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, 159 insertions, 0 deletions

diff --git a/include/net/libeth/tx.h b/include/net/libeth/tx.h
new file mode 100644
index 000000000..c3db5c6f1
--- /dev/null
+++ b/include/net/libeth/tx.h
@@ -0,0 +1,159 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/* Copyright (C) 2024-2025 Intel Corporation */
+
+#ifndef __LIBETH_TX_H
+#define __LIBETH_TX_H
+
+#include <linux/skbuff.h>
+
+#include <net/libeth/types.h>
+
+/* Tx buffer completion */
+
+/**
+ * enum libeth_sqe_type - type of &libeth_sqe to act on Tx completion
+ * @LIBETH_SQE_EMPTY: unused/empty OR XDP_TX/XSk frame, no action required
+ * @LIBETH_SQE_CTX: context descriptor with empty SQE, no action required
+ * @LIBETH_SQE_SLAB: kmalloc-allocated buffer, unmap and kfree()
+ * @LIBETH_SQE_FRAG: mapped skb frag, only unmap DMA
+ * @LIBETH_SQE_SKB: &sk_buff, unmap and napi_consume_skb(), update stats
+ * @__LIBETH_SQE_XDP_START: separator between skb and XDP types
+ * @LIBETH_SQE_XDP_TX: &skb_shared_info, libeth_xdp_return_buff_bulk(), stats
+ * @LIBETH_SQE_XDP_XMIT: &xdp_frame, unmap and xdp_return_frame_bulk(), stats
+ * @LIBETH_SQE_XDP_XMIT_FRAG: &xdp_frame frag, only unmap DMA
+ * @LIBETH_SQE_XSK_TX: &libeth_xdp_buff on XSk queue, xsk_buff_free(), stats
+ * @LIBETH_SQE_XSK_TX_FRAG: &libeth_xdp_buff frag on XSk queue, xsk_buff_free()
+ */
+enum libeth_sqe_type {
+	LIBETH_SQE_EMPTY		= 0U,
+	LIBETH_SQE_CTX,
+	LIBETH_SQE_SLAB,
+	LIBETH_SQE_FRAG,
+	LIBETH_SQE_SKB,
+
+	__LIBETH_SQE_XDP_START,
+	LIBETH_SQE_XDP_TX		= __LIBETH_SQE_XDP_START,
+	LIBETH_SQE_XDP_XMIT,
+	LIBETH_SQE_XDP_XMIT_FRAG,
+	LIBETH_SQE_XSK_TX,
+	LIBETH_SQE_XSK_TX_FRAG,
+};
+
+/**
+ * struct libeth_sqe - represents a Send Queue Element / Tx buffer
+ * @type: type of the buffer, see the enum above
+ * @rs_idx: index of the last buffer from the batch this one was sent in
+ * @raw: slab buffer to free via kfree()
+ * @skb: &sk_buff to consume
+ * @sinfo: skb shared info of an XDP_TX frame
+ * @xdpf: XDP frame from ::ndo_xdp_xmit()
+ * @xsk: XSk Rx frame from XDP_TX action
+ * @dma: DMA address to unmap
+ * @len: length of the mapped region to unmap
+ * @nr_frags: number of frags in the frame this buffer belongs to
+ * @packets: number of physical packets sent for this frame
+ * @bytes: number of physical bytes sent for this frame
+ * @priv: driver-private scratchpad
+ */
+struct libeth_sqe {
+	enum libeth_sqe_type		type:32;
+	u32				rs_idx;
+
+	union {
+		void				*raw;
+		struct sk_buff			*skb;
+		struct skb_shared_info		*sinfo;
+		struct xdp_frame		*xdpf;
+		struct libeth_xdp_buff		*xsk;
+	};
+
+	DEFINE_DMA_UNMAP_ADDR(dma);
+	DEFINE_DMA_UNMAP_LEN(len);
+
+	u32				nr_frags;
+	u32				packets;
+	u32				bytes;
+
+	unsigned long			priv;
+} __aligned_largest;
+
+/**
+ * LIBETH_SQE_CHECK_PRIV - check the driver's private SQE data
+ * @p: type or name of the object the driver wants to fit into &libeth_sqe
+ *
+ * Make sure the driver's private data fits into libeth_sqe::priv. To be used
+ * right after its declaration.
+ */
+#define LIBETH_SQE_CHECK_PRIV(p)					  \
+	static_assert(sizeof(p) <= sizeof_field(struct libeth_sqe, priv))
+
+/**
+ * struct libeth_cq_pp - completion queue poll params
+ * @dev: &device to perform DMA unmapping
+ * @bq: XDP frame bulk to combine return operations
+ * @ss: onstack NAPI stats to fill
+ * @xss: onstack XDPSQ NAPI stats to fill
+ * @xdp_tx: number of XDP-not-XSk frames processed
+ * @napi: whether it's called from the NAPI context
+ *
+ * libeth uses this structure to access objects needed for performing full
+ * Tx complete operation without passing lots of arguments and change the
+ * prototypes each time a new one is added.
+ */
+struct libeth_cq_pp {
+	struct device			*dev;
+	struct xdp_frame_bulk		*bq;
+
+	union {
+		struct libeth_sq_napi_stats	*ss;
+		struct libeth_xdpsq_napi_stats	*xss;
+	};
+	u32				xdp_tx;
+
+	bool				napi;
+};
+
+/**
+ * libeth_tx_complete - perform Tx completion for one SQE
+ * @sqe: SQE to complete
+ * @cp: poll params
+ *
+ * Do Tx complete for all the types of buffers, incl. freeing, unmapping,
+ * updating the stats etc.
+ */
+static inline void libeth_tx_complete(struct libeth_sqe *sqe,
+				      const struct libeth_cq_pp *cp)
+{
+	switch (sqe->type) {
+	case LIBETH_SQE_EMPTY:
+		return;
+	case LIBETH_SQE_SKB:
+	case LIBETH_SQE_FRAG:
+	case LIBETH_SQE_SLAB:
+		dma_unmap_page(cp->dev, dma_unmap_addr(sqe, dma),
+			       dma_unmap_len(sqe, len), DMA_TO_DEVICE);
+		break;
+	default:
+		break;
+	}
+
+	switch (sqe->type) {
+	case LIBETH_SQE_SKB:
+		cp->ss->packets += sqe->packets;
+		cp->ss->bytes += sqe->bytes;
+
+		napi_consume_skb(sqe->skb, cp->napi);
+		break;
+	case LIBETH_SQE_SLAB:
+		kfree(sqe->raw);
+		break;
+	default:
+		break;
+	}
+
+	sqe->type = LIBETH_SQE_EMPTY;
+}
+
+void libeth_tx_complete_any(struct libeth_sqe *sqe, struct libeth_cq_pp *cp);
+
+#endif /* __LIBETH_TX_H */