hfi_ipcc_scores) *
> + hfi_features.nr_classes,
> + sizeof(*hfi_ipcc_scores));
> +
> + return !hfi_ipcc_scores;
I would do
if (!hfi_ipcc_scores)
return -ENOMEM;
return 0;
Or make the function return bool.
> +}
> +
> +static void set_hfi_ipcc_score(void *caps, int cpu)
> +{
> + int i, *hfi_class;
> +
> + if (!cpu_feature_enabled(X86_FEATURE_ITD))
> + return;
> +
> + hfi_class = per_cpu_ptr(hfi_ipcc_scores, cpu);
> +
> + for (i = 0; i < hfi_features.nr_classes; i++) {
> + struct hfi_cpu_data *class_caps;
> +
> + class_caps = caps + i * hfi_features.class_stride;
> + WRITE_ONCE(hfi_class[i], class_caps->perf_cap);
As it stands, it is unclear why WRITE_ONCE() is needed here.
> + }
> +}
> +
> +#else
> +static int alloc_hfi_ipcc_scores(void) { return 0; }
> +static void set_hfi_ipcc_score(void *caps, int cpu) { }
> +#endif /* CONFIG_IPC_CLASSES */
> +
> static void get_hfi_caps(struct hfi_instance *hfi_instance,
> struct thermal_genl_cpu_caps *cpu_caps)
> {
> @@ -192,6 +230,8 @@ static void get_hfi_caps(struct hfi_instance *hfi_instance,
> cpu_caps[i].efficiency = caps->ee_cap << 2;
>
> ++i;
> +
> + set_hfi_ipcc_score(caps, cpu);
> }
> raw_spin_unlock_irq(&hfi_instance->table_lock);
> }
> @@ -580,8 +620,14 @@ void __init intel_hfi_init(void)
> if (!hfi_updates_wq)
> goto err_nomem;
>
> + if (alloc_hfi_ipcc_scores())
> + goto err_ipcc;
> +
> return;
>
> +err_ipcc:
> + destroy_workqueue(hfi_updates_wq);
> +
> err_nomem:
> for (j = 0; j < i; ++j) {
> hfi_instance = &hfi_instances[j];
> --
On Mon, Mar 27, 2023 at 06:37:32PM +0200, Rafael J. Wysocki wrote:
> On Tue, Feb 7, 2023 at 6:02 AM Ricardo Neri
> <ricardo.neri-calderon@linux.intel.com> wrote:
> >
> > The scheduler reads the IPCC scores when balancing load. These reads can
> > be quite frequent. Hardware can also update the HFI table frequently.
> > Concurrent access may cause a lot of lock contention. It gets worse as the
> > number of CPUs increases.
> >
> > Instead, create separate per-CPU IPCC scores that the scheduler can read
> > without the HFI table lock.
> >
> > Cc: Ben Segall <bsegall@google.com>
> > Cc: Daniel Bristot de Oliveira <bristot@redhat.com>
> > Cc: Dietmar Eggemann <dietmar.eggemann@arm.com>
> > Cc: Ionela Voinescu <ionela.voinescu@arm.com>
> > Cc: Joel Fernandes (Google) <joel@joelfernandes.org>
> > Cc: Len Brown <len.brown@intel.com>
> > Cc: Lukasz Luba <lukasz.luba@arm.com>
> > Cc: Mel Gorman <mgorman@suse.de>
> > Cc: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
> > Cc: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
> > Cc: Steven Rostedt <rostedt@goodmis.org>
> > Cc: Tim C. Chen <tim.c.chen@intel.com>
> > Cc: Valentin Schneider <vschneid@redhat.com>
> > Cc: x86@kernel.org
> > Cc: linux-pm@vger.kernel.org
> > Cc: linux-kernel@vger.kernel.org
> > Suggested-by: Peter Zijlstra (Intel) <peterz@infradead.org>
> > Signed-off-by: Ricardo Neri <ricardo.neri-calderon@linux.intel.com>
> > ---
> > Changes since v2:
> > * Only create these per-CPU variables when Intel Thread Director is
> > supported.
> >
> > Changes since v1:
> > * Added this patch.
> > ---
> > drivers/thermal/intel/intel_hfi.c | 46 +++++++++++++++++++++++++++++++
> > 1 file changed, 46 insertions(+)
> >
> > diff --git a/drivers/thermal/intel/intel_hfi.c b/drivers/thermal/intel/intel_hfi.c
> > index 2527ae3836c7..b06021828892 100644
> > --- a/drivers/thermal/intel/intel_hfi.c
> > +++ b/drivers/thermal/intel/intel_hfi.c
> > @@ -29,6 +29,7 @@
> > #include <linux/kernel.h>
> > #include <linux/math.h>
> > #include <linux/mutex.h>
> > +#include <linux/percpu.h>
> > #include <linux/percpu-defs.h>
> > #include <linux/printk.h>
> > #include <linux/processor.h>
> > @@ -170,6 +171,43 @@ static struct workqueue_struct *hfi_updates_wq;
> > #define HFI_UPDATE_INTERVAL HZ
> > #define HFI_MAX_THERM_NOTIFY_COUNT 16
> >
> > +#ifdef CONFIG_IPC_CLASSES
>
> It would be good to provide a (concise) description of this variable.
