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, 306 insertions, 0 deletions

diff --git a/io_uring/fs.c b/io_uring/fs.c
new file mode 100644
index 000000000..26ea841a2
--- /dev/null
+++ b/io_uring/fs.c
@@ -0,0 +1,306 @@
+// SPDX-License-Identifier: GPL-2.0
+#include <linux/kernel.h>
+#include <linux/errno.h>
+#include <linux/fs.h>
+#include <linux/file.h>
+#include <linux/mm.h>
+#include <linux/slab.h>
+#include <linux/namei.h>
+#include <linux/io_uring.h>
+
+#include <uapi/linux/io_uring.h>
+
+#include "../fs/internal.h"
+
+#include "io_uring.h"
+#include "fs.h"
+
+struct io_rename {
+	struct file			*file;
+	int				old_dfd;
+	int				new_dfd;
+	struct delayed_filename		oldpath;
+	struct delayed_filename		newpath;
+	int				flags;
+};
+
+struct io_unlink {
+	struct file			*file;
+	int				dfd;
+	int				flags;
+	struct delayed_filename		filename;
+};
+
+struct io_mkdir {
+	struct file			*file;
+	int				dfd;
+	umode_t				mode;
+	struct delayed_filename		filename;
+};
+
+struct io_link {
+	struct file			*file;
+	int				old_dfd;
+	int				new_dfd;
+	struct delayed_filename		oldpath;
+	struct delayed_filename		newpath;
+	int				flags;
+};
+
+int io_renameat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
+{
+	struct io_rename *ren = io_kiocb_to_cmd(req, struct io_rename);
+	const char __user *oldf, *newf;
+	int err;
+
+	if (sqe->buf_index || sqe->splice_fd_in)
+		return -EINVAL;
+	if (unlikely(req->flags & REQ_F_FIXED_FILE))
+		return -EBADF;
+
+	ren->old_dfd = READ_ONCE(sqe->fd);
+	oldf = u64_to_user_ptr(READ_ONCE(sqe->addr));
+	newf = u64_to_user_ptr(READ_ONCE(sqe->addr2));
+	ren->new_dfd = READ_ONCE(sqe->len);
+	ren->flags = READ_ONCE(sqe->rename_flags);
+
+	err = delayed_getname(&ren->oldpath, oldf);
+	if (unlikely(err))
+		return err;
+
+	err = delayed_getname(&ren->newpath, newf);
+	if (unlikely(err)) {
+		dismiss_delayed_filename(&ren->oldpath);
+		return err;
+	}
+
+	req->flags |= REQ_F_NEED_CLEANUP;
+	req->flags |= REQ_F_FORCE_ASYNC;
+	return 0;
+}
+
+int io_renameat(struct io_kiocb *req, unsigned int issue_flags)
+{
+	struct io_rename *ren = io_kiocb_to_cmd(req, struct io_rename);
+	CLASS(filename_complete_delayed, old)(&ren->oldpath);
+	CLASS(filename_complete_delayed, new)(&ren->newpath);
+	int ret;
+
+	WARN_ON_ONCE(issue_flags & IO_URING_F_NONBLOCK);
+
+	ret = filename_renameat2(ren->old_dfd, old,
+				 ren->new_dfd, new, ren->flags);
+
+	req->flags &= ~REQ_F_NEED_CLEANUP;
+	io_req_set_res(req, ret, 0);
+	return IOU_COMPLETE;
+}
+
+void io_renameat_cleanup(struct io_kiocb *req)
+{
+	struct io_rename *ren = io_kiocb_to_cmd(req, struct io_rename);
+
+	dismiss_delayed_filename(&ren->oldpath);
+	dismiss_delayed_filename(&ren->newpath);
+}
+
+int io_unlinkat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
+{
+	struct io_unlink *un = io_kiocb_to_cmd(req, struct io_unlink);
+	const char __user *fname;
+	int err;
+
+	if (sqe->off || sqe->len || sqe->buf_index || sqe->splice_fd_in ||
+	    sqe->addr3 || sqe->__pad2[0])
+		return -EINVAL;
+	if (unlikely(req->flags & REQ_F_FIXED_FILE))
+		return -EBADF;
+
+	un->dfd = READ_ONCE(sqe->fd);
+
+	un->flags = READ_ONCE(sqe->unlink_flags);
+	if (un->flags & ~AT_REMOVEDIR)
+		return -EINVAL;
+
+	fname = u64_to_user_ptr(READ_ONCE(sqe->addr));
+	err = delayed_getname(&un->filename, fname);
+	if (unlikely(err))
+		return err;
+
+	req->flags |= REQ_F_NEED_CLEANUP;
+	req->flags |= REQ_F_FORCE_ASYNC;
+	return 0;
+}
+
+int io_unlinkat(struct io_kiocb *req, unsigned int issue_flags)
+{
+	struct io_unlink *un = io_kiocb_to_cmd(req, struct io_unlink);
+	CLASS(filename_complete_delayed, name)(&un->filename);
+	int ret;
+
+	WARN_ON_ONCE(issue_flags & IO_URING_F_NONBLOCK);
+
+	if (un->flags & AT_REMOVEDIR)
+		ret = filename_rmdir(un->dfd, name);
+	else
+		ret = filename_unlinkat(un->dfd, name);
+
+	req->flags &= ~REQ_F_NEED_CLEANUP;
+	io_req_set_res(req, ret, 0);
+	return IOU_COMPLETE;
+}
+
+void io_unlinkat_cleanup(struct io_kiocb *req)
+{
+	struct io_unlink *ul = io_kiocb_to_cmd(req, struct io_unlink);
+
