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, 102 insertions, 0 deletions

diff --git a/include/crypto/chacha.h b/include/crypto/chacha.h
new file mode 100644
index 000000000..1cc301a48
--- /dev/null
+++ b/include/crypto/chacha.h
@@ -0,0 +1,102 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Common values and helper functions for the ChaCha and XChaCha stream ciphers.
+ *
+ * XChaCha extends ChaCha's nonce to 192 bits, while provably retaining ChaCha's
+ * security.  Here they share the same key size, tfm context, and setkey
+ * function; only their IV size and encrypt/decrypt function differ.
+ *
+ * The ChaCha paper specifies 20, 12, and 8-round variants.  In general, it is
+ * recommended to use the 20-round variant ChaCha20.  However, the other
+ * variants can be needed in some performance-sensitive scenarios.  The generic
+ * ChaCha code currently allows only the 20 and 12-round variants.
+ */
+
+#ifndef _CRYPTO_CHACHA_H
+#define _CRYPTO_CHACHA_H
+
+#include <linux/unaligned.h>
+#include <linux/string.h>
+#include <linux/types.h>
+
+/* 32-bit stream position, then 96-bit nonce (RFC7539 convention) */
+#define CHACHA_IV_SIZE		16
+
+#define CHACHA_KEY_SIZE		32
+#define CHACHA_BLOCK_SIZE	64
+#define CHACHAPOLY_IV_SIZE	12
+
+#define CHACHA_KEY_WORDS	8
+#define CHACHA_STATE_WORDS	16
+#define HCHACHA_OUT_WORDS	8
+
+/* 192-bit nonce, then 64-bit stream position */
+#define XCHACHA_IV_SIZE		32
+
+struct chacha_state {
+	u32 x[CHACHA_STATE_WORDS];
+};
+
+void chacha_block_generic(struct chacha_state *state,
+			  u8 out[at_least CHACHA_BLOCK_SIZE], int nrounds);
+static inline void chacha20_block(struct chacha_state *state,
+				  u8 out[at_least CHACHA_BLOCK_SIZE])
+{
+	chacha_block_generic(state, out, 20);
+}
+
+void hchacha_block_generic(const struct chacha_state *state,
+			   u32 out[at_least HCHACHA_OUT_WORDS], int nrounds);
+
+void hchacha_block(const struct chacha_state *state,
+		   u32 out[at_least HCHACHA_OUT_WORDS], int nrounds);
+
+enum chacha_constants { /* expand 32-byte k */
+	CHACHA_CONSTANT_EXPA = 0x61707865U,
+	CHACHA_CONSTANT_ND_3 = 0x3320646eU,
+	CHACHA_CONSTANT_2_BY = 0x79622d32U,
+	CHACHA_CONSTANT_TE_K = 0x6b206574U
+};
+
+static inline void chacha_init_consts(struct chacha_state *state)
+{
+	state->x[0]  = CHACHA_CONSTANT_EXPA;
+	state->x[1]  = CHACHA_CONSTANT_ND_3;
+	state->x[2]  = CHACHA_CONSTANT_2_BY;
+	state->x[3]  = CHACHA_CONSTANT_TE_K;
+}
+
+static inline void chacha_init(struct chacha_state *state,
+			       const u32 key[at_least CHACHA_KEY_WORDS],
+			       const u8 iv[at_least CHACHA_IV_SIZE])
+{
+	chacha_init_consts(state);
+	state->x[4]  = key[0];
+	state->x[5]  = key[1];
+	state->x[6]  = key[2];
+	state->x[7]  = key[3];
+	state->x[8]  = key[4];
+	state->x[9]  = key[5];
+	state->x[10] = key[6];
+	state->x[11] = key[7];
+	state->x[12] = get_unaligned_le32(iv +  0);
+	state->x[13] = get_unaligned_le32(iv +  4);
+	state->x[14] = get_unaligned_le32(iv +  8);
+	state->x[15] = get_unaligned_le32(iv + 12);
+}
+
+void chacha_crypt(struct chacha_state *state, u8 *dst, const u8 *src,
+		  unsigned int bytes, int nrounds);
+
+static inline void chacha20_crypt(struct chacha_state *state,
+				  u8 *dst, const u8 *src, unsigned int bytes)
+{
+	chacha_crypt(state, dst, src, bytes, 20);
+}
+
+static inline void chacha_zeroize_state(struct chacha_state *state)
+{
+	memzero_explicit(state, sizeof(*state));
+}
+
+#endif /* _CRYPTO_CHACHA_H */