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, 351 insertions, 0 deletions

diff --git a/crypto/rmd160.c b/crypto/rmd160.c
new file mode 100644
index 000000000..9860b60c9
--- /dev/null
+++ b/crypto/rmd160.c
@@ -0,0 +1,351 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Cryptographic API.
+ *
+ * RIPEMD-160 - RACE Integrity Primitives Evaluation Message Digest.
+ *
+ * Based on the reference implementation by Antoon Bosselaers, ESAT-COSIC
+ *
+ * Copyright (c) 2008 Adrian-Ken Rueegsegger <ken@codelabs.ch>
+ */
+#include <crypto/internal/hash.h>
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/string.h>
+#include "ripemd.h"
+
+struct rmd160_ctx {
+	u64 byte_count;
+	u32 state[5];
+};
+
+#define K1  RMD_K1
+#define K2  RMD_K2
+#define K3  RMD_K3
+#define K4  RMD_K4
+#define K5  RMD_K5
+#define KK1 RMD_K6
+#define KK2 RMD_K7
+#define KK3 RMD_K8
+#define KK4 RMD_K9
+#define KK5 RMD_K1
+
+#define F1(x, y, z) (x ^ y ^ z)		/* XOR */
+#define F2(x, y, z) (z ^ (x & (y ^ z)))	/* x ? y : z */
+#define F3(x, y, z) ((x | ~y) ^ z)
+#define F4(x, y, z) (y ^ (z & (x ^ y)))	/* z ? x : y */
+#define F5(x, y, z) (x ^ (y | ~z))
+
+#define ROUND(a, b, c, d, e, f, k, x, s)  { \
+	(a) += f((b), (c), (d)) + le32_to_cpup(&(x)) + (k); \
+	(a) = rol32((a), (s)) + (e); \
+	(c) = rol32((c), 10); \
+}
+
+static void rmd160_transform(u32 *state, const __le32 *in)
+{
+	u32 aa, bb, cc, dd, ee, aaa, bbb, ccc, ddd, eee;
+
+	/* Initialize left lane */
+	aa = state[0];
+	bb = state[1];
+	cc = state[2];
+	dd = state[3];
+	ee = state[4];
+
+	/* Initialize right lane */
+	aaa = state[0];
+	bbb = state[1];
+	ccc = state[2];
+	ddd = state[3];
+	eee = state[4];
+
+	/* round 1: left lane */
+	ROUND(aa, bb, cc, dd, ee, F1, K1, in[0],  11);
+	ROUND(ee, aa, bb, cc, dd, F1, K1, in[1],  14);
+	ROUND(dd, ee, aa, bb, cc, F1, K1, in[2],  15);
+	ROUND(cc, dd, ee, aa, bb, F1, K1, in[3],  12);
+	ROUND(bb, cc, dd, ee, aa, F1, K1, in[4],   5);
+	ROUND(aa, bb, cc, dd, ee, F1, K1, in[5],   8);
+	ROUND(ee, aa, bb, cc, dd, F1, K1, in[6],   7);
+	ROUND(dd, ee, aa, bb, cc, F1, K1, in[7],   9);
+	ROUND(cc, dd, ee, aa, bb, F1, K1, in[8],  11);
+	ROUND(bb, cc, dd, ee, aa, F1, K1, in[9],  13);
+	ROUND(aa, bb, cc, dd, ee, F1, K1, in[10], 14);
+	ROUND(ee, aa, bb, cc, dd, F1, K1, in[11], 15);
+	ROUND(dd, ee, aa, bb, cc, F1, K1, in[12],  6);
+	ROUND(cc, dd, ee, aa, bb, F1, K1, in[13],  7);
+	ROUND(bb, cc, dd, ee, aa, F1, K1, in[14],  9);
+	ROUND(aa, bb, cc, dd, ee, F1, K1, in[15],  8);
+
+	/* round 2: left lane" */
+	ROUND(ee, aa, bb, cc, dd, F2, K2, in[7],   7);
+	ROUND(dd, ee, aa, bb, cc, F2, K2, in[4],   6);
