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, 225 insertions, 0 deletions

diff --git a/drivers/crypto/omap-crypto.c b/drivers/crypto/omap-crypto.c
new file mode 100644
index 000000000..7bd6869c2
--- /dev/null
+++ b/drivers/crypto/omap-crypto.c
@@ -0,0 +1,225 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * OMAP Crypto driver common support routines.
+ *
+ * Copyright (c) 2017 Texas Instruments Incorporated
+ *   Tero Kristo <t-kristo@ti.com>
+ */
+
+#include <linux/module.h>
+#include <linux/kernel.h>
+#include <linux/scatterlist.h>
+#include <crypto/scatterwalk.h>
+
+#include "omap-crypto.h"
+
+static int omap_crypto_copy_sg_lists(int total, int bs,
+				     struct scatterlist **sg,
+				     struct scatterlist *new_sg, u16 flags)
+{
+	int n = sg_nents(*sg);
+	struct scatterlist *tmp;
+
+	if (!(flags & OMAP_CRYPTO_FORCE_SINGLE_ENTRY)) {
+		new_sg = kmalloc_objs(*new_sg, n);
+		if (!new_sg)
+			return -ENOMEM;
+
+		sg_init_table(new_sg, n);
+	}
+
+	tmp = new_sg;
+
+	while (*sg && total) {
+		int len = (*sg)->length;
+
+		if (total < len)
+			len = total;
+
+		if (len > 0) {
+			total -= len;
+			sg_set_page(tmp, sg_page(*sg), len, (*sg)->offset);
+			if (total <= 0)
+				sg_mark_end(tmp);
+			tmp = sg_next(tmp);
+		}
+
+		*sg = sg_next(*sg);
+	}
+
+	*sg = new_sg;
+
+	return 0;
+}
+
+static int omap_crypto_copy_sgs(int total, int bs, struct scatterlist **sg,
+				struct scatterlist *new_sg, u16 flags)
+{
+	void *buf;
+	int pages;
+	int new_len;
+
+	new_len = ALIGN(total, bs);
+	pages = get_order(new_len);
+
+	buf = (void *)__get_free_pages(GFP_ATOMIC, pages);
+	if (!buf) {
+		pr_err("%s: Couldn't allocate pages for unaligned cases.\n",
+		       __func__);
+		return -ENOMEM;
+	}
+
+	if (flags & OMAP_CRYPTO_COPY_DATA) {
+		scatterwalk_map_and_copy(buf, *sg, 0, total, 0);
+		if (flags & OMAP_CRYPTO_ZERO_BUF)
+			memset(buf + total, 0, new_len - total);
+	}
+
+	if (!(flags & OMAP_CRYPTO_FORCE_SINGLE_ENTRY))
+		sg_init_table(new_sg, 1);
+
+	sg_set_buf(new_sg, buf, new_len);
+
+	*sg = new_sg;
+
+	return 0;
+}
+
+static int omap_crypto_check_sg(struct scatterlist *sg, int total, int bs,
+				u16 flags)
+{
+	int len = 0;
+	int num_sg = 0;
+
+	if (!IS_ALIGNED(total, bs))
+		return OMAP_CRYPTO_NOT_ALIGNED;
+
+	while (sg) {
+		num_sg++;
+
+		if (!IS_ALIGNED(sg->offset, 4))
+			return OMAP_CRYPTO_NOT_ALIGNED;
+		if (!IS_ALIGNED(sg->length, bs))
+			return OMAP_CRYPTO_NOT_ALIGNED;
+#ifdef CONFIG_ZONE_DMA
+		if (page_zonenum(sg_page(sg)) != ZONE_DMA)
+			return OMAP_CRYPTO_NOT_ALIGNED;
+#endif
+
+		len += sg->length;
+		sg = sg_next(sg);
+
+		if (len >= total)
+			break;
+	}
+
