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, 168 insertions, 0 deletions

diff --git a/drivers/usb/cdns3/host.c b/drivers/usb/cdns3/host.c
new file mode 100644
index 000000000..f0df114c2
--- /dev/null
+++ b/drivers/usb/cdns3/host.c
@@ -0,0 +1,168 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Cadence USBSS and USBSSP DRD Driver - host side
+ *
+ * Copyright (C) 2018-2019 Cadence Design Systems.
+ * Copyright (C) 2017-2018 NXP
+ *
+ * Authors: Peter Chen <peter.chen@nxp.com>
+ *          Pawel Laszczak <pawell@cadence.com>
+ */
+
+#include <linux/platform_device.h>
+#include <linux/slab.h>
+#include "core.h"
+#include "drd.h"
+#include "host-export.h"
+#include <linux/usb/hcd.h>
+#include "../host/xhci.h"
+#include "../host/xhci-plat.h"
+
+/*
+ * The XECP_PORT_CAP_REG and XECP_AUX_CTRL_REG1 exist only
+ * in Cadence USB3 dual-role controller, so it can't be used
+ * with Cadence CDNSP dual-role controller.
+ */
+#define XECP_PORT_CAP_REG	0x8000
+#define XECP_AUX_CTRL_REG1	0x8120
+
+#define CFG_RXDET_P3_EN		BIT(15)
+#define LPM_2_STB_SWITCH_EN	BIT(25)
+
+static void xhci_cdns3_plat_start(struct usb_hcd *hcd)
+{
+	struct xhci_hcd *xhci = hcd_to_xhci(hcd);
+	u32 value;
+
+	/* set usbcmd.EU3S */
+	value = readl(&xhci->op_regs->command);
+	value |= CMD_PM_INDEX;
+	writel(value, &xhci->op_regs->command);
+
+	if (hcd->regs) {
+		value = readl(hcd->regs + XECP_AUX_CTRL_REG1);
+		value |= CFG_RXDET_P3_EN;
+		writel(value, hcd->regs + XECP_AUX_CTRL_REG1);
+
+		value = readl(hcd->regs + XECP_PORT_CAP_REG);
+		value |= LPM_2_STB_SWITCH_EN;
+		writel(value, hcd->regs + XECP_PORT_CAP_REG);
+	}
+}
+
+static int xhci_cdns3_resume_quirk(struct usb_hcd *hcd)
+{
+	xhci_cdns3_plat_start(hcd);
+	return 0;
+}
+
+static const struct xhci_plat_priv xhci_plat_cdns3_xhci = {
+	.quirks = XHCI_SKIP_PHY_INIT | XHCI_AVOID_BEI,
+	.plat_start = xhci_cdns3_plat_start,
+	.resume_quirk = xhci_cdns3_resume_quirk,
+};
+
+static const struct xhci_plat_priv xhci_plat_cdnsp_xhci = {
+	.quirks = XHCI_CDNS_SCTX_QUIRK,
+};
+
+static int __cdns_host_init(struct cdns *cdns)
+{
+	struct platform_device *xhci;
+	int ret;
+	struct usb_hcd *hcd;
+
+	cdns_drd_host_on(cdns);
+
+	xhci = platform_device_alloc("xhci-hcd", PLATFORM_DEVID_AUTO);
+	if (!xhci) {
+		dev_err(cdns->dev, "couldn't allocate xHCI device\n");
+		return -ENOMEM;
+	}
+
+	xhci->dev.parent = cdns->dev;
+	cdns->host_dev = xhci;
+
+	ret = platform_device_add_resources(xhci, cdns->xhci_res,
+					    CDNS_XHCI_RESOURCES_NUM);
+	if (ret) {
+		dev_err(cdns->dev, "couldn't add resources to xHCI device\n");
+		goto err1;
+	}
+
+	if (cdns->version < CDNSP_CONTROLLER_V2)
+		cdns->xhci_plat_data = kmemdup(&xhci_plat_cdns3_xhci,
+				sizeof(struct xhci_plat_priv), GFP_KERNEL);
+	else
+		cdns->xhci_plat_data = kmemdup(&xhci_plat_cdnsp_xhci,
+				sizeof(struct xhci_plat_priv), GFP_KERNEL);
+
+	if (!cdns->xhci_plat_data) {
+		ret = -ENOMEM;
+		goto err1;
+	}
+
+	if (cdns->pdata && (cdns->pdata->quirks & CDNS3_DEFAULT_PM_RUNTIME_ALLOW))
+		cdns->xhci_plat_data->quirks |= XHCI_DEFAULT_PM_RUNTIME_ALLOW;
+
+	ret = platform_device_add_data(xhci, cdns->xhci_plat_data,
+			sizeof(struct xhci_plat_priv));
+	if (ret)
+		goto free_memory;
+
+	ret = platform_device_add(xhci);
+	if (ret) {
+		dev_err(cdns->dev, "failed to register xHCI device\n");
+		goto free_memory;
+	}
+
+	/* Glue needs to access xHCI region register for Power management */
+	hcd = platform_get_drvdata(xhci);
+	if (hcd)
+		cdns->xhci_regs = hcd->regs;
+
+	return 0;
+
+free_memory:
+	kfree(cdns->xhci_plat_data);
+err1:
+	platform_device_put(xhci);
+	return ret;
+}
+
+static void cdns_host_exit(struct cdns *cdns)
+{
+	kfree(cdns->xhci_plat_data);
+	platform_device_unregister(cdns->host_dev);
+	cdns->host_dev = NULL;
+	cdns_drd_host_off(cdns);
+}
+
+static int cdns_host_resume(struct cdns *cdns, bool power_lost)
+{
+	struct usb_hcd *hcd = platform_get_drvdata(cdns->host_dev);
+	struct xhci_plat_priv *priv = hcd_to_xhci_priv(hcd);
+
+	priv->power_lost = power_lost;
+
+	return 0;
+}
+
+int cdns_host_init(struct cdns *cdns)
+{
+	struct cdns_role_driver *rdrv;
+
+	rdrv = devm_kzalloc(cdns->dev, sizeof(*rdrv), GFP_KERNEL);
+	if (!rdrv)
+		return -ENOMEM;
+
+	rdrv->start	= __cdns_host_init;
+	rdrv->stop	= cdns_host_exit;
+	rdrv->resume	= cdns_host_resume;
+	rdrv->state	= CDNS_ROLE_STATE_INACTIVE;
+	rdrv->name	= "host";
+
+	cdns->roles[USB_ROLE_HOST] = rdrv;
+
+	return 0;
+}