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, 191 insertions, 0 deletions

diff --git a/crypto/xctr.c b/crypto/xctr.c
new file mode 100644
index 000000000..607ab82cb
--- /dev/null
+++ b/crypto/xctr.c
@@ -0,0 +1,191 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * XCTR: XOR Counter mode - Adapted from ctr.c
+ *
+ * (C) Copyright IBM Corp. 2007 - Joy Latten <latten@us.ibm.com>
+ * Copyright 2021 Google LLC
+ */
+
+/*
+ * XCTR mode is a blockcipher mode of operation used to implement HCTR2. XCTR is
+ * closely related to the CTR mode of operation; the main difference is that CTR
+ * generates the keystream using E(CTR + IV) whereas XCTR generates the
+ * keystream using E(CTR ^ IV). This allows implementations to avoid dealing
+ * with multi-limb integers (as is required in CTR mode). XCTR is also specified
+ * using little-endian arithmetic which makes it slightly faster on LE machines.
+ *
+ * See the HCTR2 paper for more details:
+ *	Length-preserving encryption with HCTR2
+ *      (https://eprint.iacr.org/2021/1441.pdf)
+ */
+
+#include <crypto/algapi.h>
+#include <crypto/internal/cipher.h>
+#include <crypto/internal/skcipher.h>
+#include <linux/err.h>
+#include <linux/init.h>
+#include <linux/kernel.h>
+#include <linux/module.h>
+#include <linux/slab.h>
+
+/* For now this implementation is limited to 16-byte blocks for simplicity */
+#define XCTR_BLOCKSIZE 16
+
+static void crypto_xctr_crypt_final(struct skcipher_walk *walk,
+				   struct crypto_cipher *tfm, u32 byte_ctr)
+{
+	u8 keystream[XCTR_BLOCKSIZE];
+	const u8 *src = walk->src.virt.addr;
+	u8 *dst = walk->dst.virt.addr;
+	unsigned int nbytes = walk->nbytes;
+	__le32 ctr32 = cpu_to_le32(byte_ctr / XCTR_BLOCKSIZE + 1);
+
+	crypto_xor(walk->iv, (u8 *)&ctr32, sizeof(ctr32));
+	crypto_cipher_encrypt_one(tfm, keystream, walk->iv);
+	crypto_xor_cpy(dst, keystream, src, nbytes);
+	crypto_xor(walk->iv, (u8 *)&ctr32, sizeof(ctr32));
+}
+
+static int crypto_xctr_crypt_segment(struct skcipher_walk *walk,
+				    struct crypto_cipher *tfm, u32 byte_ctr)
+{
+	void (*fn)(struct crypto_tfm *, u8 *, const u8 *) =
+		   crypto_cipher_alg(tfm)->cia_encrypt;
+	const u8 *src = walk->src.virt.addr;
+	u8 *dst = walk->dst.virt.addr;
+	unsigned int nbytes = walk->nbytes;
+	__le32 ctr32 = cpu_to_le32(byte_ctr / XCTR_BLOCKSIZE + 1);
+
+	do {
+		crypto_xor(walk->iv, (u8 *)&ctr32, sizeof(ctr32));
+		fn(crypto_cipher_tfm(tfm), dst, walk->iv);
+		crypto_xor(dst, src, XCTR_BLOCKSIZE);
+		crypto_xor(walk->iv, (u8 *)&ctr32, sizeof(ctr32));
+
+		le32_add_cpu(&ctr32, 1);
+
+		src += XCTR_BLOCKSIZE;
+		dst += XCTR_BLOCKSIZE;
+	} while ((nbytes -= XCTR_BLOCKSIZE) >= XCTR_BLOCKSIZE);
+
+	return nbytes;
+}
+
+static int crypto_xctr_crypt_inplace(struct skcipher_walk *walk,
+				    struct crypto_cipher *tfm, u32 byte_ctr)
+{
+	void (*fn)(struct crypto_tfm *, u8 *, const u8 *) =
+		   crypto_cipher_alg(tfm)->cia_encrypt;
