firmware: arm_scmi: Add SCMI System Power Control driver
authorCristian Marussi <cristian.marussi@arm.com>
Mon, 4 Jul 2022 10:19:33 +0000 (11:19 +0100)
committerSudeep Holla <sudeep.holla@arm.com>
Mon, 4 Jul 2022 13:28:42 +0000 (14:28 +0100)
Add an SCMI System Power control driver to handle platform's requests
carried by SYSTEM_POWER_STATE_NOTIFIER notifications: such platform
requested system-wide power state transitions are handled accordingly,
gracefully or forcefully, depending on the notifications' message flags.

Graceful requests are relayed to userspace using the same Kernel API used
to handle ACPI Shutdown bus events.

Link: https://lore.kernel.org/r/20220704101933.2981635-5-cristian.marussi@arm.com
Reviewed-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Cristian Marussi <cristian.marussi@arm.com>
Signed-off-by: Sudeep Holla <sudeep.holla@arm.com>
drivers/firmware/arm_scmi/Kconfig
drivers/firmware/arm_scmi/Makefile
drivers/firmware/arm_scmi/scmi_power_control.c [new file with mode: 0644]

index 1e7b7fe..a14f654 100644 (file)
@@ -149,4 +149,16 @@ config ARM_SCMI_POWER_DOMAIN
          will be called scmi_pm_domain. Note this may needed early in boot
          before rootfs may be available.
 
+config ARM_SCMI_POWER_CONTROL
+       tristate "SCMI system power control driver"
+       depends on ARM_SCMI_PROTOCOL || (COMPILE_TEST && OF)
+       help
+         This enables System Power control logic which binds system shutdown or
+         reboot actions to SCMI System Power notifications generated by SCP
+         firmware.
+
+         This driver can also be built as a module.  If so, the module will be
+         called scmi_power_control. Note this may needed early in boot to catch
+         early shutdown/reboot SCMI requests.
+
 endmenu
index 8d4afad..faacd0e 100644 (file)
@@ -12,6 +12,7 @@ scmi-module-objs := $(scmi-bus-y) $(scmi-driver-y) $(scmi-protocols-y) \
                    $(scmi-transport-y)
 obj-$(CONFIG_ARM_SCMI_PROTOCOL) += scmi-module.o
 obj-$(CONFIG_ARM_SCMI_POWER_DOMAIN) += scmi_pm_domain.o
+obj-$(CONFIG_ARM_SCMI_POWER_CONTROL) += scmi_power_control.o
 
 ifeq ($(CONFIG_THUMB2_KERNEL)$(CONFIG_CC_IS_CLANG),yy)
 # The use of R7 in the SMCCC conflicts with the compiler's use of R7 as a frame
diff --git a/drivers/firmware/arm_scmi/scmi_power_control.c b/drivers/firmware/arm_scmi/scmi_power_control.c
new file mode 100644 (file)
index 0000000..6eb7d2a
--- /dev/null
@@ -0,0 +1,362 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * SCMI Generic SystemPower Control driver.
+ *
+ * Copyright (C) 2020-2022 ARM Ltd.
+ */
+/*
+ * In order to handle platform originated SCMI SystemPower requests (like
+ * shutdowns or cold/warm resets) we register an SCMI Notification notifier
+ * block to react when such SCMI SystemPower events are emitted by platform.
+ *
+ * Once such a notification is received we act accordingly to perform the
+ * required system transition depending on the kind of request.
+ *
+ * Graceful requests are routed to userspace through the same API methods
+ * (orderly_poweroff/reboot()) used by ACPI when handling ACPI Shutdown bus
+ * events.
+ *
+ * Direct forceful requests are not supported since are not meant to be sent
+ * by the SCMI platform to an OSPM like Linux.
+ *
+ * Additionally, graceful request notifications can carry an optional timeout
+ * field stating the maximum amount of time allowed by the platform for
+ * completion after which they are converted to forceful ones: the assumption
+ * here is that even graceful requests can be upper-bound by a maximum final
+ * timeout strictly enforced by the platform itself which can ultimately cut
+ * the power off at will anytime; in order to avoid such extreme scenario, we
+ * track progress of graceful requests through the means of a reboot notifier
+ * converting timed-out graceful requests to forceful ones, so at least we
+ * try to perform a clean sync and shutdown/restart before the power is cut.
