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/swphy.c | 175 ++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 175 insertions(+) create mode 100644 drivers/net/phy/swphy.c (limited to 'drivers/net/phy/swphy.c') diff --git a/drivers/net/phy/swphy.c b/drivers/net/phy/swphy.c new file mode 100644 index 000000000..59f1ba4d4 --- /dev/null +++ b/drivers/net/phy/swphy.c @@ -0,0 +1,175 @@ +// SPDX-License-Identifier: GPL-2.0+ +/* + * Software PHY emulation + * + * Code taken from fixed_phy.c by Russell King. + * + * Author: Vitaly Bordug + * Anton Vorontsov + * + * Copyright (c) 2006-2007 MontaVista Software, Inc. + */ +#include +#include +#include +#include + +#include "swphy.h" + +#define MII_REGS_NUM 29 + +struct swmii_regs { + u16 bmsr; + u16 lpa; + u16 lpagb; + u16 estat; +}; + +enum { + SWMII_SPEED_10 = 0, + SWMII_SPEED_100, + SWMII_SPEED_1000, + SWMII_DUPLEX_HALF = 0, + SWMII_DUPLEX_FULL, +}; + +/* + * These two tables get bitwise-anded together to produce the final result. + * This means the speed table must contain both duplex settings, and the + * duplex table must contain all speed settings. + */ +static const struct swmii_regs speed[] = { + [SWMII_SPEED_10] = { + .lpa = LPA_10FULL | LPA_10HALF, + }, + [SWMII_SPEED_100] = { + .bmsr = BMSR_100FULL | BMSR_100HALF, + .lpa = LPA_100FULL | LPA_100HALF, + }, + [SWMII_SPEED_1000] = { + .bmsr = BMSR_ESTATEN, + .lpagb = LPA_1000FULL | LPA_1000HALF, + .estat = ESTATUS_1000_TFULL | ESTATUS_1000_THALF, + }, +}; + +static const struct swmii_regs duplex[] = { + [SWMII_DUPLEX_HALF] = { + .bmsr = BMSR_ESTATEN | BMSR_100HALF, + .lpa = LPA_10HALF | LPA_100HALF, + .lpagb = LPA_1000HALF, + .estat = ESTATUS_1000_THALF, + }, + [SWMII_DUPLEX_FULL] = { + .bmsr = BMSR_ESTATEN | BMSR_100FULL, + .lpa = LPA_10FULL | LPA_100FULL, + .lpagb = LPA_1000FULL, + .estat = ESTATUS_1000_TFULL, + }, +}; + +static int swphy_decode_speed(int speed) +{ + switch (speed) { + case 1000: + return SWMII_SPEED_1000; + case 100: + return SWMII_SPEED_100; + case 10: + return SWMII_SPEED_10; + default: + return -EINVAL; + } +} + +/** + * swphy_validate_state - validate the software phy status + * @state: software phy status + * + * This checks that we can represent the state stored in @state can be + * represented in the emulated MII registers. Returns 0 if it can, + * otherwise returns -EINVAL. + */ +int swphy_validate_state(const struct fixed_phy_status *state) +{ + int err; + + if (state->link) { + err = swphy_decode_speed(state->speed); + if (err < 0) { + pr_warn("swphy: unknown speed\n"); + return -EINVAL; + } + } + return 0; +} +EXPORT_SYMBOL_GPL(swphy_validate_state); + +/** + * swphy_read_reg - return a MII register from the fixed phy state + * @reg: MII register + * @state: fixed phy status + * + * Return the MII @reg register generated from the fixed phy state @state. + */ +int swphy_read_reg(int reg, const struct fixed_phy_status *state) +{ + int speed_index, duplex_index; + u16 bmsr = BMSR_ANEGCAPABLE; + u16 estat = 0; + u16 lpagb = 0; + u16 lpa = 0; + + if (reg > MII_REGS_NUM) + return -1; + + speed_index = swphy_decode_speed(state->speed); + if (WARN_ON(speed_index < 0)) + return 0; + + duplex_index = state->duplex ? SWMII_DUPLEX_FULL : SWMII_DUPLEX_HALF; + + bmsr |= speed[speed_index].bmsr & duplex[duplex_index].bmsr; + estat |= speed[speed_index].estat & duplex[duplex_index].estat; + + if (state->link) { + bmsr |= BMSR_LSTATUS | BMSR_ANEGCOMPLETE; + + lpa |= speed[speed_index].lpa & duplex[duplex_index].lpa; + lpagb |= speed[speed_index].lpagb & duplex[duplex_index].lpagb; + + if (state->pause) + lpa |= LPA_PAUSE_CAP; + + if (state->asym_pause) + lpa |= LPA_PAUSE_ASYM; + } + + switch (reg) { + case MII_BMCR: + return BMCR_ANENABLE; + case MII_BMSR: + return bmsr; + case MII_PHYSID1: + case MII_PHYSID2: + return 0; + case MII_LPA: + return lpa; + case MII_STAT1000: + return lpagb; + case MII_ESTATUS: + return estat; + + /* + * We do not support emulating Clause 45 over Clause 22 register + * reads. Return an error instead of bogus data. + */ + case MII_MMD_CTRL: + case MII_MMD_DATA: + return -1; + + default: + return 0xffff; + } +} +EXPORT_SYMBOL_GPL(swphy_read_reg); -- cgit v1.3.1