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, 143 insertions, 0 deletions

diff --git a/arch/powerpc/boot/decompress.c b/arch/powerpc/boot/decompress.c
new file mode 100644
index 000000000..6835cb53f
--- /dev/null
+++ b/arch/powerpc/boot/decompress.c
@@ -0,0 +1,143 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Wrapper around the kernel's pre-boot decompression library.
+ *
+ * Copyright (C) IBM Corporation 2016.
+ */
+
+#include "elf.h"
+#include "page.h"
+#include "string.h"
+#include "stdio.h"
+#include "ops.h"
+#include "reg.h"
+#include "types.h"
+
+/*
+ * The decompressor_*.c files play #ifdef games so they can be used in both
+ * pre-boot and regular kernel code. We need these definitions to make the
+ * includes work.
+ */
+
+#define STATIC static
+#define INIT
+
+/*
+ * The build process will copy the required zlib source files and headers
+ * out of lib/ and "fix" the includes so they do not pull in other kernel
+ * headers.
+ */
+
+#ifdef CONFIG_KERNEL_GZIP
+#	include "decompress_inflate.c"
+#endif
+
+#ifdef CONFIG_KERNEL_XZ
+#	include "xz_config.h"
+#	include "../../../lib/decompress_unxz.c"
+#endif
+
+/* globals for tracking the state of the decompression */
+static unsigned long decompressed_bytes;
+static unsigned long limit;
+static unsigned long skip;
+static char *output_buffer;
+
+/*
+ * flush() is called by __decompress() when the decompressor's scratch buffer is
+ * full.
+ */
+static long flush(void *v, unsigned long buffer_size)
+{
+	unsigned long end = decompressed_bytes + buffer_size;
+	unsigned long size = buffer_size;
+	unsigned long offset = 0;
+	char *in = v;
+	char *out;
+
+	/*
+	 * if we hit our decompression limit, we need to fake an error to abort
+	 * the in-progress decompression.
+	 */
+	if (decompressed_bytes >= limit)
+		return -1;
+
+	/* skip this entire block */
+	if (end <= skip) {
+		decompressed_bytes += buffer_size;
+		return buffer_size;
+	}
+
+	/* skip some data at the start, but keep the rest of the block */
+	if (decompressed_bytes < skip && end > skip) {
+		offset = skip - decompressed_bytes;
+
+		in += offset;
+		size -= offset;
+		decompressed_bytes += offset;
+	}
+
+	out = &output_buffer[decompressed_bytes - skip];
+	size = min(decompressed_bytes + size, limit) - decompressed_bytes;
+
+	memcpy(out, in, size);
+	decompressed_bytes += size;
+
+	return buffer_size;
+}
+
+static void print_err(char *s)
+{
+	/* suppress the "error" when we terminate the decompressor */
+	if (decompressed_bytes >= limit)
+		return;
+
+	printf("Decompression error: '%s'\n\r", s);
+}
+
+/**
+ * partial_decompress - decompresses part or all of a compressed buffer
+ * @inbuf:       input buffer
+ * @input_size:  length of the input buffer
+ * @outbuf:      output buffer
+ * @output_size: length of the output buffer
+ * @_skip:       number of output bytes to ignore
+ *
+ * This function takes compressed data from inbuf, decompresses and write it to
+ * outbuf. Once output_size bytes are written to the output buffer, or the
+ * stream is exhausted the function will return the number of bytes that were
+ * decompressed. Otherwise it will return whatever error code the decompressor
+ * reported (NB: This is specific to each decompressor type).
+ *
+ * The skip functionality is mainly there so the program and discover
+ * the size of the compressed image so that it can ask firmware (if present)
+ * for an appropriately sized buffer.
+ */
+long partial_decompress(void *inbuf, unsigned long input_size,
+	void *outbuf, unsigned long output_size, unsigned long _skip)
+{
+	int ret;
+
+	/*
+	 * The skipped bytes needs to be included in the size of data we want
+	 * to decompress.
+	 */
+	output_size += _skip;
+
+	decompressed_bytes = 0;
+	output_buffer = outbuf;
+	limit = output_size;
+	skip = _skip;
+
+	ret = __decompress(inbuf, input_size, NULL, flush, outbuf,
+		output_size, NULL, print_err);
+
+	/*
+	 * If decompression was aborted due to an actual error rather than
+	 * a fake error that we used to abort, then we should report it.
+	 */
+	if (decompressed_bytes < limit)
+		return ret;
+
+	return decompressed_bytes - skip;
+}