sched/cache: Keep nr_pref_llc_running in the runnable domain, to fix LLC mis-scheduling buggrafted

alb_break_llc() decides whether to break LLC preference during active
load balance. It does so by testing that every runnable fair task on the
source rq prefers its LLC:

	env->src_rq->nr_pref_llc_running == env->src_rq->cfs.h_nr_runnable

But the two counters cover different sets. nr_pref_llc_running is updated
in account_llc_enqueue()/account_llc_dequeue(), next to cfs_rq->nr_queued,
so it follows queued tasks. h_nr_runnable is updated in set_delayed()/
clear_delayed() and drops delay-dequeued tasks.

So under DELAY_DEQUEUE, a preferring task that goes to sleep stays counted
in nr_pref_llc_running while h_nr_runnable falls. The equality then breaks,
alb_break_llc() returns false, and active balance is free to pull a task
off its preferred LLC. Active balance only moves runnable tasks, and this
is the only LLC check it consults: once the stopper runs, LBF_ACTIVE_LB
skips the per-task test in can_migrate_task(). The runnable set is the one
we want.

Fix it on the counter side. A task should be counted in
nr_pref_llc_running exactly while it is both queued on its preferred LLC
(pref_llc_queued) and runnable (!sched_delayed). Define that membership
once in task_pref_llc_runnable(), and adjust the counter only through
pref_llc_running_inc()/pref_llc_running_dec() from the four sites that
change either input: account_llc_enqueue(), account_llc_dequeue(),
set_delayed() and clear_delayed(). Gating every update on the same
predicate keeps the delay, wake and dequeue paths from double-counting
or underflowing; see the comments at those sites for the ordering.

nr_llc_running and sd->llc_counts are not touched and stay on queued
semantics.

Fixes: 714059f79ff0 ("sched/cache: Handle moving single tasks to/from their preferred LLC")
Closes: https://lore.kernel.org/lkml/20260827135000.735138-1-zhanxusheng@xiaomi.com/
Reported-by: Zhan Xusheng <zhanxusheng@xiaomi.com>
Suggested-by: Chen Yu <yu.c.chen@intel.com>
Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Reviewed-by: Kayra Cizmeci <kayracizmeci@gmail.com>
Cc: <stable@kernel.org> # v7.2.x
Link: https://patch.msgid.link/06af61afedac32e6477f57feb4d658f6c411c3af.1790035273.git.tim.c.chen@linux.intel.com

