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, 85 insertions, 0 deletions

diff --git a/block/partitions/sgi.c b/block/partitions/sgi.c
new file mode 100644
index 000000000..2383ca63c
--- /dev/null
+++ b/block/partitions/sgi.c
@@ -0,0 +1,85 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ *  fs/partitions/sgi.c
+ *
+ *  Code extracted from drivers/block/genhd.c
+ */
+
+#include "check.h"
+
+#define SGI_LABEL_MAGIC 0x0be5a941
+
+enum {
+	LINUX_RAID_PARTITION = 0xfd,	/* autodetect RAID partition */
+};
+
+struct sgi_disklabel {
+	__be32 magic_mushroom;		/* Big fat spliff... */
+	__be16 root_part_num;		/* Root partition number */
+	__be16 swap_part_num;		/* Swap partition number */
+	s8 boot_file[16];		/* Name of boot file for ARCS */
+	u8 _unused0[48];		/* Device parameter useless crapola.. */
+	struct sgi_volume {
+		s8 name[8];		/* Name of volume */
+		__be32 block_num;		/* Logical block number */
+		__be32 num_bytes;		/* How big, in bytes */
+	} volume[15];
+	struct sgi_partition {
+		__be32 num_blocks;		/* Size in logical blocks */
+		__be32 first_block;	/* First logical block */
+		__be32 type;		/* Type of this partition */
+	} partitions[16];
+	__be32 csum;			/* Disk label checksum */
+	__be32 _unused1;			/* Padding */
+};
+
+int sgi_partition(struct parsed_partitions *state)
+{
+	int i, csum;
+	__be32 magic;
+	int slot = 1;
+	unsigned int start, blocks;
+	__be32 *ui, cs;
+	Sector sect;
+	struct sgi_disklabel *label;
+	struct sgi_partition *p;
+
+	label = read_part_sector(state, 0, &sect);
+	if (!label)
+		return -1;
+	p = &label->partitions[0];
+	magic = label->magic_mushroom;
+	if(be32_to_cpu(magic) != SGI_LABEL_MAGIC) {
+		put_dev_sector(sect);
+		return 0;
+	}
+	ui = ((__be32 *) (label + 1)) - 1;
+	for(csum = 0; ui >= ((__be32 *) label);) {
+		cs = *ui--;
+		csum += be32_to_cpu(cs);
+	}
+	if(csum) {
+		printk(KERN_WARNING "Dev %s SGI disklabel: csum bad, label corrupted\n",
+		       state->disk->disk_name);
+		put_dev_sector(sect);
+		return 0;
+	}
+	/* All SGI disk labels have 16 partitions, disks under Linux only
+	 * have 15 minor's.  Luckily there are always a few zero length
+	 * partitions which we don't care about so we never overflow the
+	 * current_minor.
+	 */
+	for(i = 0; i < 16; i++, p++) {
+		blocks = be32_to_cpu(p->num_blocks);
+		start  = be32_to_cpu(p->first_block);
+		if (blocks) {
+			put_partition(state, slot, start, blocks);
+			if (be32_to_cpu(p->type) == LINUX_RAID_PARTITION)
+				state->parts[slot].flags = ADDPART_FLAG_RAID;
+		}
+		slot++;
+	}
+	seq_buf_puts(&state->pp_buf, "\n");
+	put_dev_sector(sect);
+	return 1;
+}