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, 413 insertions, 0 deletions

diff --git a/drivers/scsi/scsi_trace.c b/drivers/scsi/scsi_trace.c
new file mode 100644
index 000000000..b3baae91e
--- /dev/null
+++ b/drivers/scsi/scsi_trace.c
@@ -0,0 +1,413 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Copyright (C) 2010 FUJITSU LIMITED
+ * Copyright (C) 2010 Tomohiro Kusumi <kusumi.tomohiro@jp.fujitsu.com>
+ */
+#include <linux/kernel.h>
+#include <linux/trace_seq.h>
+#include <linux/unaligned.h>
+#include <trace/events/scsi.h>
+
+#define SERVICE_ACTION16(cdb) (cdb[1] & 0x1f)
+#define SERVICE_ACTION32(cdb) (get_unaligned_be16(&cdb[8]))
+
+static const char *
+scsi_trace_misc(struct trace_seq *, unsigned char *, int);
+
+static const char *
+scsi_trace_rw6(struct trace_seq *p, unsigned char *cdb, int len)
+{
+	const char *ret = trace_seq_buffer_ptr(p);
+	u32 lba, txlen;
+
+	lba = get_unaligned_be24(&cdb[1]) & 0x1fffff;
+	/*
+	 * From SBC-2: a TRANSFER LENGTH field set to zero specifies that 256
+	 * logical blocks shall be read (READ(6)) or written (WRITE(6)).
+	 */
+	txlen = cdb[4] ? cdb[4] : 256;
+
+	trace_seq_printf(p, "lba=%u txlen=%u", lba, txlen);
+	trace_seq_putc(p, 0);
+
+	return ret;
+}
+
+static const char *
+scsi_trace_rw10(struct trace_seq *p, unsigned char *cdb, int len)
+{
+	const char *ret = trace_seq_buffer_ptr(p);
+	u32 lba, txlen;
+
+	lba = get_unaligned_be32(&cdb[2]);
+	txlen = get_unaligned_be16(&cdb[7]);
+
+	trace_seq_printf(p, "lba=%u txlen=%u protect=%u", lba, txlen,
+			 cdb[1] >> 5);
+
+	if (cdb[0] == WRITE_SAME)
+		trace_seq_printf(p, " unmap=%u", cdb[1] >> 3 & 1);
+
+	trace_seq_putc(p, 0);
+
+	return ret;
+}
+
+static const char *
+scsi_trace_rw12(struct trace_seq *p, unsigned char *cdb, int len)
+{
+	const char *ret = trace_seq_buffer_ptr(p);
+	u32 lba, txlen;
+
+	lba = get_unaligned_be32(&cdb[2]);
+	txlen = get_unaligned_be32(&cdb[6]);
+
+	trace_seq_printf(p, "lba=%u txlen=%u protect=%u", lba, txlen,
+			 cdb[1] >> 5);
+	trace_seq_putc(p, 0);
+
+	return ret;
+}
+
+static const char *
+scsi_trace_rw16(struct trace_seq *p, unsigned char *cdb, int len)
+{
+	const char *ret = trace_seq_buffer_ptr(p);
+	u64 lba;
+	u32 txlen;
+
+	lba = get_unaligned_be64(&cdb[2]);
+	txlen = get_unaligned_be32(&cdb[10]);
+
+	trace_seq_printf(p, "lba=%llu txlen=%u protect=%u", lba, txlen,
+			 cdb[1] >> 5);
+
+	if (cdb[0] == WRITE_SAME_16)
+		trace_seq_printf(p, " unmap=%u", cdb[1] >> 3 & 1);
+
+	trace_seq_putc(p, 0);
+
+	return ret;
+}
+
+static const char *
+scsi_trace_rw32(struct trace_seq *p, unsigned char *cdb, int len)
+{
+	const char *ret = trace_seq_buffer_ptr(p), *cmd;
+	u64 lba;
+	u32 ei_lbrt, txlen;
+
+	switch (SERVICE_ACTION32(cdb)) {
+	case READ_32:
+		cmd = "READ";
+		break;
+	case VERIFY_32:
+		cmd = "VERIFY";
+		break;
+	case WRITE_32:
+		cmd = "WRITE";
+		break;
+	case WRITE_SAME_32:
+		cmd = "WRITE_SAME";
+		break;
+	default:
+		trace_seq_puts(p, "UNKNOWN");
+		goto out;
+	}
+
+	lba = get_unaligned_be64(&cdb[12]);
+	ei_lbrt = get_unaligned_be32(&cdb[20]);
+	txlen = get_unaligned_be32(&cdb[28]);
+
+	trace_seq_printf(p, "%s_32 lba=%llu txlen=%u protect=%u ei_lbrt=%u",
+			 cmd, lba, txlen, cdb[10] >> 5, ei_lbrt);
+
+	if (SERVICE_ACTION32(cdb) == WRITE_SAME_32)