>
> > +static int __percpu *hfi_ipcc_scores;
Do you mean hfi_ipcc_scores or CONFIG_IPC_CLASSES?
> > +
> > +static int alloc_hfi_ipcc_scores(void)
> > +{
> > + if (!cpu_feature_enabled(X86_FEATURE_ITD))
> > + return 0;
> > +
> > + hfi_ipcc_scores = __alloc_percpu(sizeof(*hfi_ipcc_scores) *
> > + hfi_features.nr_classes,
> > + sizeof(*hfi_ipcc_scores));
> > +
> > + return !hfi_ipcc_scores;
>
> I would do
>
> if (!hfi_ipcc_scores)
> return -ENOMEM;
>
> return 0;
>
> Or make the function return bool.
Sure, I can make this function return -ENOMEM.
>
> > +}
> > +
> > +static void set_hfi_ipcc_score(void *caps, int cpu)
> > +{
> > + int i, *hfi_class;
> > +
> > + if (!cpu_feature_enabled(X86_FEATURE_ITD))
> > + return;
> > +
> > + hfi_class = per_cpu_ptr(hfi_ipcc_scores, cpu);
> > +
> > + for (i = 0; i < hfi_features.nr_classes; i++) {
> > + struct hfi_cpu_data *class_caps;
> > +
> > + class_caps = caps + i * hfi_features.class_stride;
> > + WRITE_ONCE(hfi_class[i], class_caps->perf_cap);
>
> As it stands, it is unclear why WRITE_ONCE() is needed here.
The CPU handling the HFI interrupt will update all the per-CPU IPCC
scores. My intention is to ensure that a WRITE of a given IPCC score
is completed before another CPU READs an IPCC score. The corresponding
READ_ONCE happens in patch 15.
On Wed, Mar 29, 2023 at 1:32 AM Ricardo Neri <ricardo.neri-calderon@linux.intel.com> wrote: > > On Mon, Mar 27, 2023 at 06:37:32PM +0200, Rafael J. Wysocki wrote: > > On Tue, Feb 7, 2023 at 6:02 AM Ricardo Neri > > <ricardo.neri-calderon@linux.intel.com> wrote: > > > > > > The scheduler reads the IPCC scores when balancing load. These reads can > > > be quite frequent. Hardware can also update the HFI table frequently. > > > Concurrent access may cause a lot of lock contention. It gets worse as the > > > number of CPUs increases. > > > > > > Instead, create separate per-CPU IPCC scores that the scheduler can read > > > without the HFI table lock. > > > > > > Cc: Ben Segall <bsegall@google.com> > > > Cc: Daniel Bristot de Oliveira <bristot@redhat.com> > > > Cc: Dietmar Eggemann <dietmar.eggemann@arm.com> > > > Cc: Ionela Voinescu <ionela.voinescu@arm.com> > > > Cc: Joel Fernandes (Google) <joel@joelfernandes.org> > > > Cc: Len Brown <len.brown@intel.com> > > > Cc: Lukasz Luba <lukasz.luba@arm.com> > > > Cc: Mel Gorman <mgorman@suse.de> > > > Cc: Rafael J. Wysocki <rafael.j.wysocki@intel.com> > > > Cc: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com> > > > Cc: Steven Rostedt <rostedt@goodmis.org> > > > Cc: Tim C. Chen <tim.c.chen@intel.com> > > > Cc: Valentin Schneider <vschneid@redhat.com> > > > Cc: x86@kernel.org > > > Cc: linux-pm@vger.kernel.org > > > Cc: linux-kernel@vger.kernel.org > > > Suggested-by: Peter Zijlstra (Intel) <peterz@infradead.org> > > > Signed-off-by: Ricardo Neri <ricardo.neri-calderon@linux.intel.com> > > > --- > > > Changes since v2: > > > * Only create these per-CPU variables when Intel Thread Director is > > > supported. > > > > > > Changes since v1: > > > * Added this patch. > > > --- > > > drivers/thermal/intel/intel_hfi.c | 46 +++++++++++++++++++++++++++++++ > > > 1 file changed, 46 insertions(+) > > > > > > diff --git a/drivers/thermal/intel/intel_hfi.c b/drivers/thermal/intel/intel_hfi.c > > > index 2527ae3836c7..b06021828892 100644 > > > --- a/drivers/thermal/intel/intel_hfi.c > > > +++ b/drivers/thermal/intel/intel_hfi.c > > > @@ -29,6 +29,7 @@ > > > #include <linux/kernel.h> > > > #include <linux/math.h> > > > #include <linux/mutex.h> > > > +#include <linux/percpu.h> > > > #include <linux/percpu-defs.h> > > > #include <linux/printk.h> > > > #include <linux/processor.h> > > > @@ -170,6 +171,43 @@ static struct workqueue_struct *hfi_updates_wq; > > > #define HFI_UPDATE_INTERVAL HZ > > > #define HFI_MAX_THERM_NOTIFY_COUNT 16 > > > > > > +#ifdef CONFIG_IPC_CLASSES > > > > It would be good to provide a (concise) description of this variable. > > > > > +static int __percpu *hfi_ipcc_scores; > > Do you mean hfi_ipcc_scores or CONFIG_IPC_CLASSES? hfi_ipcc_scores (as the latter is not a variable).
