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

3 files changed, 163 insertions, 0 deletions

diff --git a/sound/soc/sof/xtensa/Kconfig b/sound/soc/sof/xtensa/Kconfig
new file mode 100644
index 000000000..defd6d3dc
--- /dev/null
+++ b/sound/soc/sof/xtensa/Kconfig
@@ -0,0 +1,3 @@
+# SPDX-License-Identifier: GPL-2.0-only
+config SND_SOC_SOF_XTENSA
+	tristate
diff --git a/sound/soc/sof/xtensa/Makefile b/sound/soc/sof/xtensa/Makefile
new file mode 100644
index 000000000..b9e6e8f5a
--- /dev/null
+++ b/sound/soc/sof/xtensa/Makefile
@@ -0,0 +1,5 @@
+# SPDX-License-Identifier: (GPL-2.0-only OR BSD-3-Clause)
+
+snd-sof-xtensa-dsp-y := core.o
+
+obj-$(CONFIG_SND_SOC_SOF_XTENSA) += snd-sof-xtensa-dsp.o
diff --git a/sound/soc/sof/xtensa/core.c b/sound/soc/sof/xtensa/core.c
new file mode 100644
index 000000000..50623e65f
--- /dev/null
+++ b/sound/soc/sof/xtensa/core.c
@@ -0,0 +1,155 @@
+// SPDX-License-Identifier: (GPL-2.0-only OR BSD-3-Clause)
+//
+// This file is provided under a dual BSD/GPLv2 license.  When using or
+// redistributing this file, you may do so under either license.
+//
+// Copyright(c) 2018 Intel Corporation
+//
+// Author: Pan Xiuli <xiuli.pan@linux.intel.com>
+//
+
+#include <linux/module.h>
+#include <sound/sof.h>
+#include <sound/sof/xtensa.h>
+#include "../sof-priv.h"
+
+struct xtensa_exception_cause {
+	u32 id;
+	const char *msg;
+	const char *description;
+};
+
+/*
+ * From 4.4.1.5 table 4-64 Exception Causes of Xtensa
+ * Instruction Set Architecture (ISA) Reference Manual
+ */
+static const struct xtensa_exception_cause xtensa_exception_causes[] = {
+	{0, "IllegalInstructionCause", "Illegal instruction"},
+	{1, "SyscallCause", "SYSCALL instruction"},
+	{2, "InstructionFetchErrorCause",
+	"Processor internal physical address or data error during instruction fetch"},
+	{3, "LoadStoreErrorCause",
+	"Processor internal physical address or data error during load or store"},
+	{4, "Level1InterruptCause",
+	"Level-1 interrupt as indicated by set level-1 bits in the INTERRUPT register"},
+	{5, "AllocaCause",
+	"MOVSP instruction, if caller’s registers are not in the register file"},
+	{6, "IntegerDivideByZeroCause",
+	"QUOS, QUOU, REMS, or REMU divisor operand is zero"},
+	{8, "PrivilegedCause",
+	"Attempt to execute a privileged operation when CRING ? 0"},
+	{9, "LoadStoreAlignmentCause", "Load or store to an unaligned address"},
+	{12, "InstrPIFDataErrorCause",
+	"PIF data error during instruction fetch"},
+	{13, "LoadStorePIFDataErrorCause",
+	"Synchronous PIF data error during LoadStore access"},
+	{14, "InstrPIFAddrErrorCause",
+	"PIF address error during instruction fetch"},
+	{15, "LoadStorePIFAddrErrorCause",
+	"Synchronous PIF address error during LoadStore access"},
+	{16, "InstTLBMissCause", "Error during Instruction TLB refill"},
+	{17, "InstTLBMultiHitCause",
+	"Multiple instruction TLB entries matched"},
+	{18, "InstFetchPrivilegeCause",
+	"An instruction fetch referenced a virtual address at a ring level less than CRING"},
+	{20, "InstFetchProhibitedCause",
+	"An instruction fetch referenced a page mapped with an attribute that does not permit instruction fetch"},
+	{24, "LoadStoreTLBMissCause",
+	"Error during TLB refill for a load or store"},
+	{25, "LoadStoreTLBMultiHitCause",
+	"Multiple TLB entries matched for a load or store"},
+	{26, "LoadStorePrivilegeCause",
+	"A load or store referenced a virtual address at a ring level less than CRING"},