+ *
+ * Given the peculiar nature of SCMI SystemPower protocol, that is being in
+ * charge of triggering system wide shutdown/reboot events, there should be
+ * only one SCMI platform actively emitting SystemPower events.
+ * For this reason the SCMI core takes care to enforce the creation of one
+ * single unique device associated to the SCMI System Power protocol; no matter
+ * how many SCMI platforms are defined on the system, only one can be designated
+ * to support System Power: as a consequence this driver will never be probed
+ * more than once.
+ *
+ * For similar reasons as soon as the first valid SystemPower is received by
+ * this driver and the shutdown/reboot is started, any further notification
+ * possibly emitted by the platform will be ignored.
+ */
+
+#include <linux/math.h>
+#include <linux/module.h>
+#include <linux/mutex.h>
+#include <linux/printk.h>
+#include <linux/reboot.h>
+#include <linux/scmi_protocol.h>
+#include <linux/slab.h>
+#include <linux/time64.h>
+#include <linux/timer.h>
+#include <linux/types.h>
+#include <linux/workqueue.h>
+
+#ifndef MODULE
+#include <linux/fs.h>
+#endif
+
+enum scmi_syspower_state {
+       SCMI_SYSPOWER_IDLE,
+       SCMI_SYSPOWER_IN_PROGRESS,
+       SCMI_SYSPOWER_REBOOTING
+};
+
+/**
+ * struct scmi_syspower_conf  -  Common configuration
+ *
+ * @dev: A reference device
+ * @state: Current SystemPower state
+ * @state_mtx: @state related mutex
+ * @required_transition: The requested transition as decribed in the received
+ *                      SCMI SystemPower notification
+ * @userspace_nb: The notifier_block registered against the SCMI SystemPower
+ *               notification to start the needed userspace interactions.
+ * @reboot_nb: A notifier_block optionally used to track reboot progress
+ * @forceful_work: A worker used to trigger a forceful transition once a
+ *                graceful has timed out.
+ */
+struct scmi_syspower_conf {
+       struct device *dev;
+       enum scmi_syspower_state state;
+       /* Protect access to state */
+       struct mutex state_mtx;
+       enum scmi_system_events required_transition;
+
+       struct notifier_block userspace_nb;
+       struct notifier_block reboot_nb;
+
+       struct delayed_work forceful_work;
+};
+
+#define userspace_nb_to_sconf(x)       \
+       container_of(x, struct scmi_syspower_conf, userspace_nb)
+
+#define reboot_nb_to_sconf(x)          \
+       container_of(x, struct scmi_syspower_conf, reboot_nb)
+
+#define dwork_to_sconf(x)              \
+       container_of(x, struct scmi_syspower_conf, forceful_work)
+
+/**
+ * scmi_reboot_notifier  - A reboot notifier to catch an ongoing successful
+ * system transition
+ * @nb: Reference to the related notifier block
+ * @reason: The reason for the ongoing reboot
+ * @__unused: The cmd being executed on a restart request (unused)
+ *
+ * When an ongoing system transition is detected, compatible with the one
+ * requested by SCMI, cancel the delayed work.