+	ROUND(cc, dd, ee, aa, bb, F2, K2, in[13],  8);
+	ROUND(bb, cc, dd, ee, aa, F2, K2, in[1],  13);
+	ROUND(aa, bb, cc, dd, ee, F2, K2, in[10], 11);
+	ROUND(ee, aa, bb, cc, dd, F2, K2, in[6],   9);
+	ROUND(dd, ee, aa, bb, cc, F2, K2, in[15],  7);
+	ROUND(cc, dd, ee, aa, bb, F2, K2, in[3],  15);
+	ROUND(bb, cc, dd, ee, aa, F2, K2, in[12],  7);
+	ROUND(aa, bb, cc, dd, ee, F2, K2, in[0],  12);
+	ROUND(ee, aa, bb, cc, dd, F2, K2, in[9],  15);
+	ROUND(dd, ee, aa, bb, cc, F2, K2, in[5],   9);
+	ROUND(cc, dd, ee, aa, bb, F2, K2, in[2],  11);
+	ROUND(bb, cc, dd, ee, aa, F2, K2, in[14],  7);
+	ROUND(aa, bb, cc, dd, ee, F2, K2, in[11], 13);
+	ROUND(ee, aa, bb, cc, dd, F2, K2, in[8],  12);
+
+	/* round 3: left lane" */
+	ROUND(dd, ee, aa, bb, cc, F3, K3, in[3],  11);
+	ROUND(cc, dd, ee, aa, bb, F3, K3, in[10], 13);
+	ROUND(bb, cc, dd, ee, aa, F3, K3, in[14],  6);
+	ROUND(aa, bb, cc, dd, ee, F3, K3, in[4],   7);
+	ROUND(ee, aa, bb, cc, dd, F3, K3, in[9],  14);
+	ROUND(dd, ee, aa, bb, cc, F3, K3, in[15],  9);
+	ROUND(cc, dd, ee, aa, bb, F3, K3, in[8],  13);
+	ROUND(bb, cc, dd, ee, aa, F3, K3, in[1],  15);
+	ROUND(aa, bb, cc, dd, ee, F3, K3, in[2],  14);
+	ROUND(ee, aa, bb, cc, dd, F3, K3, in[7],   8);
+	ROUND(dd, ee, aa, bb, cc, F3, K3, in[0],  13);
+	ROUND(cc, dd, ee, aa, bb, F3, K3, in[6],   6);
+	ROUND(bb, cc, dd, ee, aa, F3, K3, in[13],  5);
+	ROUND(aa, bb, cc, dd, ee, F3, K3, in[11], 12);
+	ROUND(ee, aa, bb, cc, dd, F3, K3, in[5],   7);
+	ROUND(dd, ee, aa, bb, cc, F3, K3, in[12],  5);
+
+	/* round 4: left lane" */
+	ROUND(cc, dd, ee, aa, bb, F4, K4, in[1],  11);
+	ROUND(bb, cc, dd, ee, aa, F4, K4, in[9],  12);
+	ROUND(aa, bb, cc, dd, ee, F4, K4, in[11], 14);
+	ROUND(ee, aa, bb, cc, dd, F4, K4, in[10], 15);
+	ROUND(dd, ee, aa, bb, cc, F4, K4, in[0],  14);
+	ROUND(cc, dd, ee, aa, bb, F4, K4, in[8],  15);
+	ROUND(bb, cc, dd, ee, aa, F4, K4, in[12],  9);
+	ROUND(aa, bb, cc, dd, ee, F4, K4, in[4],   8);
+	ROUND(ee, aa, bb, cc, dd, F4, K4, in[13],  9);
+	ROUND(dd, ee, aa, bb, cc, F4, K4, in[3],  14);
+	ROUND(cc, dd, ee, aa, bb, F4, K4, in[7],   5);
+	ROUND(bb, cc, dd, ee, aa, F4, K4, in[15],  6);
+	ROUND(aa, bb, cc, dd, ee, F4, K4, in[14],  8);
+	ROUND(ee, aa, bb, cc, dd, F4, K4, in[5],   6);
+	ROUND(dd, ee, aa, bb, cc, F4, K4, in[6],   5);
+	ROUND(cc, dd, ee, aa, bb, F4, K4, in[2],  12);
+
+	/* round 5: left lane" */
+	ROUND(bb, cc, dd, ee, aa, F5, K5, in[4],   9);
+	ROUND(aa, bb, cc, dd, ee, F5, K5, in[0],  15);
+	ROUND(ee, aa, bb, cc, dd, F5, K5, in[5],   5);
+	ROUND(dd, ee, aa, bb, cc, F5, K5, in[9],  11);
+	ROUND(cc, dd, ee, aa, bb, F5, K5, in[7],   6);
+	ROUND(bb, cc, dd, ee, aa, F5, K5, in[12],  8);
