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, 82 insertions, 0 deletions

diff --git a/Documentation/fb/s3fb.rst b/Documentation/fb/s3fb.rst
new file mode 100644
index 000000000..e809d69c2
--- /dev/null
+++ b/Documentation/fb/s3fb.rst
@@ -0,0 +1,82 @@
+===========================================
+s3fb - fbdev driver for S3 Trio/Virge chips
+===========================================
+
+
+Supported Hardware
+==================
+
+	S3 Trio32
+	S3 Trio64 (and variants V+, UV+, V2/DX, V2/GX)
+	S3 Virge  (and variants VX, DX, GX and GX2+)
+	S3 Plato/PX		(completely untested)
+	S3 Aurora64V+		(completely untested)
+
+	- only PCI bus supported
+	- only BIOS initialized VGA devices supported
+	- probably not working on big endian
+
+I tested s3fb on Trio64 (plain, V+ and V2/DX) and Virge (plain, VX, DX),
+all on i386.
+
+
+Supported Features
+==================
+
+	*  4 bpp pseudocolor modes (with 18bit palette, two variants)
+	*  8 bpp pseudocolor mode (with 18bit palette)
+	* 16 bpp truecolor modes (RGB 555 and RGB 565)
+	* 24 bpp truecolor mode (RGB 888) on (only on Virge VX)
+	* 32 bpp truecolor mode (RGB 888) on (not on Virge VX)
+	* text mode (activated by bpp = 0)
+	* interlaced mode variant (not available in text mode)
+	* doublescan mode variant (not available in text mode)
+	* panning in both directions
+	* suspend/resume support
+	* DPMS support
+
+Text mode is supported even in higher resolutions, but there is limitation to
+lower pixclocks (maximum usually between 50-60 MHz, depending on specific
+hardware, i get best results from plain S3 Trio32 card - about 75 MHz). This
+limitation is not enforced by driver. Text mode supports 8bit wide fonts only
+(hardware limitation) and 16bit tall fonts (driver limitation). Text mode
+support is broken on S3 Trio64 V2/DX.
+
+There are two 4 bpp modes. First mode (selected if nonstd == 0) is mode with
+packed pixels, high nibble first. Second mode (selected if nonstd == 1) is mode
+with interleaved planes (1 byte interleave), MSB first. Both modes support
+8bit wide fonts only (driver limitation).
+
+Suspend/resume works on systems that initialize video card during resume and
+if device is active (for example used by fbcon).
+
+
+Missing Features
+================
+(alias TODO list)
+
+	* secondary (not initialized by BIOS) device support
+	* big endian support
+	* Zorro bus support
+	* MMIO support
+	* 24 bpp mode support on more cards
+	* support for fontwidths != 8 in 4 bpp modes
+	* support for fontheight != 16 in text mode
+	* composite and external sync (is anyone able to test this?)
+	* hardware cursor
+	* video overlay support
+	* vsync synchronization
+	* feature connector support
+	* acceleration support (8514-like 2D, Virge 3D, busmaster transfers)
+	* better values for some magic registers (performance issues)
+
+
+Known bugs
+==========
+
+	* cursor disable in text mode doesn't work
+	* text mode broken on S3 Trio64 V2/DX
+
+
+--
+Ondrej Zajicek <santiago@crfreenet.org>