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 --- drivers/net/phy/linkmode.c | 95 ++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 95 insertions(+) create mode 100644 drivers/net/phy/linkmode.c (limited to 'drivers/net/phy/linkmode.c') diff --git a/drivers/net/phy/linkmode.c b/drivers/net/phy/linkmode.c new file mode 100644 index 000000000..f60560fe3 --- /dev/null +++ b/drivers/net/phy/linkmode.c @@ -0,0 +1,95 @@ +// SPDX-License-Identifier: GPL-2.0+ +#include + +/** + * linkmode_resolve_pause - resolve the allowable pause modes + * @local_adv: local advertisement in ethtool format + * @partner_adv: partner advertisement in ethtool format + * @tx_pause: pointer to bool to indicate whether transmit pause should be + * enabled. + * @rx_pause: pointer to bool to indicate whether receive pause should be + * enabled. + * + * Flow control is resolved according to our and the link partners + * advertisements using the following drawn from the 802.3 specs: + * Local device Link partner + * Pause AsymDir Pause AsymDir Result + * 0 X 0 X Disabled + * 0 1 1 0 Disabled + * 0 1 1 1 TX + * 1 0 0 X Disabled + * 1 X 1 X TX+RX + * 1 1 0 1 RX + */ +void linkmode_resolve_pause(const unsigned long *local_adv, + const unsigned long *partner_adv, + bool *tx_pause, bool *rx_pause) +{ + __ETHTOOL_DECLARE_LINK_MODE_MASK(m); + + linkmode_and(m, local_adv, partner_adv); + if (linkmode_test_bit(ETHTOOL_LINK_MODE_Pause_BIT, m)) { + *tx_pause = true; + *rx_pause = true; + } else if (linkmode_test_bit(ETHTOOL_LINK_MODE_Asym_Pause_BIT, m)) { + *tx_pause = linkmode_test_bit(ETHTOOL_LINK_MODE_Pause_BIT, + partner_adv); + *rx_pause = linkmode_test_bit(ETHTOOL_LINK_MODE_Pause_BIT, + local_adv); + } else { + *tx_pause = false; + *rx_pause = false; + } +} +EXPORT_SYMBOL_GPL(linkmode_resolve_pause); + +/** + * linkmode_set_pause - set the pause mode advertisement + * @advertisement: advertisement in ethtool format + * @tx: boolean from ethtool struct ethtool_pauseparam tx_pause member + * @rx: boolean from ethtool struct ethtool_pauseparam rx_pause member + * + * Configure the advertised Pause and Asym_Pause bits according to the + * capabilities of provided in @tx and @rx. + * + * We convert as follows: + * tx rx Pause AsymDir + * 0 0 0 0 + * 0 1 1 1 + * 1 0 0 1 + * 1 1 1 0 + * + * Note: this translation from ethtool tx/rx notation to the advertisement + * is actually very problematical. Here are some examples: + * + * For tx=0 rx=1, meaning transmit is unsupported, receive is supported: + * + * Local device Link partner + * Pause AsymDir Pause AsymDir Result + * 1 1 1 0 TX + RX - but we have no TX support. + * 1 1 0 1 Only this gives RX only + * + * For tx=1 rx=1, meaning we have the capability to transmit and receive + * pause frames: + * + * Local device Link partner + * Pause AsymDir Pause AsymDir Result + * 1 0 0 1 Disabled - but since we do support tx and rx, + * this should resolve to RX only. + * + * Hence, asking for: + * rx=1 tx=0 gives Pause+AsymDir advertisement, but we may end up + * resolving to tx+rx pause or only rx pause depending on + * the partners advertisement. + * rx=0 tx=1 gives AsymDir only, which will only give tx pause if + * the partners advertisement allows it. + * rx=1 tx=1 gives Pause only, which will only allow tx+rx pause + * if the other end also advertises Pause. + */ +void linkmode_set_pause(unsigned long *advertisement, bool tx, bool rx) +{ + linkmode_mod_bit(ETHTOOL_LINK_MODE_Pause_BIT, advertisement, rx); + linkmode_mod_bit(ETHTOOL_LINK_MODE_Asym_Pause_BIT, advertisement, + rx ^ tx); +} +EXPORT_SYMBOL_GPL(linkmode_set_pause); -- cgit v1.3.1