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/admin-guide/sysctl/user.rst | 84 +++++++++++++++++++++++++++++++ 1 file changed, 84 insertions(+) create mode 100644 Documentation/admin-guide/sysctl/user.rst (limited to 'Documentation/admin-guide/sysctl/user.rst') diff --git a/Documentation/admin-guide/sysctl/user.rst b/Documentation/admin-guide/sysctl/user.rst new file mode 100644 index 000000000..c45824589 --- /dev/null +++ b/Documentation/admin-guide/sysctl/user.rst @@ -0,0 +1,84 @@ +================================= +Documentation for /proc/sys/user/ +================================= + +kernel version 4.9.0 + +Copyright (c) 2016 Eric Biederman + +------------------------------------------------------------------------------ + +This file contains the documentation for the sysctl files in +/proc/sys/user. + +The files in this directory can be used to override the default +limits on the number of namespaces and other objects that have +per user per user namespace limits. + +The primary purpose of these limits is to stop programs that +malfunction and attempt to create a ridiculous number of objects, +before the malfunction becomes a system wide problem. It is the +intention that the defaults of these limits are set high enough that +no program in normal operation should run into these limits. + +The creation of per user per user namespace objects are charged to +the user in the user namespace who created the object and +verified to be below the per user limit in that user namespace. + +The creation of objects is also charged to all of the users +who created user namespaces the creation of the object happens +in (user namespaces can be nested) and verified to be below the per user +limits in the user namespaces of those users. + +This recursive counting of created objects ensures that creating a +user namespace does not allow a user to escape their current limits. + +Currently, these files are in /proc/sys/user: + +max_cgroup_namespaces +===================== + + The maximum number of cgroup namespaces that any user in the current + user namespace may create. + +max_ipc_namespaces +================== + + The maximum number of ipc namespaces that any user in the current + user namespace may create. + +max_mnt_namespaces +================== + + The maximum number of mount namespaces that any user in the current + user namespace may create. + +max_net_namespaces +================== + + The maximum number of network namespaces that any user in the + current user namespace may create. + +max_pid_namespaces +================== + + The maximum number of pid namespaces that any user in the current + user namespace may create. + +max_time_namespaces +=================== + + The maximum number of time namespaces that any user in the current + user namespace may create. + +max_user_namespaces +=================== + + The maximum number of user namespaces that any user in the current + user namespace may create. + +max_uts_namespaces +================== + + The maximum number of user namespaces that any user in the current + user namespace may create. -- cgit v1.3.1