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, 259 insertions, 0 deletions

diff --git a/drivers/crypto/ccp/dbc.c b/drivers/crypto/ccp/dbc.c
new file mode 100644
index 000000000..410084a90
--- /dev/null
+++ b/drivers/crypto/ccp/dbc.c
@@ -0,0 +1,259 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * AMD Secure Processor Dynamic Boost Control interface
+ *
+ * Copyright (C) 2023 Advanced Micro Devices, Inc.
+ *
+ * Author: Mario Limonciello <mario.limonciello@amd.com>
+ */
+
+#include <linux/mutex.h>
+
+#include "dbc.h"
+
+#define DBC_DEFAULT_TIMEOUT		(10 * MSEC_PER_SEC)
+struct error_map {
+	u32 psp;
+	int ret;
+};
+
+#define DBC_ERROR_ACCESS_DENIED		0x0001
+#define DBC_ERROR_EXCESS_DATA		0x0004
+#define DBC_ERROR_BAD_PARAMETERS	0x0006
+#define DBC_ERROR_BAD_STATE		0x0007
+#define DBC_ERROR_NOT_IMPLEMENTED	0x0009
+#define DBC_ERROR_BUSY			0x000D
+#define DBC_ERROR_MESSAGE_FAILURE	0x0307
+#define DBC_ERROR_OVERFLOW		0x300F
+#define DBC_ERROR_SIGNATURE_INVALID	0x3072
+
+static struct error_map error_codes[] = {
+	{DBC_ERROR_ACCESS_DENIED,	-EACCES},
+	{DBC_ERROR_EXCESS_DATA,		-E2BIG},
+	{DBC_ERROR_BAD_PARAMETERS,	-EINVAL},
+	{DBC_ERROR_BAD_STATE,		-EAGAIN},
+	{DBC_ERROR_MESSAGE_FAILURE,	-ENOENT},
+	{DBC_ERROR_NOT_IMPLEMENTED,	-ENOENT},
+	{DBC_ERROR_BUSY,		-EBUSY},
+	{DBC_ERROR_OVERFLOW,		-ENFILE},
+	{DBC_ERROR_SIGNATURE_INVALID,	-EPERM},
+	{0x0,	0x0},
+};
+
+static inline int send_dbc_cmd_thru_ext(struct psp_dbc_device *dbc_dev, int msg)
+{
+	dbc_dev->mbox->ext_req.header.sub_cmd_id = msg;
+
+	return psp_extended_mailbox_cmd(dbc_dev->psp,
+					DBC_DEFAULT_TIMEOUT,
+					(struct psp_ext_request *)dbc_dev->mbox);
+}
+
+static inline int send_dbc_cmd_thru_pa(struct psp_dbc_device *dbc_dev, int msg)
+{
+	return psp_send_platform_access_msg(msg,
+					    (struct psp_request *)dbc_dev->mbox);
+}
+
+static int send_dbc_cmd(struct psp_dbc_device *dbc_dev, int msg)
+{
+	int ret;
+
+	*dbc_dev->result = 0;
+	ret = dbc_dev->use_ext ? send_dbc_cmd_thru_ext(dbc_dev, msg) :
+				 send_dbc_cmd_thru_pa(dbc_dev, msg);
+	if (ret == -EIO) {
+		int i;
+
+		dev_dbg(dbc_dev->dev,
+			 "msg 0x%x failed with PSP error: 0x%x\n",
+			 msg, *dbc_dev->result);
+
+		for (i = 0; error_codes[i].psp; i++) {
+			if (*dbc_dev->result == error_codes[i].psp)
+				return error_codes[i].ret;
+		}
+	}
+
+	return ret;
+}
+
+static int send_dbc_nonce(struct psp_dbc_device *dbc_dev)
+{
+	int ret;
+
+	*dbc_dev->payload_size = dbc_dev->header_size + sizeof(struct dbc_user_nonce);
+	ret = send_dbc_cmd(dbc_dev, PSP_DYNAMIC_BOOST_GET_NONCE);
+	if (ret == -EAGAIN) {
+		dev_dbg(dbc_dev->dev, "retrying get nonce\n");
+		ret = send_dbc_cmd(dbc_dev, PSP_DYNAMIC_BOOST_GET_NONCE);
+	}
+
+	return ret;
+}
+
+static int send_dbc_parameter(struct psp_dbc_device *dbc_dev)
+{
+	struct dbc_user_param *user_param = (struct dbc_user_param *)dbc_dev->payload;
+
+	switch (user_param->msg_index) {
+	case PARAM_SET_FMAX_CAP:
+	case PARAM_SET_PWR_CAP:
+	case PARAM_SET_GFX_MODE:
+		return send_dbc_cmd(dbc_dev, PSP_DYNAMIC_BOOST_SET_PARAMETER);
+	case PARAM_GET_FMAX_CAP:
+	case PARAM_GET_PWR_CAP:
+	case PARAM_GET_CURR_TEMP:
+	case PARAM_GET_FMAX_MAX:
+	case PARAM_GET_FMAX_MIN:
+	case PARAM_GET_SOC_PWR_MAX:
+	case PARAM_GET_SOC_PWR_MIN:
+	case PARAM_GET_SOC_PWR_CUR:
+	case PARAM_GET_GFX_MODE:
+		return send_dbc_cmd(dbc_dev, PSP_DYNAMIC_BOOST_GET_PARAMETER);
+	}
+
+	return -EINVAL;
+}
+
+void dbc_dev_destroy(struct psp_device *psp)
+{
+	struct psp_dbc_device *dbc_dev = psp->dbc_data;
+
+	if (!dbc_dev)
+		return;
+
