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, 94 insertions, 0 deletions

diff --git a/Documentation/gpu/amdgpu/ptl.rst b/Documentation/gpu/amdgpu/ptl.rst
new file mode 100644
index 000000000..c7f16dea7
--- /dev/null
+++ b/Documentation/gpu/amdgpu/ptl.rst
@@ -0,0 +1,94 @@
+=======================================
+Peak Tops Limiter (PTL) sysfs Interface
+=======================================
+
+Overview
+--------
+The Peak Tops Limiter (PTL) sysfs interface enables users to control and
+configure the PTL feature for each GPU individually.  All PTL-related
+sysfs files are located under `/sys/class/drm/cardX/device/ptl/`, where
+`X` is the GPU index.  Through these files, users can enable or disable
+PTL, set preferred data formats, and query supported formats for each GPU.
+
+PTL sysfs files
+----------------
+The following files are available under `/sys/class/drm/cardX/device/ptl/`:
+
+- `ptl_enable`
+- `ptl_format`
+- `ptl_supported_formats`
+
+PTL Enable/Disable
+------------------
+File: `ptl_enable`
+Type: Read/Write (rw)
+
+Read: Returns the current PTL status as a string: `enabled` if PTL
+is active, or `disabled` if inactive.
+
+Write:
+
+- Write `1` or `enabled` to enable PTL
+- Write `0` or `disabled` to disable PTL
+
+Examples::
+
+    # Query PTL status
+    cat /sys/class/drm/card1/device/ptl/ptl_enable
+    # Output: enabled
+
+    # Enable PTL
+    sudo bash -c "echo 1 > /sys/class/drm/card1/device/ptl/ptl_enable"
+
+    # Disable PTL
+    sudo bash -c "echo 0 > /sys/class/drm/card1/device/ptl/ptl_enable"
+
+PTL Format (Preferred Data Formats)
+-----------------------------------
+File: `ptl_format`
+Type: Read/Write (rw)
+
+Read: Returns the two preferred formats, e.g. `I8,F32`.
+
+Write: Accepts two formats separated by a comma, e.g. `I8,F32`.
+
+- Both formats must be supported and different.
+- If an invalid format is provided (not supported, or both formats are the
+  same), the driver will return "write error: Invalid argument".
+
+Examples::
+
+    # Query PTL formats
+    cat /sys/class/drm/card1/device/ptl/ptl_format
+    # Output: I8,F32
+
+    # Set PTL formats
+    sudo bash -c "echo I8,F32 > /sys/class/drm/card1/device/ptl/ptl_format"
+
+Supported Formats
+-----------------
+File: `ptl_supported_formats`
+Type: Read-only (r)
+
+Read: Returns a comma-separated list of supported formats, e.g.
+`I8,F16,BF16,F32,F64`.
+
+Example::
+
+    # Check supported formats
+    cat /sys/class/drm/card1/device/ptl/ptl_supported_formats
+    # Output: I8,F16,BF16,F32,F64
+
+Behavioral Notes
+----------------
+- PTL formats can only be set when PTL is enabled.
+- If PTL is disabled, `ptl_format` returns `N/A`.
+- Only two formats can be set at a time, and they must be from the supported set and different..
+- All commands support per-GPU targeting.
+- Root permission is required to enable/disable PTL or change formats.
+- If the hardware does not support PTL, the PTL sysfs directory will not
+  be created.
+
+Implementation
+--------------
+The PTL sysfs nodes are implemented in `drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c`.