+	{28, "LoadProhibitedCause",
+	"A load referenced a page mapped with an attribute that does not permit loads"},
+	{32, "Coprocessor0Disabled",
+	"Coprocessor 0 instruction when cp0 disabled"},
+	{33, "Coprocessor1Disabled",
+	"Coprocessor 1 instruction when cp1 disabled"},
+	{34, "Coprocessor2Disabled",
+	"Coprocessor 2 instruction when cp2 disabled"},
+	{35, "Coprocessor3Disabled",
+	"Coprocessor 3 instruction when cp3 disabled"},
+	{36, "Coprocessor4Disabled",
+	"Coprocessor 4 instruction when cp4 disabled"},
+	{37, "Coprocessor5Disabled",
+	"Coprocessor 5 instruction when cp5 disabled"},
+	{38, "Coprocessor6Disabled",
+	"Coprocessor 6 instruction when cp6 disabled"},
+	{39, "Coprocessor7Disabled",
+	"Coprocessor 7 instruction when cp7 disabled"},
+};
+
+/* only need xtensa atm */
+static void xtensa_dsp_oops(struct snd_sof_dev *sdev, const char *level, void *oops)
+{
+	struct sof_ipc_dsp_oops_xtensa *xoops = oops;
+	int i;
+
+	dev_printk(level, sdev->dev, "error: DSP Firmware Oops\n");
+	for (i = 0; i < ARRAY_SIZE(xtensa_exception_causes); i++) {
+		if (xtensa_exception_causes[i].id == xoops->exccause) {
+			dev_printk(level, sdev->dev,
+				   "error: Exception Cause: %s, %s\n",
+				   xtensa_exception_causes[i].msg,
+				   xtensa_exception_causes[i].description);
+		}
+	}
+	dev_printk(level, sdev->dev,
+		   "EXCCAUSE 0x%8.8x EXCVADDR 0x%8.8x PS       0x%8.8x SAR     0x%8.8x\n",
+		   xoops->exccause, xoops->excvaddr, xoops->ps, xoops->sar);
+	dev_printk(level, sdev->dev,
+		   "EPC1     0x%8.8x EPC2     0x%8.8x EPC3     0x%8.8x EPC4    0x%8.8x",
+		   xoops->epc1, xoops->epc2, xoops->epc3, xoops->epc4);
+	dev_printk(level, sdev->dev,
+		   "EPC5     0x%8.8x EPC6     0x%8.8x EPC7     0x%8.8x DEPC    0x%8.8x",
+		   xoops->epc5, xoops->epc6, xoops->epc7, xoops->depc);
+	dev_printk(level, sdev->dev,
+		   "EPS2     0x%8.8x EPS3     0x%8.8x EPS4     0x%8.8x EPS5    0x%8.8x",
+		   xoops->eps2, xoops->eps3, xoops->eps4, xoops->eps5);
+	dev_printk(level, sdev->dev,
+		   "EPS6     0x%8.8x EPS7     0x%8.8x INTENABL 0x%8.8x INTERRU 0x%8.8x",
+		   xoops->eps6, xoops->eps7, xoops->intenable, xoops->interrupt);
+}
+
+static void xtensa_stack(struct snd_sof_dev *sdev, const char *level, void *oops,
+			 u32 *stack, u32 stack_words)
+{
+	struct sof_ipc_dsp_oops_xtensa *xoops = oops;
+	u32 stack_ptr = xoops->plat_hdr.stackptr;
+	/* 4 * 8chars + 3 ws + 1 terminating NUL */
+	unsigned char buf[4 * 8 + 3 + 1];
+	int i;
+
+	dev_printk(level, sdev->dev, "stack dump from 0x%8.8x\n", stack_ptr);
+
+	/*
+	 * example output:
+	 * 0x0049fbb0: 8000f2d0 0049fc00 6f6c6c61 00632e63
+	 */
+	for (i = 0; i < stack_words; i += 4) {
+		hex_dump_to_buffer(stack + i, 16, 16, 4,
+				   buf, sizeof(buf), false);
+		dev_printk(level, sdev->dev, "0x%08x: %s\n", stack_ptr + i * 4, buf);
+	}
+
+	if (!xoops->plat_hdr.numaregs)
+		return;
+
+	dev_printk(level, sdev->dev, "AR registers:\n");
+	/* the number of ar registers is a multiple of 4 */
+	for (i = 0; i < xoops->plat_hdr.numaregs; i += 4) {
+		hex_dump_to_buffer(xoops->ar + i, 16, 16, 4,
+				   buf, sizeof(buf), false);
+		dev_printk(level, sdev->dev, "%#x: %s\n", i * 4, buf);
+	}
+}
+
+const struct dsp_arch_ops sof_xtensa_arch_ops = {
+	.dsp_oops = xtensa_dsp_oops,
+	.dsp_stack = xtensa_stack,
+};
+EXPORT_SYMBOL_NS(sof_xtensa_arch_ops, "SND_SOC_SOF_XTENSA");
+
+MODULE_LICENSE("Dual BSD/GPL");
+MODULE_DESCRIPTION("SOF Xtensa DSP support");