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

3 files changed, 265 insertions, 0 deletions

diff --git a/arch/powerpc/sysdev/ge/Makefile b/arch/powerpc/sysdev/ge/Makefile
new file mode 100644
index 000000000..a63fdb379
--- /dev/null
+++ b/arch/powerpc/sysdev/ge/Makefile
@@ -0,0 +1,2 @@
+# SPDX-License-Identifier: GPL-2.0-only
+obj-$(CONFIG_GE_FPGA)		+= ge_pic.o
diff --git a/arch/powerpc/sysdev/ge/ge_pic.c b/arch/powerpc/sysdev/ge/ge_pic.c
new file mode 100644
index 000000000..0bc3f0b36
--- /dev/null
+++ b/arch/powerpc/sysdev/ge/ge_pic.c
@@ -0,0 +1,254 @@
+/*
+ * Interrupt handling for GE FPGA based PIC
+ *
+ * Author: Martyn Welch <martyn.welch@ge.com>
+ *
+ * 2008 (c) GE Intelligent Platforms Embedded Systems, Inc.
+ *
+ * This file is licensed under the terms of the GNU General Public License
+ * version 2.  This program is licensed "as is" without any warranty of any
+ * kind, whether express or implied.
+ */
+
+#include <linux/stddef.h>
+#include <linux/kernel.h>
+#include <linux/init.h>
+#include <linux/irq.h>
+#include <linux/irqdomain.h>
+#include <linux/interrupt.h>
+#include <linux/of_address.h>
+#include <linux/of_irq.h>
+#include <linux/spinlock.h>
+
+#include <asm/byteorder.h>
+#include <asm/io.h>
+#include <asm/irq.h>
+
+#include "ge_pic.h"
+
+#define DEBUG
+#undef DEBUG
+
+#ifdef DEBUG
+#define DBG(fmt...) do { printk(KERN_DEBUG "gef_pic: " fmt); } while (0)
+#else
+#define DBG(fmt...) do { } while (0)
+#endif
+
+#define GEF_PIC_NUM_IRQS	32
+
+/* Interrupt Controller Interface Registers */
+#define GEF_PIC_INTR_STATUS	0x0000
+
+#define GEF_PIC_INTR_MASK(cpu)	(0x0010 + (0x4 * cpu))
+#define GEF_PIC_CPU0_INTR_MASK	GEF_PIC_INTR_MASK(0)
+#define GEF_PIC_CPU1_INTR_MASK	GEF_PIC_INTR_MASK(1)
+
+#define GEF_PIC_MCP_MASK(cpu)	(0x0018 + (0x4 * cpu))
+#define GEF_PIC_CPU0_MCP_MASK	GEF_PIC_MCP_MASK(0)
+#define GEF_PIC_CPU1_MCP_MASK	GEF_PIC_MCP_MASK(1)
+
+
+static DEFINE_RAW_SPINLOCK(gef_pic_lock);
+
+static void __iomem *gef_pic_irq_reg_base;
+static struct irq_domain *gef_pic_irq_host;
+static int gef_pic_cascade_irq;
+
+/*
+ * Interrupt Controller Handling
+ *
+ * The interrupt controller handles interrupts for most on board interrupts,
+ * apart from PCI interrupts. For example on SBC610:
+ *
+ * 17:31 RO Reserved
+ * 16    RO PCI Express Doorbell 3 Status
+ * 15    RO PCI Express Doorbell 2 Status
+ * 14    RO PCI Express Doorbell 1 Status
+ * 13    RO PCI Express Doorbell 0 Status
+ * 12    RO Real Time Clock Interrupt Status
+ * 11    RO Temperature Interrupt Status
+ * 10    RO Temperature Critical Interrupt Status
+ * 9     RO Ethernet PHY1 Interrupt Status
+ * 8     RO Ethernet PHY3 Interrupt Status
+ * 7     RO PEX8548 Interrupt Status
+ * 6     RO Reserved
+ * 5     RO Watchdog 0 Interrupt Status
+ * 4     RO Watchdog 1 Interrupt Status
+ * 3     RO AXIS Message FIFO A Interrupt Status
+ * 2     RO AXIS Message FIFO B Interrupt Status
+ * 1     RO AXIS Message FIFO C Interrupt Status
+ * 0     RO AXIS Message FIFO D Interrupt Status
+ *
+ * Interrupts can be forwarded to one of two output lines. Nothing
+ * clever is done, so if the masks are incorrectly set, a single input
+ * interrupt could generate interrupts on both output lines!
+ *
+ * The dual lines are there to allow the chained interrupts to be easily
+ * passed into two different cores. We currently do not use this functionality
+ * in this driver.
+ *
+ * Controller can also be configured to generate Machine checks (MCP), again on
+ * two lines, to be attached to two different cores. It is suggested that these
+ * should be masked out.
+ */
+
+static void gef_pic_cascade(struct irq_desc *desc)
+{
+	struct irq_chip *chip = irq_desc_get_chip(desc);
+	unsigned int cascade_irq;
+
+	/*
+	 * See if we actually have an interrupt, call generic handling code if
+	 * we do.