1 files changed, 257 insertions, 0 deletions

diff --git a/drivers/hwmon/occ/sysfs.c b/drivers/hwmon/occ/sysfs.c
new file mode 100644
index 000000000..2317301fc
--- /dev/null
+++ b/drivers/hwmon/occ/sysfs.c
@@ -0,0 +1,257 @@
+// SPDX-License-Identifier: GPL-2.0+
+// Copyright IBM Corp 2019
+
+#include <linux/bitops.h>
+#include <linux/device.h>
+#include <linux/export.h>
+#include <linux/hwmon-sysfs.h>
+#include <linux/kernel.h>
+#include <linux/kstrtox.h>
+#include <linux/sysfs.h>
+
+#include "common.h"
+
+/* OCC status register */
+#define OCC_STAT_MASTER			BIT(7)
+
+/* OCC extended status register */
+#define OCC_EXT_STAT_DVFS_OT		BIT(7)
+#define OCC_EXT_STAT_DVFS_POWER		BIT(6)
+#define OCC_EXT_STAT_MEM_THROTTLE	BIT(5)
+#define OCC_EXT_STAT_QUICK_DROP		BIT(4)
+#define OCC_EXT_STAT_DVFS_VDD		BIT(3)
+#define OCC_EXT_STAT_GPU_THROTTLE	GENMASK(2, 0)
+
+static ssize_t occ_active_store(struct device *dev,
+				struct device_attribute *attr,
+				const char *buf, size_t count)
+{
+	int rc;
+	bool active;
+	struct occ *occ = dev_get_drvdata(dev);
+
+	rc = kstrtobool(buf, &active);
+	if (rc)
+		return rc;
+
+	rc = occ_active(occ, active);
+	if (rc)
+		return rc;
+
+	return count;
+}
+
+static ssize_t occ_sysfs_show(struct device *dev,
+			      struct device_attribute *attr, char *buf)
+{
+	int rc;
+	int val = 0;
+	struct occ *occ = dev_get_drvdata(dev);
+	struct occ_poll_response_header *header;
+	struct sensor_device_attribute *sattr = to_sensor_dev_attr(attr);
+
+	if (occ->active) {
+		rc = occ_update_response(occ);
+		if (rc)
+			return rc;
+
+		header = (struct occ_poll_response_header *)occ->resp.data;
+
+		switch (sattr->index) {
+		case 0:
+			val = !!(header->status & OCC_STAT_MASTER);
+			break;
+		case 1:
+			val = 1;
+			break;
+		case 2:
+			val = !!(header->ext_status & OCC_EXT_STAT_DVFS_OT);
+			break;
+		case 3:
+			val = !!(header->ext_status & OCC_EXT_STAT_DVFS_POWER);
+			break;
+		case 4:
+			val = !!(header->ext_status &
+				 OCC_EXT_STAT_MEM_THROTTLE);
+			break;
+		case 5:
+			val = !!(header->ext_status & OCC_EXT_STAT_QUICK_DROP);
+			break;
+		case 6:
+			val = header->occ_state;
+			break;
+		case 7:
+			if (header->status & OCC_STAT_MASTER)
+				val = hweight8(header->occs_present);
+			else
+				val = 1;
+			break;
+		case 8:
+			val = header->ips_status;
+			break;
+		case 9:
+			val = header->mode;
+			break;
+		case 10:
+			val = !!(header->ext_status & OCC_EXT_STAT_DVFS_VDD);
+			break;
+		case 11:
+			val = header->ext_status & OCC_EXT_STAT_GPU_THROTTLE;
+			break;
+		default:
+			return -EINVAL;
+		}
+	} else {
+		if (sattr->index == 1)
+			val = 0;
+		else if (sattr->index <= 11)
+			val = -ENODATA;
+		else
+			return -EINVAL;
+	}
+
+	return sysfs_emit(buf, "%d\n", val);
+}
+
+static ssize_t occ_error_show(struct device *dev,
+			      struct device_attribute *attr, char *buf)
+{
+	struct occ *occ = dev_get_drvdata(dev);
+
+	occ_update_response(occ);
+
+	return sysfs_emit(buf, "%d\n", occ->error);
+}
+
+static SENSOR_DEVICE_ATTR(occ_master, 0444, occ_sysfs_show, NULL, 0);
+static SENSOR_DEVICE_ATTR(occ_active, 0644, occ_sysfs_show, occ_active_store,
+			  1);
+static SENSOR_DEVICE_ATTR(occ_dvfs_overtemp, 0444, occ_sysfs_show, NULL, 2);
+static SENSOR_DEVICE_ATTR(occ_dvfs_power, 0444, occ_sysfs_show, NULL, 3);
+static SENSOR_DEVICE_ATTR(occ_mem_throttle, 0444, occ_sysfs_show, NULL, 4);
+static SENSOR_DEVICE_ATTR(occ_quick_pwr_drop, 0444, occ_sysfs_show, NULL, 5);
+static SENSOR_DEVICE_ATTR(occ_state, 0444, occ_sysfs_show, NULL, 6);
+static SENSOR_DEVICE_ATTR(occs_present, 0444, occ_sysfs_show, NULL, 7);
+static SENSOR_DEVICE_ATTR(occ_ips_status, 0444, occ_sysfs_show, NULL, 8);
+static SENSOR_DEVICE_ATTR(occ_mode, 0444, occ_sysfs_show, NULL, 9);
+static SENSOR_DEVICE_ATTR(occ_dvfs_vdd, 0444, occ_sysfs_show, NULL, 10);
+static SENSOR_DEVICE_ATTR(occ_gpu_throttle, 0444, occ_sysfs_show, NULL, 11);
