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, 56 insertions, 0 deletions

diff --git a/Documentation/fb/cmap_xfbdev.rst b/Documentation/fb/cmap_xfbdev.rst
new file mode 100644
index 000000000..5db5e9787
--- /dev/null
+++ b/Documentation/fb/cmap_xfbdev.rst
@@ -0,0 +1,56 @@
+==========================
+Understanding fbdev's cmap
+==========================
+
+These notes explain how X's dix layer uses fbdev's cmap structures.
+
+-  example of relevant structures in fbdev as used for a 3-bit grayscale cmap::
+
+    struct fb_var_screeninfo {
+	    .bits_per_pixel = 8,
+	    .grayscale      = 1,
+	    .red =          { 4, 3, 0 },
+	    .green =        { 0, 0, 0 },
+	    .blue =         { 0, 0, 0 },
+    }
+    struct fb_fix_screeninfo {
+	    .visual =       FB_VISUAL_STATIC_PSEUDOCOLOR,
+    }
+    for (i = 0; i < 8; i++)
+	info->cmap.red[i] = (((2*i)+1)*(0xFFFF))/16;
+    memcpy(info->cmap.green, info->cmap.red, sizeof(u16)*8);
+    memcpy(info->cmap.blue, info->cmap.red, sizeof(u16)*8);
+
+-  X11 apps do something like the following when trying to use grayscale::
+
+    for (i=0; i < 8; i++) {
+	char colorspec[64];
+	memset(colorspec,0,64);
+	sprintf(colorspec, "rgb:%x/%x/%x", i*36,i*36,i*36);
+	if (!XParseColor(outputDisplay, testColormap, colorspec, &wantedColor))
+		printf("Can't get color %s\n",colorspec);
+	XAllocColor(outputDisplay, testColormap, &wantedColor);
+	grays[i] = wantedColor;
+    }
+
+There's also named equivalents like gray1..x provided you have an rgb.txt.
+
+Somewhere in X's callchain, this results in a call to X code that handles the
+colormap. For example, Xfbdev hits the following:
+
+xc-011010/programs/Xserver/dix/colormap.c::
+
+  FindBestPixel(pentFirst, size, prgb, channel)
+
+  dr = (long) pent->co.local.red - prgb->red;
+  dg = (long) pent->co.local.green - prgb->green;
+  db = (long) pent->co.local.blue - prgb->blue;
+  sq = dr * dr;
+  UnsignedToBigNum (sq, &sum);
+  BigNumAdd (&sum, &temp, &sum);
+
+co.local.red are entries that were brought in through FBIOGETCMAP which come
+directly from the info->cmap.red that was listed above. The prgb is the rgb
+that the app wants to match to. The above code is doing what looks like a least
+squares matching function. That's why the cmap entries can't be set to the left
+hand side boundaries of a color range.