+		trace_seq_printf(p, " unmap=%u", cdb[10] >> 3 & 1);
+
+out:
+	trace_seq_putc(p, 0);
+
+	return ret;
+}
+
+static const char *
+scsi_trace_unmap(struct trace_seq *p, unsigned char *cdb, int len)
+{
+	const char *ret = trace_seq_buffer_ptr(p);
+	unsigned int regions = get_unaligned_be16(&cdb[7]);
+
+	trace_seq_printf(p, "regions=%u", (regions - 8) / 16);
+	trace_seq_putc(p, 0);
+
+	return ret;
+}
+
+static const char *
+scsi_trace_service_action_in(struct trace_seq *p, unsigned char *cdb, int len)
+{
+	const char *ret = trace_seq_buffer_ptr(p), *cmd;
+	u64 lba;
+	u32 alloc_len;
+
+	switch (SERVICE_ACTION16(cdb)) {
+	case SAI_READ_CAPACITY_16:
+		cmd = "READ_CAPACITY_16";
+		break;
+	case SAI_GET_LBA_STATUS:
+		cmd = "GET_LBA_STATUS";
+		break;
+	default:
+		trace_seq_puts(p, "UNKNOWN");
+		goto out;
+	}
+
+	lba = get_unaligned_be64(&cdb[2]);
+	alloc_len = get_unaligned_be32(&cdb[10]);
+
+	trace_seq_printf(p, "%s lba=%llu alloc_len=%u", cmd, lba, alloc_len);
+
+out:
+	trace_seq_putc(p, 0);
+
+	return ret;
+}
+
+static const char *
+scsi_trace_maintenance_in(struct trace_seq *p, unsigned char *cdb, int len)
+{
+	const char *ret = trace_seq_buffer_ptr(p), *cmd;
+	u32 alloc_len;
+
+	switch (SERVICE_ACTION16(cdb)) {
+	case MI_REPORT_IDENTIFYING_INFORMATION:
+		cmd = "REPORT_IDENTIFYING_INFORMATION";
+		break;
+	case MI_REPORT_TARGET_PGS:
+		cmd = "REPORT_TARGET_PORT_GROUPS";
+		break;
+	case MI_REPORT_ALIASES:
+		cmd = "REPORT_ALIASES";
+		break;
+	case MI_REPORT_SUPPORTED_OPERATION_CODES:
+		cmd = "REPORT_SUPPORTED_OPERATION_CODES";
+		break;
+	case MI_REPORT_SUPPORTED_TASK_MANAGEMENT_FUNCTIONS:
+		cmd = "REPORT_SUPPORTED_TASK_MANAGEMENT_FUNCTIONS";
+		break;
+	case MI_REPORT_PRIORITY:
+		cmd = "REPORT_PRIORITY";
+		break;
+	case MI_REPORT_TIMESTAMP:
+		cmd = "REPORT_TIMESTAMP";
+		break;
+	case MI_MANAGEMENT_PROTOCOL_IN:
+		cmd = "MANAGEMENT_PROTOCOL_IN";
+		break;
+	default:
+		trace_seq_puts(p, "UNKNOWN");
+		goto out;
+	}
+
+	alloc_len = get_unaligned_be32(&cdb[6]);
+
+	trace_seq_printf(p, "%s alloc_len=%u", cmd, alloc_len);
+
+out:
+	trace_seq_putc(p, 0);
+
+	return ret;
+}
+
+static const char *
+scsi_trace_maintenance_out(struct trace_seq *p, unsigned char *cdb, int len)
+{
+	const char *ret = trace_seq_buffer_ptr(p), *cmd;
+	u32 alloc_len;
+
+	switch (SERVICE_ACTION16(cdb)) {
+	case MO_SET_IDENTIFYING_INFORMATION:
+		cmd = "SET_IDENTIFYING_INFORMATION";
+		break;
+	case MO_SET_TARGET_PGS:
+		cmd = "SET_TARGET_PORT_GROUPS";
+		break;
+	case MO_CHANGE_ALIASES:
+		cmd = "CHANGE_ALIASES";
+		break;
+	case MO_SET_PRIORITY:
+		cmd = "SET_PRIORITY";
+		break;
+	case MO_SET_TIMESTAMP:
+		cmd = "SET_TIMESTAMP";
+		break;
+	case MO_MANAGEMENT_PROTOCOL_OUT:
+		cmd = "MANAGEMENT_PROTOCOL_OUT";
+		break;
+	default:
+		trace_seq_puts(p, "UNKNOWN");
+		goto out;
+	}
+
+	alloc_len = get_unaligned_be32(&cdb[6]);
+
+	trace_seq_printf(p, "%s alloc_len=%u", cmd, alloc_len);
+
+out:
+	trace_seq_putc(p, 0);
+
+	return ret;
+}
+
+static const char *
+scsi_trace_zbc_in(struct trace_seq *p, unsigned char *cdb, int len)
+{
+	const char *ret = trace_seq_buffer_ptr(p), *cmd;
+	u64 zone_id;
+	u32 alloc_len;
+	u8 options;
+
+	switch (SERVICE_ACTION16(cdb)) {
+	case ZI_REPORT_ZONES:
