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, 312 insertions, 0 deletions

diff --git a/include/crypto/internal/ecc.h b/include/crypto/internal/ecc.h
new file mode 100644
index 000000000..d67fe13a5
--- /dev/null
+++ b/include/crypto/internal/ecc.h
@@ -0,0 +1,312 @@
+/*
+ * Copyright (c) 2013, Kenneth MacKay
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are
+ * met:
+ *  * Redistributions of source code must retain the above copyright
+ *   notice, this list of conditions and the following disclaimer.
+ *  * Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+#ifndef _CRYPTO_ECC_H
+#define _CRYPTO_ECC_H
+
+#include <crypto/ecc_curve.h>
+#include <linux/unaligned.h>
+
+/* One digit is u64 qword. */
+#define ECC_CURVE_NIST_P192_DIGITS  3
+#define ECC_CURVE_NIST_P256_DIGITS  4
+#define ECC_CURVE_NIST_P384_DIGITS  6
+#define ECC_CURVE_NIST_P521_DIGITS  9
+#define ECC_MAX_DIGITS              DIV_ROUND_UP(521, 64) /* NIST P521 */
+
+#define ECC_DIGITS_TO_BYTES_SHIFT 3
+
+#define ECC_MAX_BYTES (ECC_MAX_DIGITS << ECC_DIGITS_TO_BYTES_SHIFT)
+
+#define ECC_POINT_INIT(x, y, ndigits)	(struct ecc_point) { x, y, ndigits }
+
+/*
+ * The integers r and s making up the signature are expected to be
+ * formatted as two consecutive u64 arrays of size ECC_MAX_BYTES.
+ * The bytes within each u64 digit are in native endianness,
+ * but the order of the u64 digits themselves is little endian.
+ * This format allows direct use by internal vli_*() functions.
+ */
+struct ecdsa_raw_sig {
+	u64 r[ECC_MAX_DIGITS];
+	u64 s[ECC_MAX_DIGITS];
+};
+
+/**
+ * ecc_swap_digits() - Copy ndigits from big endian array to native array
+ * @in:       Input array
+ * @out:      Output array
+ * @ndigits:  Number of digits to copy
+ */
+static inline void ecc_swap_digits(const void *in, u64 *out, unsigned int ndigits)
+{
+	const __be64 *src = (__force __be64 *)in;
+	int i;
+
+	for (i = 0; i < ndigits; i++)
+		out[i] = get_unaligned_be64(&src[ndigits - 1 - i]);
+}
+
+/**
+ * ecc_digits_from_bytes() - Create ndigits-sized digits array from byte array
+ * @in:       Input byte array
+ * @nbytes:   Size of input byte array
+ * @out:      Output digits array
+ * @ndigits:  Number of digits to create from byte array
+ *
+ * The first byte in the input byte array is expected to hold the most
+ * significant bits of the large integer.
+ */
+void ecc_digits_from_bytes(const u8 *in, unsigned int nbytes,
+			   u64 *out, unsigned int ndigits);
+
+/**
+ * ecc_is_key_valid() - Validate a given ECDH private key
+ *
+ * @curve_id:		id representing the curve to use
+ * @ndigits:		curve's number of digits
+ * @private_key:	private key to be used for the given curve
+ * @private_key_len:	private key length
+ *
+ * Returns: 0 if the key is acceptable, a negative value otherwise
+ */
+int ecc_is_key_valid(unsigned int curve_id, unsigned int ndigits,
+		     const u64 *private_key, unsigned int private_key_len);
+
+/**
+ * ecc_gen_privkey() -  Generates an ECC private key.
+ * The private key is a random integer in the range 0 < random < n, where n is a
+ * prime that is the order of the cyclic subgroup generated by the distinguished
+ * point G.
+ * @curve_id:		id representing the curve to use
+ * @ndigits:		curve number of digits
+ * @private_key:	buffer for storing the generated private key
+ *
+ * Returns: 0 if the private key was generated successfully, a negative value
+ * if an error occurred.
+ */
+int ecc_gen_privkey(unsigned int curve_id, unsigned int ndigits,
+		    u64 *private_key);
+
+/**
+ * ecc_make_pub_key() - Compute an ECC public key
+ *
+ * @curve_id:		id representing the curve to use
+ * @ndigits:		curve's number of digits
+ * @private_key:	pregenerated private key for the given curve
+ * @public_key:		buffer for storing the generated public key
+ *
+ * Returns: 0 if the public key was generated successfully, a negative value
+ * if an error occurred.
+ */
+int ecc_make_pub_key(const unsigned int curve_id, unsigned int ndigits,
+		     const u64 *private_key, u64 *public_key);
+
+/**
+ * crypto_ecdh_shared_secret() - Compute a shared secret
+ *
+ * @curve_id:		id representing the curve to use
+ * @ndigits:		curve's number of digits
+ * @private_key:	private key of part A
+ * @public_key:		public key of counterpart B
+ * @secret:		buffer for storing the calculated shared secret
+ *
+ * Note: It is recommended that you hash the result of crypto_ecdh_shared_secret
+ * before using it for symmetric encryption or HMAC.
+ *
+ * Returns: 0 if the shared secret was generated successfully, a negative value
+ * if an error occurred.
+ */
+int crypto_ecdh_shared_secret(unsigned int curve_id, unsigned int ndigits,
+			      const u64 *private_key, const u64 *public_key,
+			      u64 *secret);
+
+/**
+ * ecc_is_pubkey_valid_partial() - Partial public key validation
+ *
+ * @curve:		elliptic curve domain parameters
+ * @pk:			public key as a point
+ *
+ * Validate public key according to SP800-56A section 5.6.2.3.4 ECC Partial
+ * Public-Key Validation Routine.