+	ROUND(aa, bb, cc, dd, ee, F5, K5, in[2],  13);
+	ROUND(ee, aa, bb, cc, dd, F5, K5, in[10], 12);
+	ROUND(dd, ee, aa, bb, cc, F5, K5, in[14],  5);
+	ROUND(cc, dd, ee, aa, bb, F5, K5, in[1],  12);
+	ROUND(bb, cc, dd, ee, aa, F5, K5, in[3],  13);
+	ROUND(aa, bb, cc, dd, ee, F5, K5, in[8],  14);
+	ROUND(ee, aa, bb, cc, dd, F5, K5, in[11], 11);
+	ROUND(dd, ee, aa, bb, cc, F5, K5, in[6],   8);
+	ROUND(cc, dd, ee, aa, bb, F5, K5, in[15],  5);
+	ROUND(bb, cc, dd, ee, aa, F5, K5, in[13],  6);
+
+	/* round 1: right lane */
+	ROUND(aaa, bbb, ccc, ddd, eee, F5, KK1, in[5],   8);
+	ROUND(eee, aaa, bbb, ccc, ddd, F5, KK1, in[14],  9);
+	ROUND(ddd, eee, aaa, bbb, ccc, F5, KK1, in[7],   9);
+	ROUND(ccc, ddd, eee, aaa, bbb, F5, KK1, in[0],  11);
+	ROUND(bbb, ccc, ddd, eee, aaa, F5, KK1, in[9],  13);
+	ROUND(aaa, bbb, ccc, ddd, eee, F5, KK1, in[2],  15);
+	ROUND(eee, aaa, bbb, ccc, ddd, F5, KK1, in[11], 15);
+	ROUND(ddd, eee, aaa, bbb, ccc, F5, KK1, in[4],   5);
+	ROUND(ccc, ddd, eee, aaa, bbb, F5, KK1, in[13],  7);
+	ROUND(bbb, ccc, ddd, eee, aaa, F5, KK1, in[6],   7);
+	ROUND(aaa, bbb, ccc, ddd, eee, F5, KK1, in[15],  8);
+	ROUND(eee, aaa, bbb, ccc, ddd, F5, KK1, in[8],  11);
+	ROUND(ddd, eee, aaa, bbb, ccc, F5, KK1, in[1],  14);
+	ROUND(ccc, ddd, eee, aaa, bbb, F5, KK1, in[10], 14);
+	ROUND(bbb, ccc, ddd, eee, aaa, F5, KK1, in[3],  12);
+	ROUND(aaa, bbb, ccc, ddd, eee, F5, KK1, in[12],  6);
+
+	/* round 2: right lane */
+	ROUND(eee, aaa, bbb, ccc, ddd, F4, KK2, in[6],   9);
+	ROUND(ddd, eee, aaa, bbb, ccc, F4, KK2, in[11], 13);
+	ROUND(ccc, ddd, eee, aaa, bbb, F4, KK2, in[3],  15);
+	ROUND(bbb, ccc, ddd, eee, aaa, F4, KK2, in[7],   7);
+	ROUND(aaa, bbb, ccc, ddd, eee, F4, KK2, in[0],  12);
+	ROUND(eee, aaa, bbb, ccc, ddd, F4, KK2, in[13],  8);
+	ROUND(ddd, eee, aaa, bbb, ccc, F4, KK2, in[5],   9);
+	ROUND(ccc, ddd, eee, aaa, bbb, F4, KK2, in[10], 11);
+	ROUND(bbb, ccc, ddd, eee, aaa, F4, KK2, in[14],  7);
+	ROUND(aaa, bbb, ccc, ddd, eee, F4, KK2, in[15],  7);
+	ROUND(eee, aaa, bbb, ccc, ddd, F4, KK2, in[8],  12);
+	ROUND(ddd, eee, aaa, bbb, ccc, F4, KK2, in[12],  7);
+	ROUND(ccc, ddd, eee, aaa, bbb, F4, KK2, in[4],   6);
+	ROUND(bbb, ccc, ddd, eee, aaa, F4, KK2, in[9],  15);
+	ROUND(aaa, bbb, ccc, ddd, eee, F4, KK2, in[1],  13);
+	ROUND(eee, aaa, bbb, ccc, ddd, F4, KK2, in[2],  11);
+
+	/* round 3: right lane */
+	ROUND(ddd, eee, aaa, bbb, ccc, F3, KK3, in[15],  9);
+	ROUND(ccc, ddd, eee, aaa, bbb, F3, KK3, in[5],   7);
+	ROUND(bbb, ccc, ddd, eee, aaa, F3, KK3, in[1],  15);
+	ROUND(aaa, bbb, ccc, ddd, eee, F3, KK3, in[3],  11);
+	ROUND(eee, aaa, bbb, ccc, ddd, F3, KK3, in[7],   8);
