[PATCH v2 19/22] thermal: intel: hfi: Implement model-specific checks for task classification

From: Ricardo Neri
Date: Mon Nov 28 2022 - 08:14:55 EST


In Alderlake and Raptorlake, the result of thread classification is more
accurate when only one SMT sibling is busy. Classification results for
class 2 and 3 that are always reliable.

To avoid unnecessary migrations, only update the class of a task if it has
been the same during 4 consecutive ticks.

Cc: Ben Segall <bsegall@xxxxxxxxxx>
Cc: Daniel Bristot de Oliveira <bristot@xxxxxxxxxx>
Cc: Dietmar Eggemann <dietmar.eggemann@xxxxxxx>
Cc: Joel Fernandes (Google) <joel@xxxxxxxxxxxxxxxxx>
Cc: Len Brown <len.brown@xxxxxxxxx>
Cc: Mel Gorman <mgorman@xxxxxxx>
Cc: Rafael J. Wysocki <rafael.j.wysocki@xxxxxxxxx>
Cc: Srinivas Pandruvada <srinivas.pandruvada@xxxxxxxxxxxxxxx>
Cc: Steven Rostedt <rostedt@xxxxxxxxxxx>
Cc: Tim C. Chen <tim.c.chen@xxxxxxxxx>
Cc: Valentin Schneider <vschneid@xxxxxxxxxx>
Cc: x86@xxxxxxxxxx
Cc: linux-pm@xxxxxxxxxxxxxxx
Cc: linux-kernel@xxxxxxxxxxxxxxx
Signed-off-by: Ricardo Neri <ricardo.neri-calderon@xxxxxxxxxxxxxxx>
---
Changes since v1:
* Adjusted the result the classification of Intel Thread Director to start
at class 1. Class 0 for the scheduler means that the task is
unclassified.
* Used the new names of the IPC classes members in task_struct.
* Reworked helper functions to use sched_smt_siblings_idle() to query
the idle state of the SMT siblings of a CPU.
---
drivers/thermal/intel/intel_hfi.c | 60 ++++++++++++++++++++++++++++++-
1 file changed, 59 insertions(+), 1 deletion(-)

diff --git a/drivers/thermal/intel/intel_hfi.c b/drivers/thermal/intel/intel_hfi.c
index 8287bfd7d6b6..a9ae09036909 100644
--- a/drivers/thermal/intel/intel_hfi.c
+++ b/drivers/thermal/intel/intel_hfi.c
@@ -40,6 +40,7 @@
#include <linux/workqueue.h>

#include <asm/msr.h>
+#include <asm/intel-family.h>

#include "../thermal_core.h"
#include "intel_hfi.h"
@@ -216,9 +217,64 @@ int intel_hfi_has_ipc_classes(void)
return cpu_feature_enabled(X86_FEATURE_ITD);
}

+#define CLASS_DEBOUNCER_SKIPS 4
+
+/**
+ * debounce_and_update_class() - Process and update a task's classification
+ *
+ * @p: The task of which the classification will be updated
+ * @new_ipcc: The new IPC classification
+ *
+ * Update the classification of @p with the new value that hardware provides.
+ * Only update the classification of @p if it has been the same during
+ * CLASS_DEBOUNCER_SKIPS consecutive ticks.
+ */
+static void debounce_and_update_class(struct task_struct *p, u8 new_ipcc)
+{
+ u16 debounce_skip;
+
+ /* The class of @p changed, only restart the debounce counter. */
+ if (p->ipcc_tmp != new_ipcc) {
+ p->ipcc_cntr = 1;
+ goto out;
+ }
+
+ /*
+ * The class of @p did not change. Update it if it has been the same
+ * for CLASS_DEBOUNCER_SKIPS user ticks.
+ */
+ debounce_skip = p->ipcc_cntr + 1;
+ if (debounce_skip < CLASS_DEBOUNCER_SKIPS)
+ p->ipcc_cntr++;
+ else
+ p->ipcc = new_ipcc;
+
+out:
+ p->ipcc_tmp = new_ipcc;
+}
+
+static bool classification_is_accurate(u8 hfi_class, bool smt_siblings_idle)
+{
+ switch (boot_cpu_data.x86_model) {
+ case INTEL_FAM6_ALDERLAKE:
+ case INTEL_FAM6_ALDERLAKE_L:
+ case INTEL_FAM6_RAPTORLAKE:
+ case INTEL_FAM6_RAPTORLAKE_P:
+ case INTEL_FAM6_RAPTORLAKE_S:
+ if (hfi_class == 3 || hfi_class == 2 || smt_siblings_idle)
+ return true;
+
+ return false;
+
+ default:
+ return true;
+ }
+}
+
void intel_hfi_update_ipcc(struct task_struct *curr)
{
union hfi_thread_feedback_char_msr msr;
+ bool idle;

/* We should not be here if ITD is not supported. */
if (!cpu_feature_enabled(X86_FEATURE_ITD)) {
@@ -234,7 +290,9 @@ void intel_hfi_update_ipcc(struct task_struct *curr)
* 0 is a valid classification for Intel Thread Director. A scheduler
* IPCC class of 0 means that the task is unclassified. Adjust.
*/
- curr->ipcc = msr.split.classid + 1;
+ idle = sched_smt_siblings_idle(task_cpu(curr));
+ if (classification_is_accurate(msr.split.classid, idle))
+ debounce_and_update_class(curr, msr.split.classid + 1);
}

int intel_hfi_get_ipcc_score(unsigned short ipcc, int cpu)
--
2.25.1