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 --- net/mac80211/pm.c | 202 ++++++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 202 insertions(+) create mode 100644 net/mac80211/pm.c (limited to 'net/mac80211/pm.c') diff --git a/net/mac80211/pm.c b/net/mac80211/pm.c new file mode 100644 index 000000000..f63676c44 --- /dev/null +++ b/net/mac80211/pm.c @@ -0,0 +1,202 @@ +// SPDX-License-Identifier: GPL-2.0 +/* + * Portions + * Copyright (C) 2020-2021, 2023-2024 Intel Corporation + */ +#include +#include + +#include "ieee80211_i.h" +#include "mesh.h" +#include "driver-ops.h" +#include "led.h" + +static void ieee80211_sched_scan_cancel(struct ieee80211_local *local) +{ + if (ieee80211_request_sched_scan_stop(local)) + return; + cfg80211_sched_scan_stopped_locked(local->hw.wiphy, 0); +} + +int __ieee80211_suspend(struct ieee80211_hw *hw, struct cfg80211_wowlan *wowlan, + bool reset) +{ + struct ieee80211_local *local = hw_to_local(hw); + struct ieee80211_sub_if_data *sdata; + struct sta_info *sta; + + if (!local->open_count) + goto suspend; + + local->suspending = true; + mb(); /* make suspending visible before any cancellation */ + + ieee80211_scan_cancel(local); + + ieee80211_dfs_cac_cancel(local, NULL); + + ieee80211_roc_purge(local, NULL); + + ieee80211_del_virtual_monitor(local); + + if (ieee80211_hw_check(hw, AMPDU_AGGREGATION) && + !(wowlan && wowlan->any)) { + lockdep_assert_wiphy(local->hw.wiphy); + list_for_each_entry(sta, &local->sta_list, list) { + set_sta_flag(sta, WLAN_STA_BLOCK_BA); + ieee80211_sta_tear_down_BA_sessions( + sta, AGG_STOP_LOCAL_REQUEST); + } + } + + /* keep sched_scan only in case of 'any' trigger */ + if (!(wowlan && wowlan->any)) + ieee80211_sched_scan_cancel(local); + + ieee80211_stop_queues_by_reason(hw, + IEEE80211_MAX_QUEUE_MAP, + IEEE80211_QUEUE_STOP_REASON_SUSPEND, + false); + + /* flush out all packets */ + synchronize_net(); + + ieee80211_flush_queues(local, NULL, true); + + local->quiescing = true; + /* make quiescing visible to timers everywhere */ + mb(); + + flush_workqueue(local->workqueue); + + /* Don't try to run timers while suspended. */ + timer_delete_sync(&local->sta_cleanup); + + /* + * Note that this particular timer doesn't need to be + * restarted at resume. + */ + wiphy_work_cancel(local->hw.wiphy, &local->dynamic_ps_enable_work); + timer_delete_sync(&local->dynamic_ps_timer); + + local->wowlan = wowlan; + if (local->wowlan) { + int err; + + /* Drivers don't expect to suspend while some operations like + * authenticating or associating are in progress. It doesn't + * make sense anyway to accept that, since the authentication + * or association would never finish since the driver can't do + * that on its own. + * Thus, clean up in-progress auth/assoc first. + */ + list_for_each_entry(sdata, &local->interfaces, list) { + if (!ieee80211_sdata_running(sdata)) + continue; + if (sdata->vif.type != NL80211_IFTYPE_STATION) + continue; + ieee80211_mgd_quiesce(sdata); + /* If suspended during TX in progress, and wowlan + * is enabled (connection will be active) there + * can be a race where the driver is put out + * of power-save due to TX and during suspend + * dynamic_ps_timer is cancelled and TX packet + * is flushed, leaving the driver in ACTIVE even + * after resuming until dynamic_ps_timer puts + * driver back in DOZE. + */ + if (sdata->u.mgd.associated && + sdata->u.mgd.powersave && + !(local->hw.conf.flags & IEEE80211_CONF_PS)) { + local->hw.conf.flags |= IEEE80211_CONF_PS; + ieee80211_hw_config(local, -1, + IEEE80211_CONF_CHANGE_PS); + } + } + + err = drv_suspend(local, wowlan); + if (err < 0) { + local->quiescing = false; + local->wowlan = false; + if (ieee80211_hw_check(hw, AMPDU_AGGREGATION)) { + lockdep_assert_wiphy(local->hw.wiphy); + list_for_each_entry(sta, + &local->sta_list, list) { + clear_sta_flag(sta, WLAN_STA_BLOCK_BA); + } + } + ieee80211_wake_queues_by_reason(hw, + IEEE80211_MAX_QUEUE_MAP, + IEEE80211_QUEUE_STOP_REASON_SUSPEND, + false); + return err; + } else if (err > 0) { + WARN_ON(err != 1); + /* cfg80211 will call back into mac80211 to disconnect + * all interfaces, allow that to proceed properly + */ + ieee80211_wake_queues_by_reason(hw, + IEEE80211_MAX_QUEUE_MAP, + IEEE80211_QUEUE_STOP_REASON_SUSPEND, + false); + return err; + } else { + goto suspend; + } + } + + /* remove all interfaces that were created in the driver */ + list_for_each_entry(sdata, &local->interfaces, list) { + if (!ieee80211_sdata_running(sdata)) + continue; + switch (sdata->vif.type) { + case NL80211_IFTYPE_AP_VLAN: + case NL80211_IFTYPE_MONITOR: + continue; + case NL80211_IFTYPE_STATION: + ieee80211_mgd_quiesce(sdata); + break; + default: + break; + } + + wiphy_delayed_work_flush(local->hw.wiphy, + &sdata->dec_tailroom_needed_wk); + drv_remove_interface(local, sdata); + } + + /* + * We disconnected on all interfaces before suspend, all channel + * contexts should be released, but on 'reset' debugfs that's + * not true so don't check there. + */ + WARN_ON(!reset && !list_empty(&local->chanctx_list)); + + /* stop hardware - this must stop RX */ + ieee80211_stop_device(local, true); + + suspend: + local->suspended = true; + /* need suspended to be visible before quiescing is false */ + barrier(); + local->quiescing = false; + local->suspending = false; + + return 0; +} + +/* + * __ieee80211_resume() is a static inline which just calls + * ieee80211_reconfig(), which is also needed for hardware + * hang/firmware failure/etc. recovery. + */ + +void ieee80211_report_wowlan_wakeup(struct ieee80211_vif *vif, + struct cfg80211_wowlan_wakeup *wakeup, + gfp_t gfp) +{ + struct ieee80211_sub_if_data *sdata = vif_to_sdata(vif); + + cfg80211_report_wowlan_wakeup(&sdata->wdev, wakeup, gfp); +} +EXPORT_SYMBOL(ieee80211_report_wowlan_wakeup); -- cgit v1.3.1