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, 121 insertions, 0 deletions

diff --git a/drivers/misc/sram-exec.c b/drivers/misc/sram-exec.c
new file mode 100644
index 000000000..b71dbbd73
--- /dev/null
+++ b/drivers/misc/sram-exec.c
@@ -0,0 +1,121 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * SRAM protect-exec region helper functions
+ *
+ * Copyright (C) 2017 Texas Instruments Incorporated - https://www.ti.com/
+ *	Dave Gerlach
+ */
+
+#include <linux/device.h>
+#include <linux/genalloc.h>
+#include <linux/mm.h>
+#include <linux/sram.h>
+#include <linux/set_memory.h>
+
+#include <asm/fncpy.h>
+
+#include "sram.h"
+
+static DEFINE_MUTEX(exec_pool_list_mutex);
+static LIST_HEAD(exec_pool_list);
+
+int sram_check_protect_exec(struct sram_dev *sram, struct sram_reserve *block,
+			    struct sram_partition *part)
+{
+	unsigned long base = (unsigned long)part->base;
+	unsigned long end = base + block->size;
+
+	if (!PAGE_ALIGNED(base) || !PAGE_ALIGNED(end)) {
+		dev_err(sram->dev,
+			"SRAM pool marked with 'protect-exec' is not page aligned and will not be created.\n");
+		return -ENOMEM;
+	}
+
+	return 0;
+}
+
+int sram_add_protect_exec(struct sram_partition *part)
+{
+	mutex_lock(&exec_pool_list_mutex);
+	list_add_tail(&part->list, &exec_pool_list);
+	mutex_unlock(&exec_pool_list_mutex);
+
+	return 0;
+}
+
+/**
+ * sram_exec_copy - copy data to a protected executable region of sram
+ *
+ * @pool: struct gen_pool retrieved that is part of this sram
+ * @dst: Destination address for the copy, that must be inside pool
+ * @src: Source address for the data to copy
+ * @size: Size of copy to perform, which starting from dst, must reside in pool
+ *
+ * Return: Address for copied data that can safely be called through function
+ *	   pointer, or NULL if problem.
+ *
+ * This helper function allows sram driver to act as central control location
+ * of 'protect-exec' pools which are normal sram pools but are always set
+ * read-only and executable except when copying data to them, at which point
+ * they are set to read-write non-executable, to make sure no memory is
+ * writeable and executable at the same time. This region must be page-aligned
+ * and is checked during probe, otherwise page attribute manipulation would
+ * not be possible. Care must be taken to only call the returned address as
+ * dst address is not guaranteed to be safely callable.
+ *
+ * NOTE: This function uses the fncpy macro to move code to the executable
+ * region. Some architectures have strict requirements for relocating
+ * executable code, so fncpy is a macro that must be defined by any arch
+ * making use of this functionality that guarantees a safe copy of exec
+ * data and returns a safe address that can be called as a C function
+ * pointer.
+ */
+void *sram_exec_copy(struct gen_pool *pool, void *dst, void *src,
+		     size_t size)
+{
+	struct sram_partition *part = NULL, *p;
+	unsigned long base;
+	int pages;
+	void *dst_cpy;
+	int ret;
+
+	mutex_lock(&exec_pool_list_mutex);
+	list_for_each_entry(p, &exec_pool_list, list) {
+		if (p->pool == pool)
+			part = p;
+	}
+	mutex_unlock(&exec_pool_list_mutex);
+
+	if (!part)
+		return NULL;
+
+	if (!gen_pool_has_addr(pool, (unsigned long)dst, size))
+		return NULL;
+
+	base = (unsigned long)part->base;
+	pages = PAGE_ALIGN(size) / PAGE_SIZE;
+
+	mutex_lock(&part->lock);
+
+	ret = set_memory_nx((unsigned long)base, pages);
+	if (ret)
+		goto error_out;
+	ret = set_memory_rw((unsigned long)base, pages);
+	if (ret)
+		goto error_out;
+
+	dst_cpy = fncpy(dst, src, size);
+
+	ret = set_memory_rox((unsigned long)base, pages);
+	if (ret)
+		goto error_out;
+
+	mutex_unlock(&part->lock);
+
+	return dst_cpy;
+
+error_out:
+	mutex_unlock(&part->lock);
+	return NULL;
+}
+EXPORT_SYMBOL_GPL(sram_exec_copy);