diff options
| author | Tim Chen <tim.c.chen@linux.intel.com> | 2026-09-21 17:37:22 -0700 |
|---|---|---|
| committer | Ingo Molnar <mingo@kernel.org> | 2026-09-22 10:49:53 +0200 |
| commit | 0d6526f82c3cdefcca47f73f5fc08dc6f335eac6 (patch) | |
| tree | 176dc2d685debcfb082c151ee94f2425ac96b364 /Documentation/userspace-api/netlink/c-code-gen.rst | |
| download | linux-stable-0d6526f82c3cdefcca47f73f5fc08dc6f335eac6.tar.gz linux-stable-0d6526f82c3cdefcca47f73f5fc08dc6f335eac6.zip | |
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
Diffstat (limited to 'Documentation/userspace-api/netlink/c-code-gen.rst')
| -rw-r--r-- | Documentation/userspace-api/netlink/c-code-gen.rst | 109 |
1 files changed, 109 insertions, 0 deletions
diff --git a/Documentation/userspace-api/netlink/c-code-gen.rst b/Documentation/userspace-api/netlink/c-code-gen.rst new file mode 100644 index 000000000..46415e6d6 --- /dev/null +++ b/Documentation/userspace-api/netlink/c-code-gen.rst @@ -0,0 +1,109 @@ +.. SPDX-License-Identifier: BSD-3-Clause + +============================== +Netlink spec C code generation +============================== + +This document describes how Netlink specifications are used to render +C code (uAPI, policies etc.). It also defines the additional properties +allowed in older families by the ``genetlink-c`` protocol level, +to control the naming. + +For brevity this document refers to ``name`` properties of various +objects by the object type. For example ``$attr`` is the value +of ``name`` in an attribute, and ``$family`` is the name of the +family (the global ``name`` property). + +The upper case is used to denote literal values, e.g. ``$family-CMD`` +means the concatenation of ``$family``, a dash character, and the literal +``CMD``. + +The names of ``#defines`` and enum values are always converted to upper case, +and with dashes (``-``) replaced by underscores (``_``). + +If the constructed name is a C keyword, an extra underscore is +appended (``do`` -> ``do_``). + +Globals +======= + +``c-family-name`` controls the name of the ``#define`` for the family +name, default is ``$family-FAMILY-NAME``. + +``c-version-name`` controls the name of the ``#define`` for the version +of the family, default is ``$family-FAMILY-VERSION``. + +``max-by-define`` selects if max values for enums are defined as a +``#define`` rather than inside the enum. + +Definitions +=========== + +Constants +--------- + +Every constant is rendered as a ``#define``. +The name of the constant is ``$family-$constant`` and the value +is rendered as a string or integer according to its type in the spec. + +Enums and flags +--------------- + +Enums are named ``$family-$enum``. The full name can be set directly +or suppressed by specifying the ``enum-name`` property. +Default entry name is ``$family-$enum-$entry``. +If ``name-prefix`` is specified it replaces the ``$family-$enum`` +portion of the entry name. + +Boolean ``render-max`` controls creation of the max values +(which are enabled by default for attribute enums). These max +values are named ``__$pfx-MAX`` and ``$pfx-MAX``. The name +of the first value can be overridden via ``enum-cnt-name`` property. + +Attributes +========== + +Each attribute set (excluding fractional sets) is rendered as an enum. + +Attribute enums are traditionally unnamed in netlink headers. +If naming is desired ``enum-name`` can be used to specify the name. + +The default attribute name prefix is ``$family-A`` if the name of the set +is the same as the name of the family and ``$family-A-$set`` if the names +differ. The prefix can be overridden by the ``name-prefix`` property of a set. +The rest of the section will refer to the prefix as ``$pfx``. + +Attributes are named ``$pfx-$attribute``. + +Attribute enums end with two special values ``__$pfx-MAX`` and ``$pfx-MAX`` +which are used for sizing attribute tables. +These two names can be specified directly with the ``attr-cnt-name`` +and ``attr-max-name`` properties respectively. + +If ``max-by-define`` is set to ``true`` at the global level ``attr-max-name`` +will be specified as a ``#define`` rather than an enum value. + +Operations +========== + +Operations are named ``$family-CMD-$operation``. +If ``name-prefix`` is specified it replaces the ``$family-CMD`` +portion of the name. + +Similarly to attribute enums operation enums end with special count and max +attributes. For operations those attributes can be renamed with +``cmd-cnt-name`` and ``cmd-max-name``. Max will be a define if ``max-by-define`` +is ``true``. + +Multicast groups +================ + +Each multicast group gets a define rendered into the kernel uAPI header. +The name of the define is ``$family-MCGRP-$group``, and can be overwritten +with the ``c-define-name`` property. + +Code generation +=============== + +uAPI header is assumed to come from ``<linux/$family.h>`` in the default header +search path. It can be changed using the ``uapi-header`` global property. |
