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, 109 insertions, 0 deletions

diff --git a/lib/crypto/mpi/mpi-inline.h b/lib/crypto/mpi/mpi-inline.h
new file mode 100644
index 000000000..980b6b940
--- /dev/null
+++ b/lib/crypto/mpi/mpi-inline.h
@@ -0,0 +1,109 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+/* mpi-inline.h  -  Internal to the Multi Precision Integers
+ *	Copyright (C) 1994, 1996, 1998, 1999 Free Software Foundation, Inc.
+ *
+ * This file is part of GnuPG.
+ *
+ * Note: This code is heavily based on the GNU MP Library.
+ *	 Actually it's the same code with only minor changes in the
+ *	 way the data is stored; this is to support the abstraction
+ *	 of an optional secure memory allocation which may be used
+ *	 to avoid revealing of sensitive data due to paging etc.
+ *	 The GNU MP Library itself is published under the LGPL;
+ *	 however I decided to publish this code under the plain GPL.
+ */
+
+#ifndef G10_MPI_INLINE_H
+#define G10_MPI_INLINE_H
+
+#ifndef G10_MPI_INLINE_DECL
+#define G10_MPI_INLINE_DECL  static inline
+#endif
+
+G10_MPI_INLINE_DECL mpi_limb_t
+mpihelp_add_1(mpi_ptr_t res_ptr, mpi_ptr_t s1_ptr,
+	      mpi_size_t s1_size, mpi_limb_t s2_limb)
+{
+	mpi_limb_t x;
+
+	x = *s1_ptr++;
+	s2_limb += x;
+	*res_ptr++ = s2_limb;
+	if (s2_limb < x) {	/* sum is less than the left operand: handle carry */
+		while (--s1_size) {
+			x = *s1_ptr++ + 1;	/* add carry */
+			*res_ptr++ = x;	/* and store */
+			if (x)	/* not 0 (no overflow): we can stop */
+				goto leave;
+		}
+		return 1;	/* return carry (size of s1 to small) */
+	}
+
+leave:
+	if (res_ptr != s1_ptr) {	/* not the same variable */
+		mpi_size_t i;	/* copy the rest */
+		for (i = 0; i < s1_size - 1; i++)
+			res_ptr[i] = s1_ptr[i];
+	}
+	return 0;		/* no carry */
+}
+
+G10_MPI_INLINE_DECL mpi_limb_t
+mpihelp_add(mpi_ptr_t res_ptr, mpi_ptr_t s1_ptr, mpi_size_t s1_size,
+	    mpi_ptr_t s2_ptr, mpi_size_t s2_size)
+{
+	mpi_limb_t cy = 0;
+
+	if (s2_size)
+		cy = mpihelp_add_n(res_ptr, s1_ptr, s2_ptr, s2_size);
+
+	if (s1_size - s2_size)
+		cy = mpihelp_add_1(res_ptr + s2_size, s1_ptr + s2_size,
+				   s1_size - s2_size, cy);
+	return cy;
+}
+
+G10_MPI_INLINE_DECL mpi_limb_t
+mpihelp_sub_1(mpi_ptr_t res_ptr, mpi_ptr_t s1_ptr,
+	      mpi_size_t s1_size, mpi_limb_t s2_limb)
+{
+	mpi_limb_t x;
+
+	x = *s1_ptr++;
+	s2_limb = x - s2_limb;
+	*res_ptr++ = s2_limb;
+	if (s2_limb > x) {
+		while (--s1_size) {
+			x = *s1_ptr++;
+			*res_ptr++ = x - 1;
+			if (x)
+				goto leave;
+		}
+		return 1;
+	}
+
+leave:
+	if (res_ptr != s1_ptr) {
+		mpi_size_t i;
+		for (i = 0; i < s1_size - 1; i++)
+			res_ptr[i] = s1_ptr[i];
+	}
+	return 0;
+}
+
+G10_MPI_INLINE_DECL mpi_limb_t
+mpihelp_sub(mpi_ptr_t res_ptr, mpi_ptr_t s1_ptr, mpi_size_t s1_size,
+	    mpi_ptr_t s2_ptr, mpi_size_t s2_size)
+{
+	mpi_limb_t cy = 0;
+
+	if (s2_size)
+		cy = mpihelp_sub_n(res_ptr, s1_ptr, s2_ptr, s2_size);
+
+	if (s1_size - s2_size)
+		cy = mpihelp_sub_1(res_ptr + s2_size, s1_ptr + s2_size,
+				   s1_size - s2_size, cy);
+	return cy;
+}
+
+#endif /*G10_MPI_INLINE_H */