+static DEVICE_ATTR_RO(occ_error);
+
+static struct attribute *occ_attributes[] = {
+	&sensor_dev_attr_occ_master.dev_attr.attr,
+	&sensor_dev_attr_occ_active.dev_attr.attr,
+	&sensor_dev_attr_occ_dvfs_overtemp.dev_attr.attr,
+	&sensor_dev_attr_occ_dvfs_power.dev_attr.attr,
+	&sensor_dev_attr_occ_mem_throttle.dev_attr.attr,
+	&sensor_dev_attr_occ_quick_pwr_drop.dev_attr.attr,
+	&sensor_dev_attr_occ_state.dev_attr.attr,
+	&sensor_dev_attr_occs_present.dev_attr.attr,
+	&sensor_dev_attr_occ_ips_status.dev_attr.attr,
+	&sensor_dev_attr_occ_mode.dev_attr.attr,
+	&sensor_dev_attr_occ_dvfs_vdd.dev_attr.attr,
+	&sensor_dev_attr_occ_gpu_throttle.dev_attr.attr,
+	&dev_attr_occ_error.attr,
+	NULL
+};
+
+static const struct attribute_group occ_sysfs = {
+	.attrs = occ_attributes,
+};
+
+void occ_sysfs_poll_done(struct occ *occ)
+{
+	const char *name;
+	struct occ_poll_response_header *header =
+		(struct occ_poll_response_header *)occ->resp.data;
+
+	/*
+	 * On the first poll response, we haven't yet created the sysfs
+	 * attributes, so don't make any notify calls.
+	 */
+	if (!occ->active)
+		goto done;
+
+	if ((header->status & OCC_STAT_MASTER) !=
+	    (occ->prev_stat & OCC_STAT_MASTER)) {
+		name = sensor_dev_attr_occ_master.dev_attr.attr.name;
+		sysfs_notify(&occ->bus_dev->kobj, NULL, name);
+	}
+
+	if ((header->ext_status & OCC_EXT_STAT_DVFS_OT) !=
+	    (occ->prev_ext_stat & OCC_EXT_STAT_DVFS_OT)) {
+		name = sensor_dev_attr_occ_dvfs_overtemp.dev_attr.attr.name;
+		sysfs_notify(&occ->bus_dev->kobj, NULL, name);
+	}
+
+	if ((header->ext_status & OCC_EXT_STAT_DVFS_POWER) !=
+	    (occ->prev_ext_stat & OCC_EXT_STAT_DVFS_POWER)) {
+		name = sensor_dev_attr_occ_dvfs_power.dev_attr.attr.name;
+		sysfs_notify(&occ->bus_dev->kobj, NULL, name);
+	}
+
+	if ((header->ext_status & OCC_EXT_STAT_MEM_THROTTLE) !=
+	    (occ->prev_ext_stat & OCC_EXT_STAT_MEM_THROTTLE)) {
+		name = sensor_dev_attr_occ_mem_throttle.dev_attr.attr.name;
+		sysfs_notify(&occ->bus_dev->kobj, NULL, name);
+	}
+
+	if ((header->ext_status & OCC_EXT_STAT_QUICK_DROP) !=
+	    (occ->prev_ext_stat & OCC_EXT_STAT_QUICK_DROP)) {
+		name = sensor_dev_attr_occ_quick_pwr_drop.dev_attr.attr.name;
+		sysfs_notify(&occ->bus_dev->kobj, NULL, name);
+	}
+
+	if ((header->ext_status & OCC_EXT_STAT_DVFS_VDD) !=
+	    (occ->prev_ext_stat & OCC_EXT_STAT_DVFS_VDD)) {
+		name = sensor_dev_attr_occ_dvfs_vdd.dev_attr.attr.name;
+		sysfs_notify(&occ->bus_dev->kobj, NULL, name);
+	}
+
+	if ((header->ext_status & OCC_EXT_STAT_GPU_THROTTLE) !=
+	    (occ->prev_ext_stat & OCC_EXT_STAT_GPU_THROTTLE)) {
+		name = sensor_dev_attr_occ_gpu_throttle.dev_attr.attr.name;
+		sysfs_notify(&occ->bus_dev->kobj, NULL, name);
+	}
+
+	if ((header->status & OCC_STAT_MASTER) &&
+	    header->occs_present != occ->prev_occs_present) {
+		name = sensor_dev_attr_occs_present.dev_attr.attr.name;
+		sysfs_notify(&occ->bus_dev->kobj, NULL, name);
+	}
+
+	if (header->ips_status != occ->prev_ips_status) {
+		name = sensor_dev_attr_occ_ips_status.dev_attr.attr.name;
+		sysfs_notify(&occ->bus_dev->kobj, NULL, name);
+	}
+
+	if (header->mode != occ->prev_mode) {
+		name = sensor_dev_attr_occ_mode.dev_attr.attr.name;
+		sysfs_notify(&occ->bus_dev->kobj, NULL, name);
+	}
+
+	if (occ->error && occ->error != occ->prev_error) {
+		name = dev_attr_occ_error.attr.name;
+		sysfs_notify(&occ->bus_dev->kobj, NULL, name);
+	}
+
+	/* no notifications for OCC state; doesn't indicate error condition */
+
+done:
+	occ->prev_error = occ->error;
+	occ->prev_stat = header->status;
+	occ->prev_ext_stat = header->ext_status;
+	occ->prev_occs_present = header->occs_present;
+	occ->prev_ips_status = header->ips_status;
+	occ->prev_mode = header->mode;
+}
+
+int occ_setup_sysfs(struct occ *occ)
+{
+	return sysfs_create_group(&occ->bus_dev->kobj, &occ_sysfs);
+}
+
+void occ_shutdown_sysfs(struct occ *occ)
+{
+	sysfs_remove_group(&occ->bus_dev->kobj, &occ_sysfs);
+}