On Wed, Mar 29, 2023 at 02:08:30PM +0200, Rafael J. Wysocki wrote: > On Wed, Mar 29, 2023 at 1:32 AM Ricardo Neri > <ricardo.neri-calderon@linux.intel.com> wrote: > > > > On Mon, Mar 27, 2023 at 06:37:32PM +0200, Rafael J. Wysocki wrote: > > > On Tue, Feb 7, 2023 at 6:02 AM Ricardo Neri > > > <ricardo.neri-calderon@linux.intel.com> wrote: > > > > > > > > The scheduler reads the IPCC scores when balancing load. These reads can > > > > be quite frequent. Hardware can also update the HFI table frequently. > > > > Concurrent access may cause a lot of lock contention. It gets worse as the > > > > number of CPUs increases. > > > > > > > > Instead, create separate per-CPU IPCC scores that the scheduler can read > > > > without the HFI table lock. > > > > > > > > Cc: Ben Segall <bsegall@google.com> > > > > Cc: Daniel Bristot de Oliveira <bristot@redhat.com> > > > > Cc: Dietmar Eggemann <dietmar.eggemann@arm.com> > > > > Cc: Ionela Voinescu <ionela.voinescu@arm.com> > > > > Cc: Joel Fernandes (Google) <joel@joelfernandes.org> > > > > Cc: Len Brown <len.brown@intel.com> > > > > Cc: Lukasz Luba <lukasz.luba@arm.com> > > > > Cc: Mel Gorman <mgorman@suse.de> > > > > Cc: Rafael J. Wysocki <rafael.j.wysocki@intel.com> > > > > Cc: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com> > > > > Cc: Steven Rostedt <rostedt@goodmis.org> > > > > Cc: Tim C. Chen <tim.c.chen@intel.com> > > > > Cc: Valentin Schneider <vschneid@redhat.com> > > > > Cc: x86@kernel.org > > > > Cc: linux-pm@vger.kernel.org > > > > Cc: linux-kernel@vger.kernel.org > > > > Suggested-by: Peter Zijlstra (Intel) <peterz@infradead.org> > > > > Signed-off-by: Ricardo Neri <ricardo.neri-calderon@linux.intel.com> > > > > --- > > > > Changes since v2: > > > > * Only create these per-CPU variables when Intel Thread Director is > > > > supported. > > > > > > > > Changes since v1: > > > > * Added this patch. > > > > --- > > > > drivers/thermal/intel/intel_hfi.c | 46 +++++++++++++++++++++++++++++++ > > > > 1 file changed, 46 insertions(+) > > > > > > > > diff --git a/drivers/thermal/intel/intel_hfi.c b/drivers/thermal/intel/intel_hfi.c > > > > index 2527ae3836c7..b06021828892 100644 > > > > --- a/drivers/thermal/intel/intel_hfi.c > > > > +++ b/drivers/thermal/intel/intel_hfi.c > > > > @@ -29,6 +29,7 @@ > > > > #include <linux/kernel.h> > > > > #include <linux/math.h> > > > > #include <linux/mutex.h> > > > > +#include <linux/percpu.h> > > > > #include <linux/percpu-defs.h> > > > > #include <linux/printk.h> > > > > #include <linux/processor.h> > > > > @@ -170,6 +171,43 @@ static struct workqueue_struct *hfi_updates_wq; > > > > #define HFI_UPDATE_INTERVAL HZ > > > > #define HFI_MAX_THERM_NOTIFY_COUNT 16 > > > > > > > > +#ifdef CONFIG_IPC_CLASSES > > > > > > It would be good to provide a (concise) description of this variable. > > > > > > > +static int __percpu *hfi_ipcc_scores; > > > > Do you mean hfi_ipcc_scores or CONFIG_IPC_CLASSES? > > hfi_ipcc_scores (as the latter is not a variable). I thought so. Thank you for clarifying.
© 2016 - 2026 Red Hat, Inc.