+ *
+ * Return: NOTIFY_OK in any case
+ */
+static int scmi_reboot_notifier(struct notifier_block *nb,
+                               unsigned long reason, void *__unused)
+{
+       struct scmi_syspower_conf *sc = reboot_nb_to_sconf(nb);
+
+       mutex_lock(&sc->state_mtx);
+       switch (reason) {
+       case SYS_HALT:
+       case SYS_POWER_OFF:
+               if (sc->required_transition == SCMI_SYSTEM_SHUTDOWN)
+                       sc->state = SCMI_SYSPOWER_REBOOTING;
+               break;
+       case SYS_RESTART:
+               if (sc->required_transition == SCMI_SYSTEM_COLDRESET ||
+                   sc->required_transition == SCMI_SYSTEM_WARMRESET)
+                       sc->state = SCMI_SYSPOWER_REBOOTING;
+               break;
+       default:
+               break;
+       }
+
+       if (sc->state == SCMI_SYSPOWER_REBOOTING) {
+               dev_dbg(sc->dev, "Reboot in progress...cancel delayed work.\n");
+               cancel_delayed_work_sync(&sc->forceful_work);
+       }
+       mutex_unlock(&sc->state_mtx);
+
+       return NOTIFY_OK;
+}
+
+/**
+ * scmi_request_forceful_transition  - Request forceful SystemPower transition
+ * @sc: A reference to the configuration data
+ *
+ * Initiates the required SystemPower transition without involving userspace:
+ * just trigger the action at the kernel level after issuing an emergency
+ * sync. (if possible at all)
+ */
+static inline void
+scmi_request_forceful_transition(struct scmi_syspower_conf *sc)
+{
+       dev_dbg(sc->dev, "Serving forceful request:%d\n",
+               sc->required_transition);
+
+#ifndef MODULE
+       emergency_sync();
+#endif
+       switch (sc->required_transition) {
+       case SCMI_SYSTEM_SHUTDOWN:
+               kernel_power_off();
+               break;
+       case SCMI_SYSTEM_COLDRESET:
+       case SCMI_SYSTEM_WARMRESET:
+               kernel_restart(NULL);
+               break;
+       default:
+               break;
+       }
+}
+
+static void scmi_forceful_work_func(struct work_struct *work)
+{
+       struct scmi_syspower_conf *sc;
+       struct delayed_work *dwork;
+
+       if (system_state > SYSTEM_RUNNING)
+               return;
+
+       dwork = to_delayed_work(work);
+       sc = dwork_to_sconf(dwork);
+
+       dev_dbg(sc->dev, "Graceful request timed out...forcing !\n");
+       mutex_lock(&sc->state_mtx);
+       /* avoid deadlock by unregistering reboot notifier first */
+       unregister_reboot_notifier(&sc->reboot_nb);
+       if (sc->state == SCMI_SYSPOWER_IN_PROGRESS)
+               scmi_request_forceful_transition(sc);
+       mutex_unlock(&sc->state_mtx);
+}
+
+/**
+ * scmi_request_graceful_transition  - Request graceful SystemPower transition
+ * @sc: A reference to the configuration data
+ * @timeout_ms: The desired timeout to wait for the shutdown to complete before
+ *             system is forcibly shutdown.
+ *
+ * Initiates the required SystemPower transition, requesting userspace
+ * co-operation: it uses the same orderly_ methods used by ACPI Shutdown event
+ * processing.
+ *
+ * Takes care also to register a reboot notifier and to schedule a delayed work
+ * in order to detect if userspace actions are taking too long and in such a
+ * case to trigger a forceful transition.
+ */
+static void scmi_request_graceful_transition(struct scmi_syspower_conf *sc,
+                                            unsigned int timeout_ms)
+{
+       unsigned int adj_timeout_ms = 0;
+
+       if (timeout_ms) {
+               int ret;
+
+               sc->reboot_nb.notifier_call = &scmi_reboot_notifier;
+               ret = register_reboot_notifier(&sc->reboot_nb);
+               if (!ret) {
+                       /* Wait only up to 75% of the advertised timeout */
+                       adj_timeout_ms = mult_frac(timeout_ms, 3, 4);
+                       INIT_DELAYED_WORK(&sc->forceful_work,
+                                         scmi_forceful_work_func);
+                       schedule_delayed_work(&sc->forceful_work,
+                                             msecs_to_jiffies(adj_timeout_ms));
+               } else {
+                       /* Carry on best effort even without a reboot notifier */
+                       dev_warn(sc->dev,
+                                "Cannot register reboot notifier !\n");
+               }
+       }
+
+       dev_dbg(sc->dev,
+               "Serving graceful req:%d (timeout_ms:%u  adj_timeout_ms:%u)\n",
+               sc->required_transition, timeout_ms, adj_timeout_ms);
+
+       switch (sc->required_transition) {
+       case SCMI_SYSTEM_SHUTDOWN:
+               /*
+                * When triggered early at boot-time the 'orderly' call will
+                * partially fail due to the lack of userspace itself, but
+                * the force=true argument will start anyway a successful
+                * forced shutdown.
