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, 187 insertions, 0 deletions

diff --git a/drivers/mtd/devices/mtdram.c b/drivers/mtd/devices/mtdram.c
new file mode 100644
index 000000000..3fff3cdd9
--- /dev/null
+++ b/drivers/mtd/devices/mtdram.c
@@ -0,0 +1,187 @@
+/*
+ * mtdram - a test mtd device
+ * Author: Alexander Larsson <alex@cendio.se>
+ *
+ * Copyright (c) 1999 Alexander Larsson <alex@cendio.se>
+ * Copyright (c) 2005 Joern Engel <joern@wh.fh-wedel.de>
+ *
+ * This code is GPL
+ *
+ */
+
+#include <linux/module.h>
+#include <linux/slab.h>
+#include <linux/ioport.h>
+#include <linux/vmalloc.h>
+#include <linux/mm.h>
+#include <linux/init.h>
+#include <linux/mtd/mtd.h>
+#include <linux/mtd/mtdram.h>
+
+static unsigned long total_size = CONFIG_MTDRAM_TOTAL_SIZE;
+static unsigned long erase_size = CONFIG_MTDRAM_ERASE_SIZE;
+static unsigned long writebuf_size = 64;
+#define MTDRAM_TOTAL_SIZE (total_size * 1024)
+#define MTDRAM_ERASE_SIZE (erase_size * 1024)
+
+module_param(total_size, ulong, 0);
+MODULE_PARM_DESC(total_size, "Total device size in KiB");
+module_param(erase_size, ulong, 0);
+MODULE_PARM_DESC(erase_size, "Device erase block size in KiB");
+module_param(writebuf_size, ulong, 0);
+MODULE_PARM_DESC(writebuf_size, "Device write buf size in Bytes (Default: 64)");
+
+// We could store these in the mtd structure, but we only support 1 device..
+static struct mtd_info *mtd_info;
+
+static int check_offs_len(struct mtd_info *mtd, loff_t ofs, uint64_t len)
+{
+	int ret = 0;
+
+	/* Start address must align on block boundary */
+	if (mtd_mod_by_eb(ofs, mtd)) {
+		pr_debug("%s: unaligned address\n", __func__);
+		ret = -EINVAL;
+	}
+
+	/* Length must align on block boundary */
+	if (mtd_mod_by_eb(len, mtd)) {
+		pr_debug("%s: length not block aligned\n", __func__);
+		ret = -EINVAL;
+	}
+
+	return ret;
+}
+
+static int ram_erase(struct mtd_info *mtd, struct erase_info *instr)
+{
+	if (check_offs_len(mtd, instr->addr, instr->len))
+		return -EINVAL;
+	memset((char *)mtd->priv + instr->addr, 0xff, instr->len);
+
+	return 0;
+}
+
+static int ram_point(struct mtd_info *mtd, loff_t from, size_t len,
+		size_t *retlen, void **virt, resource_size_t *phys)
+{
+	*virt = mtd->priv + from;
+	*retlen = len;
+
+	if (phys) {
+		/* limit retlen to the number of contiguous physical pages */
+		unsigned long page_ofs = offset_in_page(*virt);
+		void *addr = *virt - page_ofs;
+		unsigned long pfn1, pfn0 = vmalloc_to_pfn(addr);
+
+		*phys = __pfn_to_phys(pfn0) + page_ofs;
+		len += page_ofs;
+		while (len > PAGE_SIZE) {
+			len -= PAGE_SIZE;
+			addr += PAGE_SIZE;
+			pfn0++;
+			pfn1 = vmalloc_to_pfn(addr);
+			if (pfn1 != pfn0) {
+				*retlen = addr - *virt;
+				break;
+			}
+		}
+	}
+
+	return 0;
+}
+
+static int ram_unpoint(struct mtd_info *mtd, loff_t from, size_t len)
+{
+	return 0;
+}
+
+static int ram_read(struct mtd_info *mtd, loff_t from, size_t len,
+		size_t *retlen, u_char *buf)
+{
+	memcpy(buf, mtd->priv + from, len);
+	*retlen = len;
+	return 0;
+}
+
+static int ram_write(struct mtd_info *mtd, loff_t to, size_t len,
+		size_t *retlen, const u_char *buf)
+{
+	memcpy((char *)mtd->priv + to, buf, len);
+	*retlen = len;
+	return 0;
+}
+
+static void __exit cleanup_mtdram(void)
+{
+	if (mtd_info) {
+		mtd_device_unregister(mtd_info);
+		vfree(mtd_info->priv);
+		kfree(mtd_info);
+	}
+}
+
+int mtdram_init_device(struct mtd_info *mtd, void *mapped_address,
+		unsigned long size, const char *name)
+{
+	memset(mtd, 0, sizeof(*mtd));
+
+	/* Setup the MTD structure */
+	mtd->name = name;
+	mtd->type = MTD_RAM;
+	mtd->flags = MTD_CAP_RAM;
+	mtd->size = size;
+	mtd->writesize = 1;
+	mtd->writebufsize = writebuf_size;
+	mtd->erasesize = MTDRAM_ERASE_SIZE;
+	mtd->priv = mapped_address;
+
+	mtd->owner = THIS_MODULE;
+	mtd->_erase = ram_erase;
+	mtd->_point = ram_point;
+	mtd->_unpoint = ram_unpoint;
+	mtd->_read = ram_read;
+	mtd->_write = ram_write;
+
+	if (mtd_device_register(mtd, NULL, 0))
+		return -EIO;
+
+	return 0;
+}
+
+static int __init init_mtdram(void)
+{
+	void *addr;
+	int err;
+
+	if (!total_size)
+		return -EINVAL;
+
+	/* Allocate some memory */
+	mtd_info = kmalloc_obj(struct mtd_info);
+	if (!mtd_info)
+		return -ENOMEM;
+
+	addr = vmalloc(MTDRAM_TOTAL_SIZE);
+	if (!addr) {
+		kfree(mtd_info);
+		mtd_info = NULL;
+		return -ENOMEM;
+	}
+	err = mtdram_init_device(mtd_info, addr, MTDRAM_TOTAL_SIZE, "mtdram test device");
+	if (err) {
+		vfree(addr);
+		kfree(mtd_info);
+		mtd_info = NULL;
+		return err;
+	}
+	memset(mtd_info->priv, 0xff, MTDRAM_TOTAL_SIZE);
+	return err;
+}
+
+module_init(init_mtdram);
+module_exit(cleanup_mtdram);
+
+MODULE_LICENSE("GPL");
+MODULE_AUTHOR("Alexander Larsson <alexl@redhat.com>");
+MODULE_DESCRIPTION("Simulated MTD driver for testing");