+	if ((flags & OMAP_CRYPTO_FORCE_SINGLE_ENTRY) && num_sg > 1)
+		return OMAP_CRYPTO_NOT_ALIGNED;
+
+	if (len != total)
+		return OMAP_CRYPTO_BAD_DATA_LENGTH;
+
+	return 0;
+}
+
+int omap_crypto_align_sg(struct scatterlist **sg, int total, int bs,
+			 struct scatterlist *new_sg, u16 flags,
+			 u8 flags_shift, unsigned long *dd_flags)
+{
+	int ret;
+
+	*dd_flags &= ~(OMAP_CRYPTO_COPY_MASK << flags_shift);
+
+	if (flags & OMAP_CRYPTO_FORCE_COPY)
+		ret = OMAP_CRYPTO_NOT_ALIGNED;
+	else
+		ret = omap_crypto_check_sg(*sg, total, bs, flags);
+
+	if (ret == OMAP_CRYPTO_NOT_ALIGNED) {
+		ret = omap_crypto_copy_sgs(total, bs, sg, new_sg, flags);
+		if (ret)
+			return ret;
+		*dd_flags |= OMAP_CRYPTO_DATA_COPIED << flags_shift;
+	} else if (ret == OMAP_CRYPTO_BAD_DATA_LENGTH) {
+		ret = omap_crypto_copy_sg_lists(total, bs, sg, new_sg, flags);
+		if (ret)
+			return ret;
+		if (!(flags & OMAP_CRYPTO_FORCE_SINGLE_ENTRY))
+			*dd_flags |= OMAP_CRYPTO_SG_COPIED << flags_shift;
+	} else if (flags & OMAP_CRYPTO_FORCE_SINGLE_ENTRY) {
+		sg_set_buf(new_sg, sg_virt(*sg), (*sg)->length);
+	}
+
+	return 0;
+}
+EXPORT_SYMBOL_GPL(omap_crypto_align_sg);
+
+static void omap_crypto_copy_data(struct scatterlist *src,
+				  struct scatterlist *dst,
+				  int offset, int len)
+{
+	int amt;
+	void *srcb, *dstb;
+	int srco = 0, dsto = offset;
+
+	while (src && dst && len) {
+		if (srco >= src->length) {
+			srco -= src->length;
+			src = sg_next(src);
+			continue;
+		}
+
+		if (dsto >= dst->length) {
+			dsto -= dst->length;
+			dst = sg_next(dst);
+			continue;
+		}
+
+		amt = min(src->length - srco, dst->length - dsto);
+		amt = min(len, amt);
+
+		srcb = kmap_atomic(sg_page(src)) + srco + src->offset;
+		dstb = kmap_atomic(sg_page(dst)) + dsto + dst->offset;
+
+		memcpy(dstb, srcb, amt);
+
+		flush_dcache_page(sg_page(dst));
+
+		kunmap_atomic(srcb);
+		kunmap_atomic(dstb);
+
+		srco += amt;
+		dsto += amt;
+		len -= amt;
+	}
+}
+
+void omap_crypto_cleanup(struct scatterlist *sg, struct scatterlist *orig,
+			 int offset, int len, u8 flags_shift,
+			 unsigned long flags)
+{
+	void *buf;
+	int pages;
+
+	flags >>= flags_shift;
+	flags &= OMAP_CRYPTO_COPY_MASK;
+
+	if (!flags)
+		return;
+
+	buf = sg_virt(sg);
+	pages = get_order(len);
+
+	if (orig && (flags & OMAP_CRYPTO_DATA_COPIED))
+		omap_crypto_copy_data(sg, orig, offset, len);
+
+	if (flags & OMAP_CRYPTO_DATA_COPIED)
+		free_pages((unsigned long)buf, pages);
+	else if (flags & OMAP_CRYPTO_SG_COPIED)
+		kfree(sg);
+}
+EXPORT_SYMBOL_GPL(omap_crypto_cleanup);
+
+MODULE_DESCRIPTION("OMAP crypto support library.");
+MODULE_LICENSE("GPL v2");
+MODULE_AUTHOR("Tero Kristo <t-kristo@ti.com>");