+	misc_deregister(&dbc_dev->char_dev);
+	mutex_destroy(&dbc_dev->ioctl_mutex);
+	psp->dbc_data = NULL;
+}
+
+static long dbc_ioctl(struct file *filp, unsigned int cmd, unsigned long arg)
+{
+	struct psp_device *psp_master = psp_get_master_device();
+	void __user *argp = (void __user *)arg;
+	struct psp_dbc_device *dbc_dev;
+	int ret;
+
+	if (!psp_master || !psp_master->dbc_data)
+		return -ENODEV;
+	dbc_dev = psp_master->dbc_data;
+
+	guard(mutex)(&dbc_dev->ioctl_mutex);
+
+	switch (cmd) {
+	case DBCIOCNONCE:
+		if (copy_from_user(dbc_dev->payload, argp, sizeof(struct dbc_user_nonce)))
+			return -EFAULT;
+
+		ret = send_dbc_nonce(dbc_dev);
+		if (ret)
+			return ret;
+
+		if (copy_to_user(argp, dbc_dev->payload, sizeof(struct dbc_user_nonce)))
+			return -EFAULT;
+		break;
+	case DBCIOCUID:
+		if (copy_from_user(dbc_dev->payload, argp, sizeof(struct dbc_user_setuid)))
+			return -EFAULT;
+
+		*dbc_dev->payload_size = dbc_dev->header_size + sizeof(struct dbc_user_setuid);
+		ret = send_dbc_cmd(dbc_dev, PSP_DYNAMIC_BOOST_SET_UID);
+		if (ret)
+			return ret;
+
+		if (copy_to_user(argp, dbc_dev->payload, sizeof(struct dbc_user_setuid)))
+			return -EFAULT;
+		break;
+	case DBCIOCPARAM:
+		if (copy_from_user(dbc_dev->payload, argp, sizeof(struct dbc_user_param)))
+			return -EFAULT;
+
+		*dbc_dev->payload_size = dbc_dev->header_size + sizeof(struct dbc_user_param);
+		ret = send_dbc_parameter(dbc_dev);
+		if (ret)
+			return ret;
+
+		if (copy_to_user(argp, dbc_dev->payload, sizeof(struct dbc_user_param)))
+			return -EFAULT;
+		break;
+	default:
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+static const struct file_operations dbc_fops = {
+	.owner	= THIS_MODULE,
+	.unlocked_ioctl = dbc_ioctl,
+};
+
+int dbc_dev_init(struct psp_device *psp)
+{
+	struct device *dev = psp->dev;
+	struct psp_dbc_device *dbc_dev;
+	int ret;
+
+	dbc_dev = devm_kzalloc(dev, sizeof(*dbc_dev), GFP_KERNEL);
+	if (!dbc_dev)
+		return -ENOMEM;
+
+	BUILD_BUG_ON(sizeof(union dbc_buffer) > PAGE_SIZE);
+	dbc_dev->mbox = (void *)devm_get_free_pages(dev, GFP_KERNEL | __GFP_ZERO, 0);
+	if (!dbc_dev->mbox) {
+		ret = -ENOMEM;
+		goto cleanup_dev;
+	}
+
+	psp->dbc_data = dbc_dev;
+	dbc_dev->dev = dev;
+	dbc_dev->psp = psp;
+
+	if (psp->capability.dbc_thru_ext) {
+		dbc_dev->use_ext = true;
+		dbc_dev->payload_size = &dbc_dev->mbox->ext_req.header.payload_size;
+		dbc_dev->result = &dbc_dev->mbox->ext_req.header.status;
+		dbc_dev->payload = &dbc_dev->mbox->ext_req.buf;
+		dbc_dev->header_size = sizeof(struct psp_ext_req_buffer_hdr);
+	} else {
+		dbc_dev->payload_size = &dbc_dev->mbox->pa_req.header.payload_size;
+		dbc_dev->result = &dbc_dev->mbox->pa_req.header.status;
+		dbc_dev->payload = &dbc_dev->mbox->pa_req.buf;
+		dbc_dev->header_size = sizeof(struct psp_req_buffer_hdr);
+	}
+
+	ret = send_dbc_nonce(dbc_dev);
+	if (ret == -EACCES) {
+		dev_dbg(dbc_dev->dev,
+			"dynamic boost control was previously authenticated\n");
+		ret = 0;
+	}
+	dev_dbg(dbc_dev->dev, "dynamic boost control is %savailable\n",
+		ret ? "un" : "");
+	if (ret) {
+		ret = 0;
+		goto cleanup_mbox;
+	}
+
+	dbc_dev->char_dev.minor = MISC_DYNAMIC_MINOR;
+	dbc_dev->char_dev.name = "dbc";
+	dbc_dev->char_dev.fops = &dbc_fops;
+	dbc_dev->char_dev.mode = 0600;
+	ret = misc_register(&dbc_dev->char_dev);
+	if (ret)
+		goto cleanup_mbox;
+
+	mutex_init(&dbc_dev->ioctl_mutex);
+
+	return 0;
+
+cleanup_mbox:
+	devm_free_pages(dev, (unsigned long)dbc_dev->mbox);
+
+cleanup_dev:
+	psp->dbc_data = NULL;
+	devm_kfree(dev, dbc_dev);
+
+	return ret;
+}