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/reset/reset-mpfs.c | 207 +++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 207 insertions(+) create mode 100644 drivers/reset/reset-mpfs.c (limited to 'drivers/reset/reset-mpfs.c') diff --git a/drivers/reset/reset-mpfs.c b/drivers/reset/reset-mpfs.c new file mode 100644 index 000000000..8ffcc54ee --- /dev/null +++ b/drivers/reset/reset-mpfs.c @@ -0,0 +1,207 @@ +// SPDX-License-Identifier: GPL-2.0-only +/* + * PolarFire SoC (MPFS) Peripheral Clock Reset Controller + * + * Author: Conor Dooley + * Copyright (c) 2022 Microchip Technology Inc. and its subsidiaries. + * + */ +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +/* + * The ENVM reset is the lowest bit in the register & I am using the CLK_FOO + * defines in the dt to make things easier to configure - so this is accounting + * for the offset of 3 there. + */ +#define MPFS_PERIPH_OFFSET CLK_ENVM +#define MPFS_NUM_RESETS 30u +#define MPFS_SLEEP_MIN_US 100 +#define MPFS_SLEEP_MAX_US 200 + +#define REG_SUBBLK_RESET_CR 0x88u + +struct mpfs_reset { + struct regmap *regmap; + struct reset_controller_dev rcdev; +}; + +static inline struct mpfs_reset *to_mpfs_reset(struct reset_controller_dev *rcdev) +{ + return container_of(rcdev, struct mpfs_reset, rcdev); +} + +/* + * Peripheral clock resets + */ +static int mpfs_assert(struct reset_controller_dev *rcdev, unsigned long id) +{ + struct mpfs_reset *rst = to_mpfs_reset(rcdev); + + return regmap_set_bits(rst->regmap, REG_SUBBLK_RESET_CR, BIT(id)); + +} + +static int mpfs_deassert(struct reset_controller_dev *rcdev, unsigned long id) +{ + struct mpfs_reset *rst = to_mpfs_reset(rcdev); + + return regmap_clear_bits(rst->regmap, REG_SUBBLK_RESET_CR, BIT(id)); + +} + +static int mpfs_status(struct reset_controller_dev *rcdev, unsigned long id) +{ + struct mpfs_reset *rst = to_mpfs_reset(rcdev); + u32 reg; + + regmap_read(rst->regmap, REG_SUBBLK_RESET_CR, ®); + + /* + * It is safe to return here as MPFS_NUM_RESETS makes sure the sign bit + * is never hit. + */ + return (reg & BIT(id)); +} + +static int mpfs_reset(struct reset_controller_dev *rcdev, unsigned long id) +{ + mpfs_assert(rcdev, id); + + usleep_range(MPFS_SLEEP_MIN_US, MPFS_SLEEP_MAX_US); + + mpfs_deassert(rcdev, id); + + return 0; +} + +static const struct reset_control_ops mpfs_reset_ops = { + .reset = mpfs_reset, + .assert = mpfs_assert, + .deassert = mpfs_deassert, + .status = mpfs_status, +}; + +static int mpfs_reset_xlate(struct reset_controller_dev *rcdev, + const struct of_phandle_args *reset_spec) +{ + unsigned int index = reset_spec->args[0]; + + /* + * CLK_RESERVED does not map to a clock, but it does map to a reset, + * so it has to be accounted for here. It is the reset for the fabric, + * so if this reset gets called - do not reset it. + */ + if (index == CLK_RESERVED) { + dev_err(rcdev->dev, "Resetting the fabric is not supported\n"); + return -EINVAL; + } + + if (index < MPFS_PERIPH_OFFSET || index >= (MPFS_PERIPH_OFFSET + rcdev->nr_resets)) { + dev_err(rcdev->dev, "Invalid reset index %u\n", index); + return -EINVAL; + } + + return index - MPFS_PERIPH_OFFSET; +} + +static int mpfs_reset_mfd_probe(struct platform_device *pdev) +{ + struct reset_controller_dev *rcdev; + struct device *dev = &pdev->dev; + struct mpfs_reset *rst; + + rst = devm_kzalloc(dev, sizeof(*rst), GFP_KERNEL); + if (!rst) + return -ENOMEM; + + rcdev = &rst->rcdev; + rcdev->dev = dev; + rcdev->ops = &mpfs_reset_ops; + + rcdev->of_node = pdev->dev.parent->of_node; + rcdev->of_reset_n_cells = 1; + rcdev->of_xlate = mpfs_reset_xlate; + rcdev->nr_resets = MPFS_NUM_RESETS; + + rst->regmap = device_node_to_regmap(pdev->dev.parent->of_node); + if (IS_ERR(rst->regmap)) + return dev_err_probe(dev, PTR_ERR(rst->regmap), + "Failed to find syscon regmap\n"); + + return devm_reset_controller_register(dev, rcdev); +} + +static struct platform_driver mpfs_reset_mfd_driver = { + .probe = mpfs_reset_mfd_probe, + .driver = { + .name = "mpfs-reset", + }, +}; +module_platform_driver(mpfs_reset_mfd_driver); + +static int mpfs_reset_adev_probe(struct auxiliary_device *adev, + const struct auxiliary_device_id *id) +{ + struct reset_controller_dev *rcdev; + struct device *dev = &adev->dev; + struct mpfs_reset *rst; + + rst = devm_kzalloc(dev, sizeof(*rst), GFP_KERNEL); + if (!rst) + return -ENOMEM; + + rst->regmap = (struct regmap *)adev->dev.platform_data; + + rcdev = &rst->rcdev; + rcdev->dev = dev; + rcdev->ops = &mpfs_reset_ops; + + rcdev->of_node = dev->parent->of_node; + rcdev->of_reset_n_cells = 1; + rcdev->of_xlate = mpfs_reset_xlate; + rcdev->nr_resets = MPFS_NUM_RESETS; + + return devm_reset_controller_register(dev, rcdev); +} + +int mpfs_reset_controller_register(struct device *clk_dev, struct regmap *map) +{ + struct auxiliary_device *adev; + + adev = devm_auxiliary_device_create(clk_dev, "reset-mpfs", (void *)map); + if (!adev) + return -ENODEV; + + return 0; +} +EXPORT_SYMBOL_NS_GPL(mpfs_reset_controller_register, "MCHP_CLK_MPFS"); + +static const struct auxiliary_device_id mpfs_reset_ids[] = { + { + .name = "reset_mpfs.reset-mpfs", + }, + { } +}; +MODULE_DEVICE_TABLE(auxiliary, mpfs_reset_ids); + +static struct auxiliary_driver mpfs_reset_aux_driver = { + .probe = mpfs_reset_adev_probe, + .id_table = mpfs_reset_ids, +}; + +module_auxiliary_driver(mpfs_reset_aux_driver); + +MODULE_DESCRIPTION("Microchip PolarFire SoC Reset Driver"); +MODULE_AUTHOR("Conor Dooley "); +MODULE_IMPORT_NS("MCHP_CLK_MPFS"); -- cgit v1.3.1