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, 115 insertions, 0 deletions

diff --git a/tools/testing/memblock/README b/tools/testing/memblock/README
new file mode 100644
index 000000000..b435f48d8
--- /dev/null
+++ b/tools/testing/memblock/README
@@ -0,0 +1,115 @@
+==================
+Memblock simulator
+==================
+
+Introduction
+============
+
+Memblock is a boot time memory allocator[1] that manages memory regions before
+the actual memory management is initialized. Its APIs allow to register physical
+memory regions, mark them as available or reserved, allocate a block of memory
+within the requested range and/or in specific NUMA node, and many more.
+
+Because it is used so early in the booting process, testing and debugging it is
+difficult. This test suite, usually referred as memblock simulator, is
+an attempt at testing the memblock mechanism. It runs one monolithic test that
+consist of a series of checks that exercise both the basic operations and
+allocation functionalities of memblock. The main data structure of the boot time
+memory allocator is initialized at the build time, so the checks here reuse its
+instance throughout the duration of the test. To ensure that tests don't affect
+each other, region arrays are reset in between.
+
+As this project uses the actual memblock code and has to run in user space,
+some of the kernel definitions were stubbed by the initial commit that
+introduced memblock simulator (commit 16802e55dea9 ("memblock tests: Add
+skeleton of the memblock simulator")) and a few preparation commits just
+before it. Most of them don't match the kernel implementation, so one should
+consult them first before making any significant changes to the project.
+
+Usage
+=====
+
+To run the tests, build the main target and run it:
+
+$ make && ./main
+
+A successful run produces no output. It is possible to control the behavior
+by passing options from command line. For example, to include verbose output,
+append the `-v` options when you run the tests:
+
+$ ./main -v
+
+This will print information about which functions are being tested and the
+number of test cases that passed.
+
+For the full list of options from command line, see `./main --help`.
+
+It is also possible to override different configuration parameters to change
+the test functions. For example, to simulate enabled NUMA, use:
+
+$ make NUMA=1
+
+For the full list of build options, see `make help`.
+
+Project structure
+=================
+
+The project has one target, main, which calls a group of checks for basic and
+allocation functions. Tests for each group are defined in dedicated files, as it
+can be seen here:
+
+memblock
+|-- asm       ------------------,
+|-- lib                         |-- implement function and struct stubs
+|-- linux     ------------------'
+|-- scripts
+|    |-- Makefile.include        -- handles `make` parameters
+|-- tests
+|    |-- alloc_api.(c|h)         -- memblock_alloc tests
+|    |-- alloc_helpers_api.(c|h) -- memblock_alloc_from tests
+|    |-- alloc_nid_api.(c|h)     -- memblock_alloc_try_nid tests
+|    |-- basic_api.(c|h)         -- memblock_add/memblock_reserve/... tests
+|    |-- common.(c|h)            -- helper functions for resetting memblock;
+|-- main.c        --------------.   dummy physical memory definition
+|-- Makefile                     `- test runner
+|-- README
+|-- TODO
+|-- .gitignore
+
+Simulating physical memory
+==========================
+
+Some allocation functions clear the memory in the process, so it is required for
+memblock to track valid memory ranges. To achieve this, the test suite registers
+with memblock memory stored by test_memory struct. It is a small wrapper that
+points to a block of memory allocated via malloc. For each group of allocation
+tests, dummy physical memory is allocated, added to memblock, and then released
+at the end of the test run. The structure of a test runner checking allocation
+functions is as follows:
+
+int memblock_alloc_foo_checks(void)
+{
+	reset_memblock_attributes();     /* data structure reset */
+	dummy_physical_memory_init();    /* allocate and register memory */
+
+	(...allocation checks...)
+
+	dummy_physical_memory_cleanup(); /* free the memory */
+}
+
+There's no need to explicitly free the dummy memory from memblock via
+memblock_free() call. The entry will be erased by reset_memblock_regions(),
+called at the beginning of each test.
+
+Known issues
+============
+
+1. Tests for memblock_alloc_low() can't be easily implemented. The function uses
+   ARCH_LOW_ADDRESS_LIMIT marco, which can't be changed to point at the low
+   memory of the memory_block.
+
+References
+==========
+
+1. Boot time memory management documentation page:
+   https://www.kernel.org/doc/html/latest/core-api/boot-time-mm.html