+		cmd = "REPORT_ZONES";
+		break;
+	default:
+		trace_seq_puts(p, "UNKNOWN");
+		goto out;
+	}
+
+	zone_id = get_unaligned_be64(&cdb[2]);
+	alloc_len = get_unaligned_be32(&cdb[10]);
+	options = cdb[14] & 0x3f;
+
+	trace_seq_printf(p, "%s zone=%llu alloc_len=%u options=%u partial=%u",
+			 cmd, (unsigned long long)zone_id, alloc_len,
+			 options, (cdb[14] >> 7) & 1);
+
+out:
+	trace_seq_putc(p, 0);
+
+	return ret;
+}
+
+static const char *
+scsi_trace_zbc_out(struct trace_seq *p, unsigned char *cdb, int len)
+{
+	const char *ret = trace_seq_buffer_ptr(p), *cmd;
+	u64 zone_id;
+
+	switch (SERVICE_ACTION16(cdb)) {
+	case ZO_CLOSE_ZONE:
+		cmd = "CLOSE_ZONE";
+		break;
+	case ZO_FINISH_ZONE:
+		cmd = "FINISH_ZONE";
+		break;
+	case ZO_OPEN_ZONE:
+		cmd = "OPEN_ZONE";
+		break;
+	case ZO_RESET_WRITE_POINTER:
+		cmd = "RESET_WRITE_POINTER";
+		break;
+	default:
+		trace_seq_puts(p, "UNKNOWN");
+		goto out;
+	}
+
+	zone_id = get_unaligned_be64(&cdb[2]);
+
+	trace_seq_printf(p, "%s zone=%llu all=%u", cmd,
+			 (unsigned long long)zone_id, cdb[14] & 1);
+
+out:
+	trace_seq_putc(p, 0);
+
+	return ret;
+}
+
+static const char *
+scsi_trace_atomic_write16_out(struct trace_seq *p, unsigned char *cdb, int len)
+{
+	const char *ret = trace_seq_buffer_ptr(p);
+	unsigned int boundary_size;
+	unsigned int nr_blocks;
+	sector_t lba;
+
+	lba = get_unaligned_be64(&cdb[2]);
+	boundary_size = get_unaligned_be16(&cdb[10]);
+	nr_blocks = get_unaligned_be16(&cdb[12]);
+
+	trace_seq_printf(p, "lba=%llu txlen=%u boundary_size=%u",
+			  lba, nr_blocks, boundary_size);
+
+	trace_seq_putc(p, 0);
+
+	return ret;
+}
+
+static const char *
+scsi_trace_varlen(struct trace_seq *p, unsigned char *cdb, int len)
+{
+	switch (SERVICE_ACTION32(cdb)) {
+	case READ_32:
+	case VERIFY_32:
+	case WRITE_32:
+	case WRITE_SAME_32:
+		return scsi_trace_rw32(p, cdb, len);
+	default:
+		return scsi_trace_misc(p, cdb, len);
+	}
+}
+
+static const char *
+scsi_trace_misc(struct trace_seq *p, unsigned char *cdb, int len)
+{
+	const char *ret = trace_seq_buffer_ptr(p);
+
+	trace_seq_putc(p, '-');
+	trace_seq_putc(p, 0);
+
+	return ret;
+}
+
+const char *
+scsi_trace_parse_cdb(struct trace_seq *p, unsigned char *cdb, int len)
+{
+	switch (cdb[0]) {
+	case READ_6:
+	case WRITE_6:
+		return scsi_trace_rw6(p, cdb, len);
+	case READ_10:
+	case VERIFY:
+	case WRITE_10:
+	case WRITE_SAME:
+		return scsi_trace_rw10(p, cdb, len);
+	case READ_12:
+	case VERIFY_12:
+	case WRITE_12:
+		return scsi_trace_rw12(p, cdb, len);
+	case READ_16:
+	case VERIFY_16:
+	case WRITE_16:
+	case WRITE_SAME_16:
+		return scsi_trace_rw16(p, cdb, len);
+	case UNMAP:
+		return scsi_trace_unmap(p, cdb, len);
+	case SERVICE_ACTION_IN_16:
+		return scsi_trace_service_action_in(p, cdb, len);
+	case VARIABLE_LENGTH_CMD:
+		return scsi_trace_varlen(p, cdb, len);
+	case MAINTENANCE_IN:
+		return scsi_trace_maintenance_in(p, cdb, len);
+	case MAINTENANCE_OUT:
+		return scsi_trace_maintenance_out(p, cdb, len);
+	case ZBC_IN:
+		return scsi_trace_zbc_in(p, cdb, len);
+	case ZBC_OUT:
+		return scsi_trace_zbc_out(p, cdb, len);
+	case WRITE_ATOMIC_16:
+		return scsi_trace_atomic_write16_out(p, cdb, len);
+	default:
+		return scsi_trace_misc(p, cdb, len);
+	}
+}