sched/cache: Keep nr_pref_llc_running in the runnable domain, to fix LLC mis-scheduling buggrafted

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 <zhanxusheng@xiaomi.com>
Suggested-by: Chen Yu <yu.c.chen@intel.com>
Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Reviewed-by: Kayra Cizmeci <kayracizmeci@gmail.com>
Cc: <stable@kernel.org> # v7.2.x
Link: https://patch.msgid.link/06af61afedac32e6477f57feb4d658f6c411c3af.1790035273.git.tim.c.chen@linux.intel.com

1 files changed, 34 insertions, 0 deletions

diff --git a/Documentation/filesystems/ext4/bigalloc.rst b/Documentation/filesystems/ext4/bigalloc.rst
new file mode 100644
index 000000000..976a180b2
--- /dev/null
+++ b/Documentation/filesystems/ext4/bigalloc.rst
@@ -0,0 +1,34 @@
+.. SPDX-License-Identifier: GPL-2.0
+
+Bigalloc
+--------
+
+At the moment, the default size of a block is 4KiB, which is a commonly
+supported page size on most MMU-capable hardware. This is fortunate, as
+ext4 code is not prepared to handle the case where the block size
+exceeds the page size. However, for a filesystem of mostly huge files,
+it is desirable to be able to allocate disk blocks in units of multiple
+blocks to reduce both fragmentation and metadata overhead. The
+bigalloc feature provides exactly this ability.
+
+The bigalloc feature (EXT4_FEATURE_RO_COMPAT_BIGALLOC) changes ext4 to
+use clustered allocation, so that each bit in the ext4 block allocation
+bitmap addresses a power of two number of blocks. For example, if the
+file system is mainly going to be storing large files in the 4-32
+megabyte range, it might make sense to set a cluster size of 1 megabyte.
+This means that each bit in the block allocation bitmap now addresses
+256 4k blocks. This shrinks the total size of the block allocation
+bitmaps for a 2T file system from 64 megabytes to 256 kilobytes. It also
+means that a block group addresses 32 gigabytes instead of 128 megabytes,
+also shrinking the amount of file system overhead for metadata.
+
+The administrator can set a block cluster size at mkfs time (which is
+stored in the s_log_cluster_size field in the superblock); from then
+on, the block bitmaps track clusters, not individual blocks. This means
+that block groups can be several gigabytes in size (instead of just
+128MiB); however, the minimum allocation unit becomes a cluster, not a
+block, even for directories. TaoBao had a patchset to extend the “use
+units of clusters instead of blocks” to the extent tree, though it is
+not clear where those patches went-- they eventually morphed into
+“extent tree v2” but that code has not landed as of May 2015.
+