+	 */
+	cascade_irq = gef_pic_get_irq();
+
+	if (cascade_irq)
+		generic_handle_irq(cascade_irq);
+
+	chip->irq_eoi(&desc->irq_data);
+}
+
+static void gef_pic_mask(struct irq_data *d)
+{
+	unsigned long flags;
+	unsigned int hwirq = irqd_to_hwirq(d);
+	u32 mask;
+
+	raw_spin_lock_irqsave(&gef_pic_lock, flags);
+	mask = in_be32(gef_pic_irq_reg_base + GEF_PIC_INTR_MASK(0));
+	mask &= ~(1 << hwirq);
+	out_be32(gef_pic_irq_reg_base + GEF_PIC_INTR_MASK(0), mask);
+	raw_spin_unlock_irqrestore(&gef_pic_lock, flags);
+}
+
+static void gef_pic_mask_ack(struct irq_data *d)
+{
+	/* Don't think we actually have to do anything to ack an interrupt,
+	 * we just need to clear down the devices interrupt and it will go away
+	 */
+	gef_pic_mask(d);
+}
+
+static void gef_pic_unmask(struct irq_data *d)
+{
+	unsigned long flags;
+	unsigned int hwirq = irqd_to_hwirq(d);
+	u32 mask;
+
+	raw_spin_lock_irqsave(&gef_pic_lock, flags);
+	mask = in_be32(gef_pic_irq_reg_base + GEF_PIC_INTR_MASK(0));
+	mask |= (1 << hwirq);
+	out_be32(gef_pic_irq_reg_base + GEF_PIC_INTR_MASK(0), mask);
+	raw_spin_unlock_irqrestore(&gef_pic_lock, flags);
+}
+
+static struct irq_chip gef_pic_chip = {
+	.name		= "gefp",
+	.irq_mask	= gef_pic_mask,
+	.irq_mask_ack	= gef_pic_mask_ack,
+	.irq_unmask	= gef_pic_unmask,
+};
+
+
+/* When an interrupt is being configured, this call allows some flexibility
+ * in deciding which irq_chip structure is used
+ */
+static int gef_pic_host_map(struct irq_domain *h, unsigned int virq,
+			  irq_hw_number_t hwirq)
+{
+	/* All interrupts are LEVEL sensitive */
+	irq_set_status_flags(virq, IRQ_LEVEL);
+	irq_set_chip_and_handler(virq, &gef_pic_chip, handle_level_irq);
+
+	return 0;
+}
+
+static int gef_pic_host_xlate(struct irq_domain *h, struct device_node *ct,
+			    const u32 *intspec, unsigned int intsize,
+			    irq_hw_number_t *out_hwirq, unsigned int *out_flags)
+{
+
+	*out_hwirq = intspec[0];
+	if (intsize > 1)
+		*out_flags = intspec[1];
+	else
+		*out_flags = IRQ_TYPE_LEVEL_HIGH;
+
+	return 0;
+}
+
+static const struct irq_domain_ops gef_pic_host_ops = {
+	.map	= gef_pic_host_map,
+	.xlate	= gef_pic_host_xlate,
+};
+
+
+/*
+ * Initialisation of PIC, this should be called in BSP
+ */
+void __init gef_pic_init(struct device_node *np)
+{
+	unsigned long flags;
+
+	/* Map the devices registers into memory */
+	gef_pic_irq_reg_base = of_iomap(np, 0);
+
+	raw_spin_lock_irqsave(&gef_pic_lock, flags);
+
+	/* Initialise everything as masked. */
+	out_be32(gef_pic_irq_reg_base + GEF_PIC_CPU0_INTR_MASK, 0);
+	out_be32(gef_pic_irq_reg_base + GEF_PIC_CPU1_INTR_MASK, 0);
+
+	out_be32(gef_pic_irq_reg_base + GEF_PIC_CPU0_MCP_MASK, 0);
+	out_be32(gef_pic_irq_reg_base + GEF_PIC_CPU1_MCP_MASK, 0);
+
+	raw_spin_unlock_irqrestore(&gef_pic_lock, flags);
+
+	/* Map controller */
+	gef_pic_cascade_irq = irq_of_parse_and_map(np, 0);
+	if (!gef_pic_cascade_irq) {
+		printk(KERN_ERR "SBC610: failed to map cascade interrupt");
+		return;
+	}
+
+	/* Setup an irq_domain structure */
+	gef_pic_irq_host = irq_domain_create_linear(of_fwnode_handle(np),
+						    GEF_PIC_NUM_IRQS,
+						    &gef_pic_host_ops, NULL);
+	if (gef_pic_irq_host == NULL)
+		return;
+
+	/* Chain with parent controller */
+	irq_set_chained_handler(gef_pic_cascade_irq, gef_pic_cascade);
+}
+
+/*
+ * This is called when we receive an interrupt with apparently comes from this
+ * chip - check, returning the highest interrupt generated or return 0.
+ */
+unsigned int gef_pic_get_irq(void)
+{
+	u32 cause, mask, active;
+	unsigned int virq = 0;
+	int hwirq;
+
+	cause = in_be32(gef_pic_irq_reg_base + GEF_PIC_INTR_STATUS);
+
+	mask = in_be32(gef_pic_irq_reg_base + GEF_PIC_INTR_MASK(0));
+
+	active = cause & mask;
+
+	if (active) {
+		for (hwirq = GEF_PIC_NUM_IRQS - 1; hwirq > -1; hwirq--) {
+			if (active & (0x1 << hwirq))
+				break;
+		}
+		virq = irq_find_mapping(gef_pic_irq_host,
+			(irq_hw_number_t)hwirq);
+	}
+
+	return virq;
+}
+
diff --git a/arch/powerpc/sysdev/ge/ge_pic.h b/arch/powerpc/sysdev/ge/ge_pic.h
new file mode 100644
index 000000000..923dedba3
--- /dev/null
+++ b/arch/powerpc/sysdev/ge/ge_pic.h
@@ -0,0 +1,9 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef __GEF_PIC_H__
+#define __GEF_PIC_H__
+
+unsigned int gef_pic_get_irq(void);
+void gef_pic_init(struct device_node *);
+
+#endif /* __GEF_PIC_H__ */
+