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/mtd/mtdsuper.c | 179 +++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 179 insertions(+) create mode 100644 drivers/mtd/mtdsuper.c (limited to 'drivers/mtd/mtdsuper.c') diff --git a/drivers/mtd/mtdsuper.c b/drivers/mtd/mtdsuper.c new file mode 100644 index 000000000..c709ff7aa --- /dev/null +++ b/drivers/mtd/mtdsuper.c @@ -0,0 +1,179 @@ +// SPDX-License-Identifier: GPL-2.0-or-later +/* MTD-based superblock management + * + * Copyright © 2001-2007 Red Hat, Inc. All Rights Reserved. + * Copyright © 2001-2010 David Woodhouse + * + * Written by: David Howells + * David Woodhouse + */ + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include "mtdcore.h" + +/* + * get a superblock on an MTD-backed filesystem + */ +static int mtd_get_sb(struct fs_context *fc, + struct mtd_info *mtd, + int (*fill_super)(struct super_block *, + struct fs_context *)) +{ + struct super_block *sb; + int ret; + + sb = sget_dev(fc, MKDEV(MTD_BLOCK_MAJOR, mtd->index)); + if (IS_ERR(sb)) + return PTR_ERR(sb); + + if (sb->s_root) { + /* new mountpoint for an already mounted superblock */ + pr_debug("MTDSB: Device %d (\"%s\") is already mounted\n", + mtd->index, mtd->name); + put_mtd_device(mtd); + } else { + /* fresh new superblock */ + pr_debug("MTDSB: New superblock for device %d (\"%s\")\n", + mtd->index, mtd->name); + + /* + * Would usually have been set with @sb_lock held but in + * contrast to sb->s_bdev that's checked with only + * @sb_lock held, nothing checks sb->s_mtd without also + * holding sb->s_umount and we're holding sb->s_umount + * here. + */ + sb->s_mtd = mtd; + sb->s_bdi = bdi_get(mtd_bdi); + + ret = fill_super(sb, fc); + if (ret < 0) + goto error_sb; + + sb->s_flags |= SB_ACTIVE; + } + + BUG_ON(fc->root); + fc->root = dget(sb->s_root); + return 0; + +error_sb: + deactivate_locked_super(sb); + return ret; +} + +/* + * get a superblock on an MTD-backed filesystem by MTD device number + */ +static int mtd_get_sb_by_nr(struct fs_context *fc, int mtdnr, + int (*fill_super)(struct super_block *, + struct fs_context *)) +{ + struct mtd_info *mtd; + + mtd = get_mtd_device(NULL, mtdnr); + if (IS_ERR(mtd)) { + errorf(fc, "MTDSB: Device #%u doesn't appear to exist\n", mtdnr); + return PTR_ERR(mtd); + } + + return mtd_get_sb(fc, mtd, fill_super); +} + +/** + * get_tree_mtd - Get a superblock based on a single MTD device + * @fc: The filesystem context holding the parameters + * @fill_super: Helper to initialise a new superblock + */ +int get_tree_mtd(struct fs_context *fc, + int (*fill_super)(struct super_block *sb, + struct fs_context *fc)) +{ +#ifdef CONFIG_BLOCK + dev_t dev; + int ret; +#endif + int mtdnr; + + if (!fc->source) + return invalf(fc, "No source specified"); + + pr_debug("MTDSB: dev_name \"%s\"\n", fc->source); + + /* the preferred way of mounting in future; especially when + * CONFIG_BLOCK=n - we specify the underlying MTD device by number or + * by name, so that we don't require block device support to be present + * in the kernel. + */ + if (fc->source[0] == 'm' && + fc->source[1] == 't' && + fc->source[2] == 'd') { + if (fc->source[3] == ':') { + struct mtd_info *mtd; + + /* mount by MTD device name */ + pr_debug("MTDSB: mtd:%%s, name \"%s\"\n", + fc->source + 4); + + mtd = get_mtd_device_nm(fc->source + 4); + if (!IS_ERR(mtd)) + return mtd_get_sb(fc, mtd, fill_super); + + errorf(fc, "MTD: MTD device with name \"%s\" not found", + fc->source + 4); + + } else if (isdigit(fc->source[3])) { + /* mount by MTD device number name */ + unsigned int mtdnr_val; + + if (kstrtouint(fc->source + 3, 0, &mtdnr_val) == 0) { + mtdnr = mtdnr_val; + /* It was a valid number */ + pr_debug("MTDSB: mtd%%d, mtdnr %u\n", mtdnr_val); + return mtd_get_sb_by_nr(fc, mtdnr, fill_super); + } + } + } + +#ifdef CONFIG_BLOCK + /* try the old way - the hack where we allowed users to mount + * /dev/mtdblock$(n) but didn't actually _use_ the blockdev + */ + ret = lookup_bdev(fc->source, &dev); + if (ret) { + errorf(fc, "MTD: Couldn't look up '%s': %d", fc->source, ret); + return ret; + } + pr_debug("MTDSB: lookup_bdev() returned 0\n"); + + if (MAJOR(dev) == MTD_BLOCK_MAJOR) + return mtd_get_sb_by_nr(fc, MINOR(dev), fill_super); + +#endif /* CONFIG_BLOCK */ + + if (!(fc->sb_flags & SB_SILENT)) + errorf(fc, "MTD: Attempt to mount non-MTD device \"%s\"", + fc->source); + return -EINVAL; +} +EXPORT_SYMBOL_GPL(get_tree_mtd); + +/* + * destroy an MTD-based superblock + */ +void kill_mtd_super(struct super_block *sb) +{ + generic_shutdown_super(sb); + put_mtd_device(sb->s_mtd); + sb->s_mtd = NULL; +} + +EXPORT_SYMBOL_GPL(kill_mtd_super); -- cgit v1.3.1