+	dismiss_delayed_filename(&ul->filename);
+}
+
+int io_mkdirat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
+{
+	struct io_mkdir *mkd = io_kiocb_to_cmd(req, struct io_mkdir);
+	const char __user *fname;
+	int err;
+
+	if (sqe->off || sqe->rw_flags || sqe->buf_index || sqe->splice_fd_in)
+		return -EINVAL;
+	if (unlikely(req->flags & REQ_F_FIXED_FILE))
+		return -EBADF;
+
+	mkd->dfd = READ_ONCE(sqe->fd);
+	mkd->mode = READ_ONCE(sqe->len);
+
+	fname = u64_to_user_ptr(READ_ONCE(sqe->addr));
+	err = delayed_getname(&mkd->filename, fname);
+	if (unlikely(err))
+		return err;
+
+	req->flags |= REQ_F_NEED_CLEANUP;
+	req->flags |= REQ_F_FORCE_ASYNC;
+	return 0;
+}
+
+int io_mkdirat(struct io_kiocb *req, unsigned int issue_flags)
+{
+	struct io_mkdir *mkd = io_kiocb_to_cmd(req, struct io_mkdir);
+	CLASS(filename_complete_delayed, name)(&mkd->filename);
+	int ret;
+
+	WARN_ON_ONCE(issue_flags & IO_URING_F_NONBLOCK);
+
+	ret = filename_mkdirat(mkd->dfd, name, mkd->mode);
+
+	req->flags &= ~REQ_F_NEED_CLEANUP;
+	io_req_set_res(req, ret, 0);
+	return IOU_COMPLETE;
+}
+
+void io_mkdirat_cleanup(struct io_kiocb *req)
+{
+	struct io_mkdir *md = io_kiocb_to_cmd(req, struct io_mkdir);
+
+	dismiss_delayed_filename(&md->filename);
+}
+
+int io_symlinkat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
+{
+	struct io_link *sl = io_kiocb_to_cmd(req, struct io_link);
+	const char __user *oldpath, *newpath;
+	int err;
+
+	if (sqe->len || sqe->rw_flags || sqe->buf_index || sqe->splice_fd_in)
+		return -EINVAL;
+	if (unlikely(req->flags & REQ_F_FIXED_FILE))
+		return -EBADF;
+
+	sl->new_dfd = READ_ONCE(sqe->fd);
+	oldpath = u64_to_user_ptr(READ_ONCE(sqe->addr));
+	newpath = u64_to_user_ptr(READ_ONCE(sqe->addr2));
+
+	err = delayed_getname(&sl->oldpath, oldpath);
+	if (unlikely(err))
+		return err;
+
+	err = delayed_getname(&sl->newpath, newpath);
+	if (unlikely(err)) {
+		dismiss_delayed_filename(&sl->oldpath);
+		return err;
+	}
+
+	req->flags |= REQ_F_NEED_CLEANUP;
+	req->flags |= REQ_F_FORCE_ASYNC;
+	return 0;
+}
+
+int io_symlinkat(struct io_kiocb *req, unsigned int issue_flags)
+{
+	struct io_link *sl = io_kiocb_to_cmd(req, struct io_link);
+	CLASS(filename_complete_delayed, old)(&sl->oldpath);
+	CLASS(filename_complete_delayed, new)(&sl->newpath);
+	int ret;
+
+	WARN_ON_ONCE(issue_flags & IO_URING_F_NONBLOCK);
+
+	ret = filename_symlinkat(old, sl->new_dfd, new);
+
+	req->flags &= ~REQ_F_NEED_CLEANUP;
+	io_req_set_res(req, ret, 0);
+	return IOU_COMPLETE;
+}
+
+int io_linkat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
+{
+	struct io_link *lnk = io_kiocb_to_cmd(req, struct io_link);
+	const char __user *oldf, *newf;
+	int err;
+
+	if (sqe->buf_index || sqe->splice_fd_in)
+		return -EINVAL;
+	if (unlikely(req->flags & REQ_F_FIXED_FILE))
+		return -EBADF;
+
+	lnk->old_dfd = READ_ONCE(sqe->fd);
+	lnk->new_dfd = READ_ONCE(sqe->len);
+	oldf = u64_to_user_ptr(READ_ONCE(sqe->addr));
+	newf = u64_to_user_ptr(READ_ONCE(sqe->addr2));
+	lnk->flags = READ_ONCE(sqe->hardlink_flags);
+
+	err = delayed_getname_uflags(&lnk->oldpath, oldf, lnk->flags);
+	if (unlikely(err))
+		return err;
+
+	err = delayed_getname(&lnk->newpath, newf);
+	if (unlikely(err)) {
+		dismiss_delayed_filename(&lnk->oldpath);
+		return err;
+	}
+
+	req->flags |= REQ_F_NEED_CLEANUP;
+	req->flags |= REQ_F_FORCE_ASYNC;
+	return 0;
+}
+
+int io_linkat(struct io_kiocb *req, unsigned int issue_flags)
+{
+	struct io_link *lnk = io_kiocb_to_cmd(req, struct io_link);
+	CLASS(filename_complete_delayed, old)(&lnk->oldpath);
+	CLASS(filename_complete_delayed, new)(&lnk->newpath);
+	int ret;
+
+	WARN_ON_ONCE(issue_flags & IO_URING_F_NONBLOCK);
+
+	ret = filename_linkat(lnk->old_dfd, old, lnk->new_dfd, new, lnk->flags);
+
+	req->flags &= ~REQ_F_NEED_CLEANUP;
+	io_req_set_res(req, ret, 0);
+	return IOU_COMPLETE;
+}
+
+void io_link_cleanup(struct io_kiocb *req)
+{
+	struct io_link *sl = io_kiocb_to_cmd(req, struct io_link);
+
+	dismiss_delayed_filename(&sl->oldpath);
+	dismiss_delayed_filename(&sl->newpath);
+}