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, 78 insertions, 0 deletions

diff --git a/scripts/rustdoc_test_builder.rs b/scripts/rustdoc_test_builder.rs
new file mode 100644
index 000000000..df864437c
--- /dev/null
+++ b/scripts/rustdoc_test_builder.rs
@@ -0,0 +1,78 @@
+// SPDX-License-Identifier: GPL-2.0
+
+//! Test builder for `rustdoc`-generated tests.
+//!
+//! This script is a hack to extract the test from `rustdoc`'s output. Ideally, `rustdoc` would
+//! have an option to generate this information instead, e.g. as JSON output.
+//!
+//! The `rustdoc`-generated test names look like `{file}_{line}_{number}`, e.g.
+//! `...path_rust_kernel_sync_arc_rs_42_0`. `number` is the "test number", needed in cases like
+//! a macro that expands into items with doctests is invoked several times within the same line.
+//!
+//! However, since these names are used for bisection in CI, the line number makes it not stable
+//! at all. In the future, we would like `rustdoc` to give us the Rust item path associated with
+//! the test, plus a "test number" (for cases with several examples per item) and generate a name
+//! from that. For the moment, we generate ourselves a new name, `{file}_{number}` instead, in
+//! the `gen` script (done there since we need to be aware of all the tests in a given file).
+
+use std::io::Read;
+
+fn main() {
+    let mut stdin = std::io::stdin().lock();
+    let mut body = String::new();
+    stdin.read_to_string(&mut body).unwrap();
+
+    // Find the generated function name looking for the inner function inside `main()`.
+    //
+    // The line we are looking for looks like one of the following:
+    //
+    // ```
+    // fn main() { #[allow(non_snake_case)] fn _doctest_main_rust_kernel_file_rs_28_0() {
+    // fn main() { #[allow(non_snake_case)] fn _doctest_main_rust_kernel_file_rs_37_0() -> Result<(), impl core::fmt::Debug> {
+    // ```
+    //
+    // It should be unlikely that doctest code matches such lines (when code is formatted properly).
+    let rustdoc_function_name = body
+        .lines()
+        .find_map(|line| {
+            Some(
+                line.split_once("fn main() {")?
+                    .1
+                    .split_once("fn ")?
+                    .1
+                    .split_once("()")?
+                    .0,
+            )
+            .filter(|x| x.chars().all(|c| c.is_alphanumeric() || c == '_'))
+        })
+        .expect("No test function found in `rustdoc`'s output.");
+
+    // Replicate `rustdoc` 1.87+ behaviour [1] by fully qualifying `Result` to avoid the collision
+    // with our own `Result` coming from the prelude.
+    //
+    // [1]: https://github.com/rust-lang/rust/pull/137807
+    //
+    // TODO: Remove this when MSRV is bumped above 1.87.
+    let body = body.replace(
+        &format!("{rustdoc_function_name}() -> Result<(), impl core::fmt::Debug> {{"),
+        // This intentionally does not use absolute paths to match `rustdoc` 1.87 behaviour.
+        &format!("{rustdoc_function_name}() -> core::result::Result<(), impl core::fmt::Debug> {{"),
+    );
+
+    // For tests that get generated with `Result`, like above, `rustdoc` generates an `unwrap()` on
+    // the return value to check there were no returned errors. Instead, we use our assert macro
+    // since we want to just fail the test, not panic the kernel.
+    //
+    // We save the result in a variable so that the failed assertion message looks nicer.
+    let body = body.replace(
+        &format!("}} {rustdoc_function_name}().unwrap() }}"),
+        &format!("}} let test_return_value = {rustdoc_function_name}(); assert!(test_return_value.is_ok()); }}"),
+    );
+
+    // Figure out a smaller test name based on the generated function name.
+    let name = rustdoc_function_name.split_once("_rust_kernel_").unwrap().1;
+
+    let path = format!("rust/test/doctests/kernel/{name}");
+
+    std::fs::write(path, body.as_bytes()).unwrap();
+}