+	ROUND(ddd, eee, aaa, bbb, ccc, F3, KK3, in[14],  6);
+	ROUND(ccc, ddd, eee, aaa, bbb, F3, KK3, in[6],   6);
+	ROUND(bbb, ccc, ddd, eee, aaa, F3, KK3, in[9],  14);
+	ROUND(aaa, bbb, ccc, ddd, eee, F3, KK3, in[11], 12);
+	ROUND(eee, aaa, bbb, ccc, ddd, F3, KK3, in[8],  13);
+	ROUND(ddd, eee, aaa, bbb, ccc, F3, KK3, in[12],  5);
+	ROUND(ccc, ddd, eee, aaa, bbb, F3, KK3, in[2],  14);
+	ROUND(bbb, ccc, ddd, eee, aaa, F3, KK3, in[10], 13);
+	ROUND(aaa, bbb, ccc, ddd, eee, F3, KK3, in[0],  13);
+	ROUND(eee, aaa, bbb, ccc, ddd, F3, KK3, in[4],   7);
+	ROUND(ddd, eee, aaa, bbb, ccc, F3, KK3, in[13],  5);
+
+	/* round 4: right lane */
+	ROUND(ccc, ddd, eee, aaa, bbb, F2, KK4, in[8],  15);
+	ROUND(bbb, ccc, ddd, eee, aaa, F2, KK4, in[6],   5);
+	ROUND(aaa, bbb, ccc, ddd, eee, F2, KK4, in[4],   8);
+	ROUND(eee, aaa, bbb, ccc, ddd, F2, KK4, in[1],  11);
+	ROUND(ddd, eee, aaa, bbb, ccc, F2, KK4, in[3],  14);
+	ROUND(ccc, ddd, eee, aaa, bbb, F2, KK4, in[11], 14);
+	ROUND(bbb, ccc, ddd, eee, aaa, F2, KK4, in[15],  6);
+	ROUND(aaa, bbb, ccc, ddd, eee, F2, KK4, in[0],  14);
+	ROUND(eee, aaa, bbb, ccc, ddd, F2, KK4, in[5],   6);
+	ROUND(ddd, eee, aaa, bbb, ccc, F2, KK4, in[12],  9);
+	ROUND(ccc, ddd, eee, aaa, bbb, F2, KK4, in[2],  12);
+	ROUND(bbb, ccc, ddd, eee, aaa, F2, KK4, in[13],  9);
+	ROUND(aaa, bbb, ccc, ddd, eee, F2, KK4, in[9],  12);
+	ROUND(eee, aaa, bbb, ccc, ddd, F2, KK4, in[7],   5);
+	ROUND(ddd, eee, aaa, bbb, ccc, F2, KK4, in[10], 15);
+	ROUND(ccc, ddd, eee, aaa, bbb, F2, KK4, in[14],  8);
+
+	/* round 5: right lane */
+	ROUND(bbb, ccc, ddd, eee, aaa, F1, KK5, in[12],  8);
+	ROUND(aaa, bbb, ccc, ddd, eee, F1, KK5, in[15],  5);
+	ROUND(eee, aaa, bbb, ccc, ddd, F1, KK5, in[10], 12);
+	ROUND(ddd, eee, aaa, bbb, ccc, F1, KK5, in[4],   9);
+	ROUND(ccc, ddd, eee, aaa, bbb, F1, KK5, in[1],  12);
+	ROUND(bbb, ccc, ddd, eee, aaa, F1, KK5, in[5],   5);
+	ROUND(aaa, bbb, ccc, ddd, eee, F1, KK5, in[8],  14);
+	ROUND(eee, aaa, bbb, ccc, ddd, F1, KK5, in[7],   6);
+	ROUND(ddd, eee, aaa, bbb, ccc, F1, KK5, in[6],   8);
+	ROUND(ccc, ddd, eee, aaa, bbb, F1, KK5, in[2],  13);
+	ROUND(bbb, ccc, ddd, eee, aaa, F1, KK5, in[13],  6);
+	ROUND(aaa, bbb, ccc, ddd, eee, F1, KK5, in[14],  5);
+	ROUND(eee, aaa, bbb, ccc, ddd, F1, KK5, in[0],  15);
+	ROUND(ddd, eee, aaa, bbb, ccc, F1, KK5, in[3],  13);
+	ROUND(ccc, ddd, eee, aaa, bbb, F1, KK5, in[9],  11);
+	ROUND(bbb, ccc, ddd, eee, aaa, F1, KK5, in[11], 11);
+
+	/* combine results */
+	ddd += cc + state[1];		/* final result for state[0] */
+	state[1] = state[2] + dd + eee;
+	state[2] = state[3] + ee + aaa;
+	state[3] = state[4] + aa + bbb;
