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, 210 insertions, 0 deletions

diff --git a/include/crypto/md5.h b/include/crypto/md5.h
new file mode 100644
index 000000000..c47aedfe6
--- /dev/null
+++ b/include/crypto/md5.h
@@ -0,0 +1,210 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef _CRYPTO_MD5_H
+#define _CRYPTO_MD5_H
+
+#include <crypto/hash.h>
+#include <linux/types.h>
+
+#define MD5_DIGEST_SIZE		16
+#define MD5_HMAC_BLOCK_SIZE	64
+#define MD5_BLOCK_SIZE		64
+#define MD5_BLOCK_WORDS		16
+#define MD5_HASH_WORDS		4
+#define MD5_STATE_SIZE		24
+
+#define MD5_H0	0x67452301UL
+#define MD5_H1	0xefcdab89UL
+#define MD5_H2	0x98badcfeUL
+#define MD5_H3	0x10325476UL
+
+#define CRYPTO_MD5_STATESIZE \
+	CRYPTO_HASH_STATESIZE(MD5_STATE_SIZE, MD5_HMAC_BLOCK_SIZE)
+
+extern const u8 md5_zero_message_hash[MD5_DIGEST_SIZE];
+
+struct md5_state {
+	u32 hash[MD5_HASH_WORDS];
+	u64 byte_count;
+	u32 block[MD5_BLOCK_WORDS];
+};
+
+/* State for the MD5 compression function */
+struct md5_block_state {
+	u32 h[MD5_HASH_WORDS];
+};
+
+/**
+ * struct md5_ctx - Context for hashing a message with MD5
+ * @state: the compression function state
+ * @bytecount: number of bytes processed so far
+ * @buf: partial block buffer; bytecount % MD5_BLOCK_SIZE bytes are valid
+ */
+struct md5_ctx {
+	struct md5_block_state state;
+	u64 bytecount;
+	u8 buf[MD5_BLOCK_SIZE] __aligned(__alignof__(__le64));
+};
+
+/**
+ * md5_init() - Initialize an MD5 context for a new message
+ * @ctx: the context to initialize
+ *
+ * If you don't need incremental computation, consider md5() instead.
+ *
+ * Context: Any context.
+ */
+void md5_init(struct md5_ctx *ctx);
+
+/**
+ * md5_update() - Update an MD5 context with message data
+ * @ctx: the context to update; must have been initialized
+ * @data: the message data
+ * @len: the data length in bytes
+ *
+ * This can be called any number of times.
+ *
+ * Context: Any context.
+ */
+void md5_update(struct md5_ctx *ctx, const u8 *data, size_t len);
+
+/**
+ * md5_final() - Finish computing an MD5 message digest
+ * @ctx: the context to finalize; must have been initialized
+ * @out: (output) the resulting MD5 message digest
+ *
+ * After finishing, this zeroizes @ctx.  So the caller does not need to do it.
+ *
+ * Context: Any context.
+ */
+void md5_final(struct md5_ctx *ctx, u8 out[at_least MD5_DIGEST_SIZE]);
+
+/**
+ * md5() - Compute MD5 message digest in one shot
+ * @data: the message data
+ * @len: the data length in bytes
+ * @out: (output) the resulting MD5 message digest
+ *
+ * Context: Any context.
+ */
+void md5(const u8 *data, size_t len, u8 out[at_least MD5_DIGEST_SIZE]);
+
+/**
+ * struct hmac_md5_key - Prepared key for HMAC-MD5
+ * @istate: private
+ * @ostate: private
+ */
+struct hmac_md5_key {
+	struct md5_block_state istate;
+	struct md5_block_state ostate;
+};
+
+/**
+ * struct hmac_md5_ctx - Context for computing HMAC-MD5 of a message
+ * @hash_ctx: private
+ * @ostate: private
+ */
+struct hmac_md5_ctx {
+	struct md5_ctx hash_ctx;
+	struct md5_block_state ostate;
+};
+
+/**
+ * hmac_md5_preparekey() - Prepare a key for HMAC-MD5
+ * @key: (output) the key structure to initialize
+ * @raw_key: the raw HMAC-MD5 key
+ * @raw_key_len: the key length in bytes.  All key lengths are supported.
