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, 201 insertions, 0 deletions

diff --git a/drivers/dma/dmaengine.h b/drivers/dma/dmaengine.h
new file mode 100644
index 000000000..53f16d3f0
--- /dev/null
+++ b/drivers/dma/dmaengine.h
@@ -0,0 +1,201 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * The contents of this file are private to DMA engine drivers, and is not
+ * part of the API to be used by DMA engine users.
+ */
+#ifndef DMAENGINE_H
+#define DMAENGINE_H
+
+#include <linux/bug.h>
+#include <linux/dmaengine.h>
+
+/**
+ * dma_cookie_init - initialize the cookies for a DMA channel
+ * @chan: dma channel to initialize
+ */
+static inline void dma_cookie_init(struct dma_chan *chan)
+{
+	chan->cookie = DMA_MIN_COOKIE;
+	chan->completed_cookie = DMA_MIN_COOKIE;
+}
+
+/**
+ * dma_cookie_assign - assign a DMA engine cookie to the descriptor
+ * @tx: descriptor needing cookie
+ *
+ * Assign a unique non-zero per-channel cookie to the descriptor.
+ * Note: caller is expected to hold a lock to prevent concurrency.
+ */
+static inline dma_cookie_t dma_cookie_assign(struct dma_async_tx_descriptor *tx)
+{
+	struct dma_chan *chan = tx->chan;
+	dma_cookie_t cookie;
+
+	cookie = chan->cookie + 1;
+	if (cookie < DMA_MIN_COOKIE)
+		cookie = DMA_MIN_COOKIE;
+	tx->cookie = chan->cookie = cookie;
+
+	return cookie;
+}
+
+/**
+ * dma_cookie_complete - complete a descriptor
+ * @tx: descriptor to complete
+ *
+ * Mark this descriptor complete by updating the channels completed
+ * cookie marker.  Zero the descriptors cookie to prevent accidental
+ * repeated completions.
+ *
+ * Note: caller is expected to hold a lock to prevent concurrency.
+ */
+static inline void dma_cookie_complete(struct dma_async_tx_descriptor *tx)
+{
+	BUG_ON(tx->cookie < DMA_MIN_COOKIE);
+	tx->chan->completed_cookie = tx->cookie;
+	tx->cookie = 0;
+}
+
+/**
+ * dma_cookie_status - report cookie status
+ * @chan: dma channel
+ * @cookie: cookie we are interested in
+ * @state: dma_tx_state structure to return last/used cookies
+ *
+ * Report the status of the cookie, filling in the state structure if
+ * non-NULL.  No locking is required.
+ */
+static inline enum dma_status dma_cookie_status(struct dma_chan *chan,
+	dma_cookie_t cookie, struct dma_tx_state *state)
+{
+	dma_cookie_t used, complete;
+
+	used = chan->cookie;
+	complete = chan->completed_cookie;
+	barrier();
+	if (state) {
+		state->last = complete;
+		state->used = used;
+		state->residue = 0;
+		state->in_flight_bytes = 0;
+	}
+	return dma_async_is_complete(cookie, complete, used);
+}
+
+static inline void dma_set_residue(struct dma_tx_state *state, u32 residue)
+{
+	if (state)
+		state->residue = residue;
+}
+
+static inline void dma_set_in_flight_bytes(struct dma_tx_state *state,
+					   u32 in_flight_bytes)
+{
+	if (state)
+		state->in_flight_bytes = in_flight_bytes;
+}
+
+struct dmaengine_desc_callback {
+	dma_async_tx_callback callback;
+	dma_async_tx_callback_result callback_result;
+	void *callback_param;
+};
+
+/**
+ * dmaengine_desc_get_callback - get the passed in callback function
+ * @tx: tx descriptor
+ * @cb: temp struct to hold the callback info
+ *
+ * Fill the passed in cb struct with what's available in the passed in
+ * tx descriptor struct
+ * No locking is required.
+ */
+static inline void
+dmaengine_desc_get_callback(struct dma_async_tx_descriptor *tx,
+			    struct dmaengine_desc_callback *cb)
+{
+	cb->callback = tx->callback;
+	cb->callback_result = tx->callback_result;
+	cb->callback_param = tx->callback_param;
+}
+
+/**
+ * dmaengine_desc_callback_invoke - call the callback function in cb struct
+ * @cb: temp struct that is holding the callback info
+ * @result: transaction result
+ *
+ * Call the callback function provided in the cb struct with the parameter
+ * in the cb struct.
+ * Locking is dependent on the driver.
+ */
+static inline void
+dmaengine_desc_callback_invoke(struct dmaengine_desc_callback *cb,
+			       const struct dmaengine_result *result)
+{
+	struct dmaengine_result dummy_result = {
+		.result = DMA_TRANS_NOERROR,
+		.residue = 0
+	};
+
+	if (cb->callback_result) {
+		if (!result)
+			result = &dummy_result;
+		cb->callback_result(cb->callback_param, result);
+	} else if (cb->callback) {
+		cb->callback(cb->callback_param);
+	}
+}
+
+/**
+ * dmaengine_desc_get_callback_invoke - get the callback in tx descriptor and
+ * 					then immediately call the callback.
+ * @tx: dma async tx descriptor
+ * @result: transaction result
+ *
+ * Call dmaengine_desc_get_callback() and dmaengine_desc_callback_invoke()
+ * in a single function since no work is necessary in between for the driver.
+ * Locking is dependent on the driver.
+ */
+static inline void
+dmaengine_desc_get_callback_invoke(struct dma_async_tx_descriptor *tx,
+				   const struct dmaengine_result *result)
+{
+	struct dmaengine_desc_callback cb;
+
+	dmaengine_desc_get_callback(tx, &cb);
+	dmaengine_desc_callback_invoke(&cb, result);
+}
+
+/**
+ * dmaengine_desc_callback_valid - verify the callback is valid in cb
+ * @cb: callback info struct
+ *
+ * Return a bool that verifies whether callback in cb is valid or not.
+ * No locking is required.
+ */
+static inline bool
+dmaengine_desc_callback_valid(struct dmaengine_desc_callback *cb)
+{
+	return cb->callback || cb->callback_result;
+}
+
+struct dma_chan *dma_get_slave_channel(struct dma_chan *chan);
+struct dma_chan *dma_get_any_slave_channel(struct dma_device *device);
+
+#ifdef CONFIG_DEBUG_FS
+#include <linux/debugfs.h>
+
+static inline struct dentry *
+dmaengine_get_debugfs_root(struct dma_device *dma_dev) {
+	return dma_dev->dbg_dev_root;
+}
+#else
+struct dentry;
+static inline struct dentry *
+dmaengine_get_debugfs_root(struct dma_device *dma_dev)
+{
+	return NULL;
+}
+#endif /* CONFIG_DEBUG_FS */
+
+#endif