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, 395 insertions, 0 deletions

diff --git a/drivers/crypto/padlock-sha.c b/drivers/crypto/padlock-sha.c
new file mode 100644
index 000000000..9214bbfc8
--- /dev/null
+++ b/drivers/crypto/padlock-sha.c
@@ -0,0 +1,395 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Cryptographic API.
+ *
+ * Support for VIA PadLock hardware crypto engine.
+ *
+ * Copyright (c) 2006  Michal Ludvig <michal@logix.cz>
+ */
+
+#include <asm/cpu_device_id.h>
+#include <crypto/internal/hash.h>
+#include <crypto/padlock.h>
+#include <crypto/sha1.h>
+#include <crypto/sha2.h>
+#include <linux/cpufeature.h>
+#include <linux/err.h>
+#include <linux/kernel.h>
+#include <linux/module.h>
+
+#define PADLOCK_SHA_DESCSIZE (128 + ((PADLOCK_ALIGNMENT - 1) & \
+				     ~(CRYPTO_MINALIGN - 1)))
+
+struct padlock_sha_ctx {
+	struct crypto_ahash *fallback;
+};
+
+static inline void *padlock_shash_desc_ctx(struct shash_desc *desc)
+{
+	return PTR_ALIGN(shash_desc_ctx(desc), PADLOCK_ALIGNMENT);
+}
+
+static int padlock_sha1_init(struct shash_desc *desc)
+{
+	struct sha1_state *sctx = padlock_shash_desc_ctx(desc);
+
+	*sctx = (struct sha1_state){
+		.state = { SHA1_H0, SHA1_H1, SHA1_H2, SHA1_H3, SHA1_H4 },
+	};
+
+	return 0;
+}
+
+static int padlock_sha256_init(struct shash_desc *desc)
+{
+	struct crypto_sha256_state *sctx = padlock_shash_desc_ctx(desc);
+
+	sha256_block_init(sctx);
+	return 0;
+}
+
+static int padlock_sha_update(struct shash_desc *desc,
+			      const u8 *data, unsigned int length)
+{
+	u8 *state = padlock_shash_desc_ctx(desc);
+	struct crypto_shash *tfm = desc->tfm;
+	int err, remain;
+
+	remain = length - round_down(length, crypto_shash_blocksize(tfm));
+	{
+		struct padlock_sha_ctx *ctx = crypto_shash_ctx(tfm);
+		HASH_REQUEST_ON_STACK(req, ctx->fallback);
+
+		ahash_request_set_callback(req, 0, NULL, NULL);
+		ahash_request_set_virt(req, data, NULL, length - remain);
+		err = crypto_ahash_import_core(req, state) ?:
+		      crypto_ahash_update(req) ?:
+		      crypto_ahash_export_core(req, state);
+		HASH_REQUEST_ZERO(req);
+	}
+
+	return err ?: remain;
+}
+
+static int padlock_sha_export(struct shash_desc *desc, void *out)
+{
+	memcpy(out, padlock_shash_desc_ctx(desc),
+	       crypto_shash_coresize(desc->tfm));
+	return 0;
+}
+
+static int padlock_sha_import(struct shash_desc *desc, const void *in)
+{
+	unsigned int bs = crypto_shash_blocksize(desc->tfm);
+	unsigned int ss = crypto_shash_coresize(desc->tfm);
+	u64 *state = padlock_shash_desc_ctx(desc);
+
+	memcpy(state, in, ss);
+
+	/* Stop evil imports from generating a fault. */
+	state[ss / 8 - 1] &= ~(bs - 1);
+
+	return 0;
+}
+
+static inline void padlock_output_block(uint32_t *src,
+		 	uint32_t *dst, size_t count)
+{
+	while (count--)
+		*dst++ = swab32(*src++);
+}
+
+static int padlock_sha_finup(struct shash_desc *desc, const u8 *in,
+			     unsigned int count, u8 *out)
+{
+	struct padlock_sha_ctx *ctx = crypto_shash_ctx(desc->tfm);
+	HASH_REQUEST_ON_STACK(req, ctx->fallback);
+
+	ahash_request_set_callback(req, 0, NULL, NULL);
+	ahash_request_set_virt(req, in, out, count);
+	return crypto_ahash_import_core(req, padlock_shash_desc_ctx(desc)) ?:
+	       crypto_ahash_finup(req);
+}
+
+static int padlock_sha1_finup(struct shash_desc *desc, const u8 *in,
+			      unsigned int count, u8 *out)
+{
+	/* We can't store directly to *out as it may be unaligned. */
+	/* BTW Don't reduce the buffer size below 128 Bytes!