+ *
+ * Note: the caller is responsible for zeroizing both the struct hmac_md5_key
+ * and the raw key once they are no longer needed.
+ *
+ * Context: Any context.
+ */
+void hmac_md5_preparekey(struct hmac_md5_key *key,
+			 const u8 *raw_key, size_t raw_key_len);
+
+/**
+ * hmac_md5_init() - Initialize an HMAC-MD5 context for a new message
+ * @ctx: (output) the HMAC context to initialize
+ * @key: the prepared HMAC key
+ *
+ * If you don't need incremental computation, consider hmac_md5() instead.
+ *
+ * Context: Any context.
+ */
+void hmac_md5_init(struct hmac_md5_ctx *ctx, const struct hmac_md5_key *key);
+
+/**
+ * hmac_md5_init_usingrawkey() - Initialize an HMAC-MD5 context for a new
+ *				  message, using a raw key
+ * @ctx: (output) the HMAC context to initialize
+ * @raw_key: the raw HMAC-MD5 key
+ * @raw_key_len: the key length in bytes.  All key lengths are supported.
+ *
+ * If you don't need incremental computation, consider hmac_md5_usingrawkey()
+ * instead.
+ *
+ * Context: Any context.
+ */
+void hmac_md5_init_usingrawkey(struct hmac_md5_ctx *ctx,
+			       const u8 *raw_key, size_t raw_key_len);
+
+/**
+ * hmac_md5_update() - Update an HMAC-MD5 context with message data
+ * @ctx: the HMAC context to update; must have been initialized
+ * @data: the message data
+ * @data_len: the data length in bytes
+ *
+ * This can be called any number of times.
+ *
+ * Context: Any context.
+ */
+static inline void hmac_md5_update(struct hmac_md5_ctx *ctx,
+				   const u8 *data, size_t data_len)
+{
+	md5_update(&ctx->hash_ctx, data, data_len);
+}
+
+/**
+ * hmac_md5_final() - Finish computing an HMAC-MD5 value
+ * @ctx: the HMAC context to finalize; must have been initialized
+ * @out: (output) the resulting HMAC-MD5 value
+ *
+ * After finishing, this zeroizes @ctx.  So the caller does not need to do it.
+ *
+ * Context: Any context.
+ */
+void hmac_md5_final(struct hmac_md5_ctx *ctx, u8 out[at_least MD5_DIGEST_SIZE]);
+
+/**
+ * hmac_md5() - Compute HMAC-MD5 in one shot, using a prepared key
+ * @key: the prepared HMAC key
+ * @data: the message data
+ * @data_len: the data length in bytes
+ * @out: (output) the resulting HMAC-MD5 value
+ *
+ * If you're using the key only once, consider using hmac_md5_usingrawkey().
+ *
+ * Context: Any context.
+ */
+void hmac_md5(const struct hmac_md5_key *key,
+	      const u8 *data, size_t data_len,
+	      u8 out[at_least MD5_DIGEST_SIZE]);
+
+/**
+ * hmac_md5_usingrawkey() - Compute HMAC-MD5 in one shot, using a raw key
+ * @raw_key: the raw HMAC-MD5 key
+ * @raw_key_len: the key length in bytes.  All key lengths are supported.
+ * @data: the message data
+ * @data_len: the data length in bytes
+ * @out: (output) the resulting HMAC-MD5 value
+ *
+ * If you're using the key multiple times, prefer to use hmac_md5_preparekey()
+ * followed by multiple calls to hmac_md5() instead.
+ *
+ * Context: Any context.
+ */
+void hmac_md5_usingrawkey(const u8 *raw_key, size_t raw_key_len,
+			  const u8 *data, size_t data_len,
+			  u8 out[at_least MD5_DIGEST_SIZE]);
+
+#endif /* _CRYPTO_MD5_H */