+                */
+               orderly_poweroff(true);
+               break;
+       case SCMI_SYSTEM_COLDRESET:
+       case SCMI_SYSTEM_WARMRESET:
+               orderly_reboot();
+               break;
+       default:
+               break;
+       }
+}
+
+/**
+ * scmi_userspace_notifier  - Notifier callback to act on SystemPower
+ * Notifications
+ * @nb: Reference to the related notifier block
+ * @event: The SystemPower notification event id
+ * @data: The SystemPower event report
+ *
+ * This callback is in charge of decoding the received SystemPower report
+ * and act accordingly triggering a graceful or forceful system transition.
+ *
+ * Note that once a valid SCMI SystemPower event starts being served, any
+ * other following SystemPower notification received from the same SCMI
+ * instance (handle) will be ignored.
+ *
+ * Return: NOTIFY_OK once a valid SystemPower event has been successfully
+ * processed.
+ */
+static int scmi_userspace_notifier(struct notifier_block *nb,
+                                  unsigned long event, void *data)
+{
+       struct scmi_system_power_state_notifier_report *er = data;
+       struct scmi_syspower_conf *sc = userspace_nb_to_sconf(nb);
+
+       if (er->system_state >= SCMI_SYSTEM_POWERUP) {
+               dev_err(sc->dev, "Ignoring unsupported system_state: 0x%X\n",
+                       er->system_state);
+               return NOTIFY_DONE;
+       }
+
+       if (!SCMI_SYSPOWER_IS_REQUEST_GRACEFUL(er->flags)) {
+               dev_err(sc->dev, "Ignoring forceful notification.\n");
+               return NOTIFY_DONE;
+       }
+
+       /*
+        * Bail out if system is already shutting down or an SCMI SystemPower
+        * requested is already being served.
+        */
+       if (system_state > SYSTEM_RUNNING)
+               return NOTIFY_DONE;
+       mutex_lock(&sc->state_mtx);
+       if (sc->state != SCMI_SYSPOWER_IDLE) {
+               dev_dbg(sc->dev,
+                       "Transition already in progress...ignore.\n");
+               mutex_unlock(&sc->state_mtx);
+               return NOTIFY_DONE;
+       }
+       sc->state = SCMI_SYSPOWER_IN_PROGRESS;
+       mutex_unlock(&sc->state_mtx);
+
+       sc->required_transition = er->system_state;
+
+       /* Leaving a trace in logs of who triggered the shutdown/reboot. */
+       dev_info(sc->dev, "Serving shutdown/reboot request: %d\n",
+                sc->required_transition);
+
+       scmi_request_graceful_transition(sc, er->timeout);
+
+       return NOTIFY_OK;
+}
+
+static int scmi_syspower_probe(struct scmi_device *sdev)
+{
+       int ret;
+       struct scmi_syspower_conf *sc;
+       struct scmi_handle *handle = sdev->handle;
+
+       if (!handle)
+               return -ENODEV;
+
+       ret = handle->devm_protocol_acquire(sdev, SCMI_PROTOCOL_SYSTEM);
+       if (ret)
+               return ret;
+
+       sc = devm_kzalloc(&sdev->dev, sizeof(*sc), GFP_KERNEL);
+       if (!sc)
+               return -ENOMEM;
+
+       sc->state = SCMI_SYSPOWER_IDLE;
+       mutex_init(&sc->state_mtx);
+       sc->required_transition = SCMI_SYSTEM_MAX;
+       sc->userspace_nb.notifier_call = &scmi_userspace_notifier;
+       sc->dev = &sdev->dev;
+
+       return handle->notify_ops->devm_event_notifier_register(sdev,
+                                                          SCMI_PROTOCOL_SYSTEM,
+                                        SCMI_EVENT_SYSTEM_POWER_STATE_NOTIFIER,
+                                                      NULL, &sc->userspace_nb);
+}
+
+static const struct scmi_device_id scmi_id_table[] = {
+       { SCMI_PROTOCOL_SYSTEM, "syspower" },
+       { },
+};
+MODULE_DEVICE_TABLE(scmi, scmi_id_table);
+
+static struct scmi_driver scmi_system_power_driver = {
+       .name = "scmi-system-power",
+       .probe = scmi_syspower_probe,
+       .id_table = scmi_id_table,
+};
+module_scmi_driver(scmi_system_power_driver);
+
+MODULE_AUTHOR("Cristian Marussi <cristian.marussi@arm.com>");
+MODULE_DESCRIPTION("ARM SCMI SystemPower Control driver");
+MODULE_LICENSE("GPL");