+ *
+ * Note: There is no check that the public key is in the correct elliptic curve
+ * subgroup.
+ *
+ * Return: 0 if validation is successful, -EINVAL if validation is failed.
+ */
+int ecc_is_pubkey_valid_partial(const struct ecc_curve *curve,
+				struct ecc_point *pk);
+
+/**
+ * ecc_is_pubkey_valid_full() - Full public key validation
+ *
+ * @curve:		elliptic curve domain parameters
+ * @pk:			public key as a point
+ *
+ * Validate public key according to SP800-56A section 5.6.2.3.3 ECC Full
+ * Public-Key Validation Routine.
+ *
+ * Return: 0 if validation is successful, -EINVAL if validation is failed.
+ */
+int ecc_is_pubkey_valid_full(const struct ecc_curve *curve,
+			     struct ecc_point *pk);
+
+/**
+ * vli_is_zero() - Determine is vli is zero
+ *
+ * @vli:		vli to check.
+ * @ndigits:		length of the @vli
+ *
+ * Returns: %true if vli == 0, %false otherwise.
+ */
+bool vli_is_zero(const u64 *vli, unsigned int ndigits);
+
+/**
+ * vli_cmp() - compare left and right vlis
+ *
+ * @left:		vli
+ * @right:		vli
+ * @ndigits:		length of both vlis
+ *
+ * Returns: sign of @left - @right, i.e. -1 if @left < @right,
+ * 0 if @left == @right, 1 if @left > @right.
+ */
+int vli_cmp(const u64 *left, const u64 *right, unsigned int ndigits);
+
+/**
+ * vli_sub() - Subtracts right from left
+ *
+ * @result:		where to write result
+ * @left:		vli
+ * @right:		vli
+ * @ndigits:		length of all vlis
+ *
+ * Note: can modify in-place.
+ *
+ * Return: carry bit.
+ */
+u64 vli_sub(u64 *result, const u64 *left, const u64 *right,
+	    unsigned int ndigits);
+
+/**
+ * vli_from_be64() - Load vli from big-endian u64 array
+ *
+ * @dest:		destination vli
+ * @src:		source array of u64 BE values
+ * @ndigits:		length of both vli and array
+ */
+void vli_from_be64(u64 *dest, const void *src, unsigned int ndigits);
+
+/**
+ * vli_from_le64() - Load vli from little-endian u64 array
+ *
+ * @dest:		destination vli
+ * @src:		source array of u64 LE values
+ * @ndigits:		length of both vli and array
+ */
+void vli_from_le64(u64 *dest, const void *src, unsigned int ndigits);
+
+/**
+ * vli_mod_inv() - Modular inversion
+ *
+ * @result:		where to write vli number
+ * @input:		vli value to operate on
+ * @mod:		modulus
+ * @ndigits:		length of all vlis
+ */
+void vli_mod_inv(u64 *result, const u64 *input, const u64 *mod,
+		 unsigned int ndigits);
+
+/**
+ * vli_mod_mult_slow() - Modular multiplication
+ *
+ * @result:		where to write result value
+ * @left:		vli number to multiply with @right
+ * @right:		vli number to multiply with @left
+ * @mod:		modulus
+ * @ndigits:		length of all vlis
+ *
+ * Note: Assumes that mod is big enough curve order.
+ */
+void vli_mod_mult_slow(u64 *result, const u64 *left, const u64 *right,
+		       const u64 *mod, unsigned int ndigits);
+
+/**
+ * vli_num_bits() - Counts the number of bits required for vli.
+ *
+ * @vli:		vli to check.
+ * @ndigits:		Length of the @vli
+ *
+ * Return: The number of bits required to represent @vli.
+ */
+unsigned int vli_num_bits(const u64 *vli, unsigned int ndigits);
+
+/**
+ * ecc_alloc_point() - Allocate ECC point.
+ *
+ * @ndigits:		Length of vlis in u64 qwords.
+ *
+ * Return: Pointer to the allocated point or NULL if allocation failed.
+ */
+struct ecc_point *ecc_alloc_point(unsigned int ndigits);
+
+/**
+ * ecc_free_point() - Free ECC point.
+ *
+ * @p:			The point to free.
+ */
+void ecc_free_point(struct ecc_point *p);
+
+/**
+ * ecc_point_is_zero() - Check if point is zero.
+ *
+ * @point:		Point to check for zero.
+ *
+ * Return: true if point is the point at infinity, false otherwise.
+ */
+bool ecc_point_is_zero(const struct ecc_point *point);
+
+/**
+ * ecc_point_mult_shamir() - Add two points multiplied by scalars
+ *
+ * @result:		resulting point
+ * @x:			scalar to multiply with @p
+ * @p:			point to multiply with @x
+ * @y:			scalar to multiply with @q
+ * @q:			point to multiply with @y
+ * @curve:		curve
+ *
+ * Returns result = x * p + x * q over the curve.
+ * This works faster than two multiplications and addition.
+ */
+void ecc_point_mult_shamir(const struct ecc_point *result,
+			   const u64 *x, const struct ecc_point *p,
+			   const u64 *y, const struct ecc_point *q,
+			   const struct ecc_curve *curve);
+
+extern struct crypto_template ecdsa_x962_tmpl;
+extern struct crypto_template ecdsa_p1363_tmpl;
+#endif