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, 149 insertions, 0 deletions

diff --git a/mm/dmapool_test.c b/mm/dmapool_test.c
new file mode 100644
index 000000000..d829d99aa
--- /dev/null
+++ b/mm/dmapool_test.c
@@ -0,0 +1,149 @@
+// SPDX-License-Identifier: GPL-2.0
+#include <linux/device.h>
+#include <linux/dma-map-ops.h>
+#include <linux/dma-mapping.h>
+#include <linux/dmapool.h>
+#include <linux/kernel.h>
+#include <linux/ktime.h>
+#include <linux/module.h>
+
+#define NR_TESTS (100)
+
+struct dma_pool_pair {
+	dma_addr_t dma;
+	void *v;
+};
+
+struct dmapool_parms {
+	size_t size;
+	size_t align;
+	size_t boundary;
+};
+
+static const struct dmapool_parms pool_parms[] = {
+	{ .size = 16, .align = 16, .boundary = 0 },
+	{ .size = 64, .align = 64, .boundary = 0 },
+	{ .size = 256, .align = 256, .boundary = 0 },
+	{ .size = 1024, .align = 1024, .boundary = 0 },
+	{ .size = 4096, .align = 4096, .boundary = 0 },
+	{ .size = 68, .align = 32, .boundary = 4096 },
+};
+
+static struct dma_pool *pool;
+static struct device test_dev;
+static u64 dma_mask;
+
+static inline int nr_blocks(int size)
+{
+	return clamp_t(int, (PAGE_SIZE / size) * 512, 1024, 8192);
+}
+
+static int dmapool_test_alloc(struct dma_pool_pair *p, int blocks)
+{
+	int i;
+
+	for (i = 0; i < blocks; i++) {
+		p[i].v = dma_pool_alloc(pool, GFP_KERNEL,
+					&p[i].dma);
+		if (!p[i].v)
+			goto pool_fail;
+	}
+
+	for (i = 0; i < blocks; i++)
+		dma_pool_free(pool, p[i].v, p[i].dma);
+
+	return 0;
+
+pool_fail:
+	for (--i; i >= 0; i--)
+		dma_pool_free(pool, p[i].v, p[i].dma);
+	return -ENOMEM;
+}
+
+static int dmapool_test_block(const struct dmapool_parms *parms)
+{
+	int blocks = nr_blocks(parms->size);
+	ktime_t start_time, end_time;
+	struct dma_pool_pair *p;
+	int i, ret;
+
+	p = kzalloc_objs(*p, blocks);
+	if (!p)
+		return -ENOMEM;
+
+	pool = dma_pool_create("test pool", &test_dev, parms->size,
+			       parms->align, parms->boundary);
+	if (!pool) {
+		ret = -ENOMEM;
+		goto free_pairs;
+	}
+
+	start_time = ktime_get();
+	for (i = 0; i < NR_TESTS; i++) {
+		ret = dmapool_test_alloc(p, blocks);
+		if (ret)
+			goto free_pool;
+		if (need_resched())
+			cond_resched();
+	}
+	end_time = ktime_get();
+
+	printk("dmapool test: size:%-4zu align:%-4zu blocks:%-4d time:%llu\n",
+		parms->size, parms->align, blocks,
+		ktime_us_delta(end_time, start_time));
+
+free_pool:
+	dma_pool_destroy(pool);
+free_pairs:
+	kfree(p);
+	return ret;
+}
+
+static void dmapool_test_release(struct device *dev)
+{
+}
+
+static int dmapool_checks(void)
+{
+	int i, ret;
+
+	ret = dev_set_name(&test_dev, "dmapool-test");
+	if (ret)
+		return ret;
+
+	ret = device_register(&test_dev);
+	if (ret) {
+		printk("%s: register failed:%d\n", __func__, ret);
+		goto put_device;
+	}
+
+	test_dev.release = dmapool_test_release;
+	set_dma_ops(&test_dev, NULL);
+	test_dev.dma_mask = &dma_mask;
+	ret = dma_set_mask_and_coherent(&test_dev, DMA_BIT_MASK(64));
+	if (ret) {
+		printk("%s: mask failed:%d\n", __func__, ret);
+		goto del_device;
+	}
+
+	for (i = 0; i < ARRAY_SIZE(pool_parms); i++) {
+		ret = dmapool_test_block(&pool_parms[i]);
+		if (ret)
+			break;
+	}
+
+del_device:
+	device_del(&test_dev);
+put_device:
+	put_device(&test_dev);
+	return ret;
+}
+
+static void dmapool_exit(void)
+{
+}
+
+module_init(dmapool_checks);
+module_exit(dmapool_exit);
+MODULE_DESCRIPTION("dma_pool timing test");
+MODULE_LICENSE("GPL");