From 0d6526f82c3cdefcca47f73f5fc08dc6f335eac6 Mon Sep 17 00:00:00 2001 From: Tim Chen Date: Mon, 21 Sep 2026 17:37:22 -0700 Subject: sched/cache: Keep nr_pref_llc_running in the runnable domain, to fix LLC mis-scheduling bug 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 Suggested-by: Chen Yu Signed-off-by: Tim Chen Signed-off-by: Peter Zijlstra (Intel) Signed-off-by: Ingo Molnar Reviewed-by: Kayra Cizmeci Cc: # v7.2.x Link: https://patch.msgid.link/06af61afedac32e6477f57feb4d658f6c411c3af.1790035273.git.tim.c.chen@linux.intel.com --- Documentation/gpu/drm-ras.rst | 113 ++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 113 insertions(+) create mode 100644 Documentation/gpu/drm-ras.rst (limited to 'Documentation/gpu/drm-ras.rst') diff --git a/Documentation/gpu/drm-ras.rst b/Documentation/gpu/drm-ras.rst new file mode 100644 index 000000000..83c21853b --- /dev/null +++ b/Documentation/gpu/drm-ras.rst @@ -0,0 +1,113 @@ +.. SPDX-License-Identifier: GPL-2.0+ + +============================ +DRM RAS over Generic Netlink +============================ + +The DRM RAS (Reliability, Availability, Serviceability) interface provides a +standardized way for GPU/accelerator drivers to expose error counters and +other reliability nodes to user space via Generic Netlink. This allows +diagnostic tools, monitoring daemons, or test infrastructure to query hardware +health in a uniform way across different DRM drivers. + +Key Goals: + +* Provide a standardized RAS solution for GPU and accelerator drivers, enabling + data center monitoring and reliability operations. +* Implement a single drm-ras Generic Netlink family to meet modern Netlink YAML + specifications and centralize all RAS-related communication in one namespace. +* Support a basic error counter interface, addressing the immediate, essential + monitoring needs. +* Offer a flexible, future-proof interface that can be extended to support + additional types of RAS data in the future. +* Allow multiple nodes per driver, enabling drivers to register separate + nodes for different IP blocks, sub-blocks, or other logical subdivisions + as applicable. + +.. contents:: + +Nodes +===== + +Nodes are logical abstractions representing an error type or error source within +the device. Currently, only error counter nodes is supported. + +Drivers are responsible for registering and unregistering nodes via the +`drm_ras_node_register()` and `drm_ras_node_unregister()` APIs. + +Node Management +------------------- + +.. kernel-doc:: drivers/gpu/drm/drm_ras.c + :doc: DRM RAS Node Management +.. kernel-doc:: drivers/gpu/drm/drm_ras.c + :internal: + +Generic Netlink Usage +===================== + +The interface is implemented as a Generic Netlink family named ``drm-ras``. +User space tools can: + +* List registered nodes with the ``list-nodes`` command. +* List all error counters in an node with the ``get-error-counter`` command with ``node-id`` + as a parameter. +* Query specific error counter values with the ``get-error-counter`` command, using both + ``node-id`` and ``error-id`` as parameters. +* Clear specific error counters with the ``clear-error-counter`` command, using both + ``node-id`` and ``error-id`` as parameters. + +YAML-based Interface +-------------------- + +The interface is described in a YAML specification ``Documentation/netlink/specs/drm_ras.yaml`` + +This YAML is used to auto-generate user space bindings via +``tools/net/ynl/pyynl/ynl_gen_c.py``, and drives the structure of netlink +attributes and operations. + +Usage Notes +----------- + +* User space must first enumerate nodes to obtain their IDs. +* Node IDs or Node names can be used for all further queries, such as error counters. +* Error counters can be queried by either the Error ID or Error name. +* Query Parameters should be defined as part of the uAPI to ensure user interface stability. +* The interface supports future extension by adding new node types and + additional attributes. + +Example: List nodes using ynl + +.. code-block:: bash + + sudo ynl --family drm_ras --dump list-nodes + [{'device-name': '0000:03:00.0', + 'node-id': 0, + 'node-name': 'correctable-errors', + 'node-type': 'error-counter'}, + {'device-name': '0000:03:00.0', + 'node-id': 1, + 'node-name': 'uncorrectable-errors', + 'node-type': 'error-counter'}] + +Example: List all error counters using ynl + +.. code-block:: bash + + sudo ynl --family drm_ras --dump get-error-counter --json '{"node-id":0}' + [{'error-id': 1, 'error-name': 'error_name1', 'error-value': 0}, + {'error-id': 2, 'error-name': 'error_name2', 'error-value': 0}] + +Example: Query an error counter for a given node + +.. code-block:: bash + + sudo ynl --family drm_ras --do get-error-counter --json '{"node-id":0, "error-id":1}' + {'error-id': 1, 'error-name': 'error_name1', 'error-value': 0} + +Example: Clear an error counter for a given node + +.. code-block:: bash + + sudo ynl --family drm_ras --do clear-error-counter --json '{"node-id":0, "error-id":1}' + None -- cgit v1.3.1