+	state[4] = state[0] + bb + ccc;
+	state[0] = ddd;
+}
+
+static int rmd160_init(struct shash_desc *desc)
+{
+	struct rmd160_ctx *rctx = shash_desc_ctx(desc);
+
+	rctx->byte_count = 0;
+
+	rctx->state[0] = RMD_H0;
+	rctx->state[1] = RMD_H1;
+	rctx->state[2] = RMD_H2;
+	rctx->state[3] = RMD_H3;
+	rctx->state[4] = RMD_H4;
+
+	return 0;
+}
+
+static int rmd160_update(struct shash_desc *desc, const u8 *data,
+			 unsigned int len)
+{
+	int remain = len - round_down(len, RMD160_BLOCK_SIZE);
+	struct rmd160_ctx *rctx = shash_desc_ctx(desc);
+	__le32 buffer[RMD160_BLOCK_SIZE / 4];
+
+	rctx->byte_count += len - remain;
+
+	do {
+		memcpy(buffer, data, sizeof(buffer));
+		rmd160_transform(rctx->state, buffer);
+		data += sizeof(buffer);
+		len -= sizeof(buffer);
+	} while (len >= sizeof(buffer));
+
+	memzero_explicit(buffer, sizeof(buffer));
+	return remain;
+}
+
+/* Add padding and return the message digest. */
+static int rmd160_finup(struct shash_desc *desc, const u8 *src,
+			unsigned int len, u8 *out)
+{
+	unsigned int bit_offset = RMD160_BLOCK_SIZE / 8 - 1;
+	struct rmd160_ctx *rctx = shash_desc_ctx(desc);
+	union {
+		__le64 l64[RMD160_BLOCK_SIZE / 4];
+		__le32 l32[RMD160_BLOCK_SIZE / 2];
+		u8 u8[RMD160_BLOCK_SIZE * 2];
+	} block = {};
+	__le32 *dst = (__le32 *)out;
+	u32 i;
+
+	rctx->byte_count += len;
+	if (len >= bit_offset * 8)
+		bit_offset += RMD160_BLOCK_SIZE / 8;
+	memcpy(&block, src, len);
+	block.u8[len] = 0x80;
+	block.l64[bit_offset] = cpu_to_le64(rctx->byte_count << 3);
+
+	rmd160_transform(rctx->state, block.l32);
+	if (bit_offset > RMD160_BLOCK_SIZE / 8)
+		rmd160_transform(rctx->state,
+				 block.l32 + RMD160_BLOCK_SIZE / 4);
+	memzero_explicit(&block, sizeof(block));
+
+	/* Store state in digest */
+	for (i = 0; i < 5; i++)
+		dst[i] = cpu_to_le32p(&rctx->state[i]);
+	return 0;
+}
+
+static struct shash_alg alg = {
+	.digestsize	=	RMD160_DIGEST_SIZE,
+	.init		=	rmd160_init,
+	.update		=	rmd160_update,
+	.finup		=	rmd160_finup,
+	.descsize	=	sizeof(struct rmd160_ctx),
+	.base		=	{
+		.cra_name	 =	"rmd160",
+		.cra_driver_name =	"rmd160-generic",
+		.cra_flags	 =	CRYPTO_AHASH_ALG_BLOCK_ONLY,
+		.cra_blocksize	 =	RMD160_BLOCK_SIZE,
+		.cra_module	 =	THIS_MODULE,
+	}
+};
+
+static int __init rmd160_mod_init(void)
+{
+	return crypto_register_shash(&alg);
+}
+
+static void __exit rmd160_mod_fini(void)
+{
+	crypto_unregister_shash(&alg);
+}
+
+module_init(rmd160_mod_init);
+module_exit(rmd160_mod_fini);
+
+MODULE_LICENSE("GPL");
+MODULE_AUTHOR("Adrian-Ken Rueegsegger <ken@codelabs.ch>");
+MODULE_DESCRIPTION("RIPEMD-160 Message Digest");
+MODULE_ALIAS_CRYPTO("rmd160");