+	 *     PadLock microcode needs it that big. */
+	struct sha1_state *state = padlock_shash_desc_ctx(desc);
+	u64 start = state->count;
+
+	if (start + count > ULONG_MAX)
+		return padlock_sha_finup(desc, in, count, out);
+
+	asm volatile (".byte 0xf3,0x0f,0xa6,0xc8" /* rep xsha1 */
+		      : \
+		      : "c"((unsigned long)start + count), \
+			"a"((unsigned long)start), \
+			"S"(in), "D"(state));
+
+	padlock_output_block(state->state, (uint32_t *)out, 5);
+	return 0;
+}
+
+static int padlock_sha256_finup(struct shash_desc *desc, const u8 *in,
+				unsigned int count, u8 *out)
+{
+	/* We can't store directly to *out as it may be unaligned. */
+	/* BTW Don't reduce the buffer size below 128 Bytes!
+	 *     PadLock microcode needs it that big. */
+	struct sha256_state *state = padlock_shash_desc_ctx(desc);
+	u64 start = state->count;
+
+	if (start + count > ULONG_MAX)
+		return padlock_sha_finup(desc, in, count, out);
+
+	asm volatile (".byte 0xf3,0x0f,0xa6,0xd0" /* rep xsha256 */
+		      : \
+		      : "c"((unsigned long)start + count), \
+			"a"((unsigned long)start), \
+			"S"(in), "D"(state));
+
+	padlock_output_block(state->state, (uint32_t *)out, 8);
+	return 0;
+}
+
+static int padlock_init_tfm(struct crypto_shash *hash)
+{
+	const char *fallback_driver_name = crypto_shash_alg_name(hash);
+	struct padlock_sha_ctx *ctx = crypto_shash_ctx(hash);
+	struct crypto_ahash *fallback_tfm;
+
+	/* Allocate a fallback and abort if it failed. */
+	fallback_tfm = crypto_alloc_ahash(fallback_driver_name, 0,
+					  CRYPTO_ALG_NEED_FALLBACK |
+					  CRYPTO_ALG_ASYNC);
+	if (IS_ERR(fallback_tfm)) {
+		printk(KERN_WARNING PFX "Fallback driver '%s' could not be loaded!\n",
+		       fallback_driver_name);
+		return PTR_ERR(fallback_tfm);
+	}
+
+	if (crypto_shash_statesize(hash) !=
+	    crypto_ahash_statesize(fallback_tfm)) {
+		crypto_free_ahash(fallback_tfm);
+		return -EINVAL;
+	}
+
+	ctx->fallback = fallback_tfm;
+
+	return 0;
+}
+
+static void padlock_exit_tfm(struct crypto_shash *hash)
+{
+	struct padlock_sha_ctx *ctx = crypto_shash_ctx(hash);
+
+	crypto_free_ahash(ctx->fallback);
+}
+
+static struct shash_alg sha1_alg = {
+	.digestsize	=	SHA1_DIGEST_SIZE,
+	.init   	= 	padlock_sha1_init,
+	.update 	=	padlock_sha_update,
+	.finup  	=	padlock_sha1_finup,
+	.export		=	padlock_sha_export,
+	.import		=	padlock_sha_import,
+	.init_tfm	=	padlock_init_tfm,
+	.exit_tfm	=	padlock_exit_tfm,
+	.descsize	=	PADLOCK_SHA_DESCSIZE,
+	.statesize	=	SHA1_STATE_SIZE,
+	.base		=	{
+		.cra_name		=	"sha1",
+		.cra_driver_name	=	"sha1-padlock",
+		.cra_priority		=	PADLOCK_CRA_PRIORITY,
+		.cra_flags		=	CRYPTO_ALG_NEED_FALLBACK |