+	unsigned long alignmask = crypto_cipher_alignmask(tfm);
+	unsigned int nbytes = walk->nbytes;
+	u8 *data = walk->dst.virt.addr;
+	u8 tmp[XCTR_BLOCKSIZE + MAX_CIPHER_ALIGNMASK];
+	u8 *keystream = PTR_ALIGN(tmp + 0, alignmask + 1);
+	__le32 ctr32 = cpu_to_le32(byte_ctr / XCTR_BLOCKSIZE + 1);
+
+	do {
+		crypto_xor(walk->iv, (u8 *)&ctr32, sizeof(ctr32));
+		fn(crypto_cipher_tfm(tfm), keystream, walk->iv);
+		crypto_xor(data, keystream, XCTR_BLOCKSIZE);
+		crypto_xor(walk->iv, (u8 *)&ctr32, sizeof(ctr32));
+
+		le32_add_cpu(&ctr32, 1);
+
+		data += XCTR_BLOCKSIZE;
+	} while ((nbytes -= XCTR_BLOCKSIZE) >= XCTR_BLOCKSIZE);
+
+	return nbytes;
+}
+
+static int crypto_xctr_crypt(struct skcipher_request *req)
+{
+	struct crypto_skcipher *tfm = crypto_skcipher_reqtfm(req);
+	struct crypto_cipher *cipher = skcipher_cipher_simple(tfm);
+	struct skcipher_walk walk;
+	unsigned int nbytes;
+	int err;
+	u32 byte_ctr = 0;
+
+	err = skcipher_walk_virt(&walk, req, false);
+
+	while (walk.nbytes >= XCTR_BLOCKSIZE) {
+		if (walk.src.virt.addr == walk.dst.virt.addr)
+			nbytes = crypto_xctr_crypt_inplace(&walk, cipher,
+							   byte_ctr);
+		else
+			nbytes = crypto_xctr_crypt_segment(&walk, cipher,
+							   byte_ctr);
+
+		byte_ctr += walk.nbytes - nbytes;
+		err = skcipher_walk_done(&walk, nbytes);
+	}
+
+	if (walk.nbytes) {
+		crypto_xctr_crypt_final(&walk, cipher, byte_ctr);
+		err = skcipher_walk_done(&walk, 0);
+	}
+
+	return err;
+}
+
+static int crypto_xctr_create(struct crypto_template *tmpl, struct rtattr **tb)
+{
+	struct skcipher_instance *inst;
+	struct crypto_alg *alg;
+	int err;
+
+	inst = skcipher_alloc_instance_simple(tmpl, tb);
+	if (IS_ERR(inst))
+		return PTR_ERR(inst);
+
+	alg = skcipher_ialg_simple(inst);
+
+	/* Block size must be 16 bytes. */
+	err = -EINVAL;
+	if (alg->cra_blocksize != XCTR_BLOCKSIZE)
+		goto out_free_inst;
+
+	/* XCTR mode is a stream cipher. */
+	inst->alg.base.cra_blocksize = 1;
+
+	/*
+	 * To simplify the implementation, configure the skcipher walk to only
+	 * give a partial block at the very end, never earlier.
+	 */
+	inst->alg.chunksize = alg->cra_blocksize;
+
+	inst->alg.encrypt = crypto_xctr_crypt;
+	inst->alg.decrypt = crypto_xctr_crypt;
+
+	err = skcipher_register_instance(tmpl, inst);
+	if (err) {
+out_free_inst:
+		inst->free(inst);
+	}
+
+	return err;
+}
+
+static struct crypto_template crypto_xctr_tmpl = {
+	.name = "xctr",
+	.create = crypto_xctr_create,
+	.module = THIS_MODULE,
+};
+
+static int __init crypto_xctr_module_init(void)
+{
+	return crypto_register_template(&crypto_xctr_tmpl);
+}
+
+static void __exit crypto_xctr_module_exit(void)
+{
+	crypto_unregister_template(&crypto_xctr_tmpl);
+}
+
+module_init(crypto_xctr_module_init);
+module_exit(crypto_xctr_module_exit);
+
+MODULE_LICENSE("GPL");
+MODULE_DESCRIPTION("XCTR block cipher mode of operation");
+MODULE_ALIAS_CRYPTO("xctr");
+MODULE_IMPORT_NS("CRYPTO_INTERNAL");