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, 155 insertions, 0 deletions

diff --git a/io_uring/query.c b/io_uring/query.c
new file mode 100644
index 000000000..7bce2ec16
--- /dev/null
+++ b/io_uring/query.c
@@ -0,0 +1,155 @@
+// SPDX-License-Identifier: GPL-2.0
+
+#include "linux/io_uring/query.h"
+
+#include "query.h"
+#include "io_uring.h"
+#include "zcrx.h"
+
+union io_query_data {
+	struct io_uring_query_opcode opcodes;
+	struct io_uring_query_zcrx zcrx;
+	struct io_uring_query_zcrx_event zcrx_notif;
+	struct io_uring_query_scq scq;
+};
+
+#define IO_MAX_QUERY_SIZE		sizeof(union io_query_data)
+#define IO_MAX_QUERY_ENTRIES		1000
+
+static ssize_t io_query_ops(union io_query_data *data)
+{
+	struct io_uring_query_opcode *e = &data->opcodes;
+
+	e->nr_request_opcodes = IORING_OP_LAST;
+	e->nr_register_opcodes = IORING_REGISTER_LAST;
+	e->feature_flags = IORING_FEAT_FLAGS;
+	e->ring_setup_flags = IORING_SETUP_FLAGS;
+	e->enter_flags = IORING_ENTER_FLAGS;
+	e->sqe_flags = SQE_VALID_FLAGS;
+	e->nr_query_opcodes = __IO_URING_QUERY_MAX;
+	e->__pad = 0;
+	return sizeof(*e);
+}
+
+static ssize_t io_query_zcrx(union io_query_data *data)
+{
+	struct io_uring_query_zcrx *e = &data->zcrx;
+
+	e->register_flags = ZCRX_SUPPORTED_REG_FLAGS;
+	e->area_flags = IORING_ZCRX_AREA_DMABUF;
+	e->nr_ctrl_opcodes = __ZCRX_CTRL_LAST;
+	e->rq_hdr_size = sizeof(struct zcrx_rq_hdr);
+	e->rq_hdr_alignment = L1_CACHE_BYTES;
+	e->features = ZCRX_FEATURES;
+	e->__resv2 = 0;
+	return sizeof(*e);
+}
+
+static ssize_t io_query_zcrx_notif(union io_query_data *data)
+{
+	struct io_uring_query_zcrx_event *e = &data->zcrx_notif;
+
+	e->event_flags = ZCRX_EVENT_TYPE_MASK;
+	e->stats_size = sizeof(struct zcrx_stats);
+	e->stats_off_alignment = __alignof__(struct zcrx_stats);
+	e->__resv1 = 0;
+	memset(&e->__resv2, 0, sizeof(e->__resv2));
+	return sizeof(*e);
+}
+
+static ssize_t io_query_scq(union io_query_data *data)
+{
+	struct io_uring_query_scq *e = &data->scq;
+
+	e->hdr_size = sizeof(struct io_rings);
+	e->hdr_alignment = SMP_CACHE_BYTES;
+	return sizeof(*e);
+}
+
+static int io_handle_query_entry(union io_query_data *data, void __user *uhdr,
+				 u64 *next_entry)
+{
+	struct io_uring_query_hdr hdr;
+	size_t usize, res_size = 0;
+	ssize_t ret = -EINVAL;
+	void __user *udata;
+
+	if (copy_from_user(&hdr, uhdr, sizeof(hdr)))
+		return -EFAULT;
+	/* copy_struct_to_user() zeros up to usize bytes */
+	if (hdr.size > PAGE_SIZE)
+		return -E2BIG;
+	usize = hdr.size;
+	hdr.size = min(hdr.size, IO_MAX_QUERY_SIZE);
+	udata = u64_to_user_ptr(hdr.query_data);
+
+	if (hdr.query_op >= __IO_URING_QUERY_MAX) {
+		ret = -EOPNOTSUPP;
+		goto out;
+	}
+	if (!mem_is_zero(hdr.__resv, sizeof(hdr.__resv)) || hdr.result || !hdr.size)
+		goto out;
+	if (copy_from_user(data, udata, hdr.size))
+		return -EFAULT;
+
+	switch (hdr.query_op) {
+	case IO_URING_QUERY_OPCODES:
+		ret = io_query_ops(data);
+		break;
+	case IO_URING_QUERY_ZCRX:
+		ret = io_query_zcrx(data);
+		break;
+	case IO_URING_QUERY_ZCRX_EVENT:
+		ret = io_query_zcrx_notif(data);
+		break;
+	case IO_URING_QUERY_SCQ:
+		ret = io_query_scq(data);
+		break;
+	}
+
+	if (ret >= 0) {
+		if (WARN_ON_ONCE(ret > IO_MAX_QUERY_SIZE))
+			return -EFAULT;
+		res_size = ret;
+		ret = 0;
+	}
+out:
+	hdr.result = ret;
+	hdr.size = min_t(size_t, usize, res_size);
+
+	if (copy_struct_to_user(udata, usize, data, hdr.size, NULL))
+		return -EFAULT;
+	if (copy_to_user(uhdr, &hdr, sizeof(hdr)))
+		return -EFAULT;
+	*next_entry = hdr.next_entry;
+	return 0;
+}
+
+int io_query(void __user *arg, unsigned nr_args)
+{
+	union io_query_data entry_buffer;
+	void __user *uhdr = arg;
+	int ret, nr = 0;
+
+	memset(&entry_buffer, 0, sizeof(entry_buffer));
+
+	if (nr_args)
+		return -EINVAL;
+
+	while (uhdr) {
+		u64 next_hdr;
+
+		ret = io_handle_query_entry(&entry_buffer, uhdr, &next_hdr);
+		if (ret)
+			return ret;
+		uhdr = u64_to_user_ptr(next_hdr);
+
+		/* Have some limit to avoid a potential cycle */
+		if (++nr >= IO_MAX_QUERY_ENTRIES)
+			return -ERANGE;
+		if (fatal_signal_pending(current))
+			return -EINTR;
+		cond_resched();
+	}
+	return 0;
+}