+						CRYPTO_AHASH_ALG_BLOCK_ONLY |
+						CRYPTO_AHASH_ALG_FINUP_MAX,
+		.cra_blocksize		=	SHA1_BLOCK_SIZE,
+		.cra_ctxsize		=	sizeof(struct padlock_sha_ctx),
+		.cra_module		=	THIS_MODULE,
+	}
+};
+
+static struct shash_alg sha256_alg = {
+	.digestsize	=	SHA256_DIGEST_SIZE,
+	.init   	= 	padlock_sha256_init,
+	.update 	=	padlock_sha_update,
+	.finup  	=	padlock_sha256_finup,
+	.init_tfm	=	padlock_init_tfm,
+	.export		=	padlock_sha_export,
+	.import		=	padlock_sha_import,
+	.exit_tfm	=	padlock_exit_tfm,
+	.descsize	=	PADLOCK_SHA_DESCSIZE,
+	.statesize	=	sizeof(struct crypto_sha256_state),
+	.base		=	{
+		.cra_name		=	"sha256",
+		.cra_driver_name	=	"sha256-padlock",
+		.cra_priority		=	PADLOCK_CRA_PRIORITY,
+		.cra_flags		=	CRYPTO_ALG_NEED_FALLBACK |
+						CRYPTO_AHASH_ALG_BLOCK_ONLY |
+						CRYPTO_AHASH_ALG_FINUP_MAX,
+		.cra_blocksize		=	SHA256_BLOCK_SIZE,
+		.cra_ctxsize		=	sizeof(struct padlock_sha_ctx),
+		.cra_module		=	THIS_MODULE,
+	}
+};
+
+/* Add two shash_alg instance for hardware-implemented *
+* multiple-parts hash supported by VIA Nano Processor.*/
+
+static int padlock_sha1_update_nano(struct shash_desc *desc,
+				    const u8 *src, unsigned int len)
+{
+	/*The PHE require the out buffer must 128 bytes and 16-bytes aligned*/
+	struct sha1_state *state = padlock_shash_desc_ctx(desc);
+	int blocks = len / SHA1_BLOCK_SIZE;
+
+	len -= blocks * SHA1_BLOCK_SIZE;
+	state->count += blocks * SHA1_BLOCK_SIZE;
+
+	/* Process the left bytes from the input data */
+	asm volatile (".byte 0xf3,0x0f,0xa6,0xc8"
+		      : "+S"(src), "+D"(state)
+		      : "a"((long)-1),
+			"c"((unsigned long)blocks));
+	return len;
+}
+
+static int padlock_sha256_update_nano(struct shash_desc *desc, const u8 *src,
+			  unsigned int len)
+{
+	/*The PHE require the out buffer must 128 bytes and 16-bytes aligned*/
+	struct crypto_sha256_state *state = padlock_shash_desc_ctx(desc);
+	int blocks = len / SHA256_BLOCK_SIZE;
+
+	len -= blocks * SHA256_BLOCK_SIZE;
+	state->count += blocks * SHA256_BLOCK_SIZE;
+
+	/* Process the left bytes from input data*/
+	asm volatile (".byte 0xf3,0x0f,0xa6,0xd0"
+		      : "+S"(src), "+D"(state)
+		      : "a"((long)-1),
+		      "c"((unsigned long)blocks));
+	return len;
+}
+
+static struct shash_alg sha1_alg_nano = {
+	.digestsize	=	SHA1_DIGEST_SIZE,
+	.init		=	padlock_sha1_init,
+	.update		=	padlock_sha1_update_nano,
+	.finup  	=	padlock_sha1_finup,
+	.export		=	padlock_sha_export,
+	.import		=	padlock_sha_import,
+	.descsize	=	PADLOCK_SHA_DESCSIZE,
+	.statesize	=	SHA1_STATE_SIZE,
+	.base		=	{
+		.cra_name		=	"sha1",
+		.cra_driver_name	=	"sha1-padlock-nano",
+		.cra_priority		=	PADLOCK_CRA_PRIORITY,
