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, 88 insertions, 0 deletions

diff --git a/include/rdma/iter.h b/include/rdma/iter.h
new file mode 100644
index 000000000..19d64ef04
--- /dev/null
+++ b/include/rdma/iter.h
@@ -0,0 +1,88 @@
+/* SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB */
+/* Copyright (c) 2026, NVIDIA CORPORATION & AFFILIATES. */
+
+#ifndef _RDMA_ITER_H_
+#define _RDMA_ITER_H_
+
+#include <linux/scatterlist.h>
+#include <rdma/ib_umem.h>
+
+/**
+ * IB block DMA iterator
+ *
+ * Iterates the DMA-mapped SGL in contiguous memory blocks aligned
+ * to a HW supported page size.
+ */
+struct ib_block_iter {
+	/* internal states */
+	struct scatterlist *__sg;	/* sg holding the current aligned block */
+	dma_addr_t __dma_addr;		/* unaligned DMA address of this block */
+	size_t __sg_numblocks;		/* ib_umem_num_dma_blocks() */
+	unsigned int __sg_nents;	/* number of SG entries */
+	unsigned int __sg_advance;	/* number of bytes to advance in sg in next step */
+	unsigned int __pg_bit;		/* alignment of current block */
+};
+
+void __rdma_block_iter_start(struct ib_block_iter *biter,
+			     struct scatterlist *sglist,
+			     unsigned int nents,
+			     unsigned long pgsz);
+bool __rdma_block_iter_next(struct ib_block_iter *biter);
+
+/**
+ * rdma_block_iter_dma_address - get the aligned dma address of the current
+ * block held by the block iterator.
+ * @biter: block iterator holding the memory block
+ */
+static inline dma_addr_t
+rdma_block_iter_dma_address(struct ib_block_iter *biter)
+{
+	return biter->__dma_addr & ~(BIT_ULL(biter->__pg_bit) - 1);
+}
+
+/**
+ * rdma_for_each_block - iterate over contiguous memory blocks of the sg list
+ * @sglist: sglist to iterate over
+ * @biter: block iterator holding the memory block
+ * @nents: maximum number of sg entries to iterate over
+ * @pgsz: best HW supported page size to use
+ *
+ * Callers may use rdma_block_iter_dma_address() to get each
+ * blocks aligned DMA address.
+ */
+#define rdma_for_each_block(sglist, biter, nents, pgsz)		\
+	for (__rdma_block_iter_start(biter, sglist, nents,	\
+				     pgsz);			\
+	     __rdma_block_iter_next(biter);)
+
+static inline void __rdma_umem_block_iter_start(struct ib_block_iter *biter,
+						struct ib_umem *umem,
+						unsigned long pgsz)
+{
+	__rdma_block_iter_start(biter, umem->sgt_append.sgt.sgl,
+				umem->sgt_append.sgt.nents, pgsz);
+	biter->__sg_advance = ib_umem_offset(umem) & ~(pgsz - 1);
+	biter->__sg_numblocks = ib_umem_num_dma_blocks(umem, pgsz);
+}
+
+static inline bool __rdma_umem_block_iter_next(struct ib_block_iter *biter)
+{
+	return __rdma_block_iter_next(biter) && biter->__sg_numblocks--;
+}
+
+/**
+ * rdma_umem_for_each_dma_block - iterate over contiguous DMA blocks of the umem
+ * @umem: umem to iterate over
+ * @pgsz: Page size to split the list into
+ *
+ * pgsz must be <= PAGE_SIZE or computed by ib_umem_find_best_pgsz(). The
+ * returned DMA blocks will be aligned to pgsz and span the range:
+ * ALIGN_DOWN(umem->address, pgsz) to ALIGN(umem->address + umem->length, pgsz)
+ *
+ * Performs exactly ib_umem_num_dma_blocks() iterations.
+ */
+#define rdma_umem_for_each_dma_block(umem, biter, pgsz)                        \
+	for (__rdma_umem_block_iter_start(biter, umem, pgsz);                  \
+	     __rdma_umem_block_iter_next(biter);)
+
+#endif /* _RDMA_ITER_H_ */