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, 143 insertions, 0 deletions

diff --git a/drivers/dma/virt-dma.c b/drivers/dma/virt-dma.c
new file mode 100644
index 000000000..7961172a7
--- /dev/null
+++ b/drivers/dma/virt-dma.c
@@ -0,0 +1,143 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Virtual DMA channel support for DMAengine
+ *
+ * Copyright (C) 2012 Russell King
+ */
+#include <linux/device.h>
+#include <linux/dmaengine.h>
+#include <linux/module.h>
+#include <linux/spinlock.h>
+
+#include "virt-dma.h"
+
+static struct virt_dma_desc *to_virt_desc(struct dma_async_tx_descriptor *tx)
+{
+	return container_of(tx, struct virt_dma_desc, tx);
+}
+
+dma_cookie_t vchan_tx_submit(struct dma_async_tx_descriptor *tx)
+{
+	struct virt_dma_chan *vc = to_virt_chan(tx->chan);
+	struct virt_dma_desc *vd = to_virt_desc(tx);
+	unsigned long flags;
+	dma_cookie_t cookie;
+
+	spin_lock_irqsave(&vc->lock, flags);
+	cookie = dma_cookie_assign(tx);
+
+	list_move_tail(&vd->node, &vc->desc_submitted);
+	spin_unlock_irqrestore(&vc->lock, flags);
+
+	dev_dbg(vc->chan.device->dev, "vchan %p: txd %p[%x]: submitted\n",
+		vc, vd, cookie);
+
+	return cookie;
+}
+EXPORT_SYMBOL_GPL(vchan_tx_submit);
+
+/**
+ * vchan_tx_desc_free - free a reusable descriptor
+ * @tx: the transfer
+ *
+ * This function frees a previously allocated reusable descriptor. The only
+ * other way is to clear the DMA_CTRL_REUSE flag and submit one last time the
+ * transfer.
+ *
+ * Returns 0 upon success
+ */
+int vchan_tx_desc_free(struct dma_async_tx_descriptor *tx)
+{
+	struct virt_dma_chan *vc = to_virt_chan(tx->chan);
+	struct virt_dma_desc *vd = to_virt_desc(tx);
+	unsigned long flags;
+
+	spin_lock_irqsave(&vc->lock, flags);
+	list_del(&vd->node);
+	spin_unlock_irqrestore(&vc->lock, flags);
+
+	dev_dbg(vc->chan.device->dev, "vchan %p: txd %p[%x]: freeing\n",
+		vc, vd, vd->tx.cookie);
+	vc->desc_free(vd);
+	return 0;
+}
+EXPORT_SYMBOL_GPL(vchan_tx_desc_free);
+
+struct virt_dma_desc *vchan_find_desc(struct virt_dma_chan *vc,
+	dma_cookie_t cookie)
+{
+	struct virt_dma_desc *vd;
+
+	list_for_each_entry(vd, &vc->desc_issued, node)
+		if (vd->tx.cookie == cookie)
+			return vd;
+
+	return NULL;
+}
+EXPORT_SYMBOL_GPL(vchan_find_desc);
+
+/*
+ * This tasklet handles the completion of a DMA descriptor by
+ * calling its callback and freeing it.
+ */
+static void vchan_complete(struct tasklet_struct *t)
+{
+	struct virt_dma_chan *vc = from_tasklet(vc, t, task);
+	struct virt_dma_desc *vd, *_vd;
+	struct dmaengine_desc_callback cb;
+	LIST_HEAD(head);
+
+	spin_lock_irq(&vc->lock);
+	list_splice_tail_init(&vc->desc_completed, &head);
+	vd = vc->cyclic;
+	if (vd) {
+		vc->cyclic = NULL;
+		dmaengine_desc_get_callback(&vd->tx, &cb);
+	} else {
+		memset(&cb, 0, sizeof(cb));
+	}
+	spin_unlock_irq(&vc->lock);
+
+	dmaengine_desc_callback_invoke(&cb, &vd->tx_result);
+
+	list_for_each_entry_safe(vd, _vd, &head, node) {
+		dmaengine_desc_get_callback(&vd->tx, &cb);
+
+		list_del(&vd->node);
+		dmaengine_desc_callback_invoke(&cb, &vd->tx_result);
+		vchan_vdesc_fini(vd);
+	}
+}
+
+void vchan_dma_desc_free_list(struct virt_dma_chan *vc, struct list_head *head)
+{
+	struct virt_dma_desc *vd, *_vd;
+
+	list_for_each_entry_safe(vd, _vd, head, node) {
+		list_del(&vd->node);
+		vchan_vdesc_fini(vd);
+	}
+}
+EXPORT_SYMBOL_GPL(vchan_dma_desc_free_list);
+
+void vchan_init(struct virt_dma_chan *vc, struct dma_device *dmadev)
+{
+	dma_cookie_init(&vc->chan);
+
+	spin_lock_init(&vc->lock);
+	INIT_LIST_HEAD(&vc->desc_allocated);
+	INIT_LIST_HEAD(&vc->desc_submitted);
+	INIT_LIST_HEAD(&vc->desc_issued);
+	INIT_LIST_HEAD(&vc->desc_completed);
+	INIT_LIST_HEAD(&vc->desc_terminated);
+
+	tasklet_setup(&vc->task, vchan_complete);
+
+	vc->chan.device = dmadev;
+	list_add_tail(&vc->chan.device_node, &dmadev->channels);
+}
+EXPORT_SYMBOL_GPL(vchan_init);
+
+MODULE_AUTHOR("Russell King");
+MODULE_DESCRIPTION("Virtual DMA channel support for DMAengine");
+MODULE_LICENSE("GPL");