+		.cra_flags		=	CRYPTO_AHASH_ALG_BLOCK_ONLY |
+						CRYPTO_AHASH_ALG_FINUP_MAX,
+		.cra_blocksize		=	SHA1_BLOCK_SIZE,
+		.cra_module		=	THIS_MODULE,
+	}
+};
+
+static struct shash_alg sha256_alg_nano = {
+	.digestsize	=	SHA256_DIGEST_SIZE,
+	.init		=	padlock_sha256_init,
+	.update		=	padlock_sha256_update_nano,
+	.finup		=	padlock_sha256_finup,
+	.export		=	padlock_sha_export,
+	.import		=	padlock_sha_import,
+	.descsize	=	PADLOCK_SHA_DESCSIZE,
+	.statesize	=	sizeof(struct crypto_sha256_state),
+	.base		=	{
+		.cra_name		=	"sha256",
+		.cra_driver_name	=	"sha256-padlock-nano",
+		.cra_priority		=	PADLOCK_CRA_PRIORITY,
+		.cra_flags		=	CRYPTO_AHASH_ALG_BLOCK_ONLY |
+						CRYPTO_AHASH_ALG_FINUP_MAX,
+		.cra_blocksize		=	SHA256_BLOCK_SIZE,
+		.cra_module		=	THIS_MODULE,
+	}
+};
+
+static const struct x86_cpu_id padlock_sha_ids[] = {
+	X86_MATCH_FEATURE(X86_FEATURE_PHE, NULL),
+	{}
+};
+MODULE_DEVICE_TABLE(x86cpu, padlock_sha_ids);
+
+static int __init padlock_init(void)
+{
+	int rc = -ENODEV;
+	struct cpuinfo_x86 *c = &cpu_data(0);
+	struct shash_alg *sha1;
+	struct shash_alg *sha256;
+
+	if (!x86_match_cpu(padlock_sha_ids) || !boot_cpu_has(X86_FEATURE_PHE_EN))
+		return -ENODEV;
+
+	/*
+	 * Skip family 0x07 and newer used by Zhaoxin processors,
+	 * as the driver's self-tests fail on these CPUs.
+	 */
+	if (c->x86 >= 0x07)
+		return -ENODEV;
+
+	/* Register the newly added algorithm module if on *
+	* VIA Nano processor, or else just do as before */
+	if (c->x86_model < 0x0f) {
+		sha1 = &sha1_alg;
+		sha256 = &sha256_alg;
+	} else {
+		sha1 = &sha1_alg_nano;
+		sha256 = &sha256_alg_nano;
+	}
+
+	rc = crypto_register_shash(sha1);
+	if (rc)
+		goto out;
+
+	rc = crypto_register_shash(sha256);
+	if (rc)
+		goto out_unreg1;
+
+	printk(KERN_NOTICE PFX "Using VIA PadLock ACE for SHA1/SHA256 algorithms.\n");
+
+	return 0;
+
+out_unreg1:
+	crypto_unregister_shash(sha1);
+
+out:
+	printk(KERN_ERR PFX "VIA PadLock SHA1/SHA256 initialization failed.\n");
+	return rc;
+}
+
+static void __exit padlock_fini(void)
+{
+	struct cpuinfo_x86 *c = &cpu_data(0);
+
+	if (c->x86_model >= 0x0f) {
+		crypto_unregister_shash(&sha1_alg_nano);
+		crypto_unregister_shash(&sha256_alg_nano);
+	} else {
+		crypto_unregister_shash(&sha1_alg);
+		crypto_unregister_shash(&sha256_alg);
+	}
+}
+
+module_init(padlock_init);
+module_exit(padlock_fini);
+
+MODULE_DESCRIPTION("VIA PadLock SHA1/SHA256 algorithms support.");
+MODULE_LICENSE("GPL");
+MODULE_AUTHOR("Michal Ludvig");
+
+MODULE_ALIAS_CRYPTO("sha1-all");
+MODULE_ALIAS_CRYPTO("sha256-all");
+MODULE_ALIAS_CRYPTO("sha1-padlock");
+MODULE_ALIAS_CRYPTO("sha256-padlock");