ACPI / sleep: Ignore spurious SCI wakeups from suspend-to-idle
authorRafael J. Wysocki <rafael.j.wysocki@intel.com>
Wed, 26 Apr 2017 21:23:03 +0000 (23:23 +0200)
committerRafael J. Wysocki <rafael.j.wysocki@intel.com>
Fri, 5 May 2017 20:54:28 +0000 (22:54 +0200)
The ACPI SCI (System Control Interrupt) is set up as a wakeup IRQ
during suspend-to-idle transitions and, consequently, any events
signaled through it wake up the system from that state.  However,
on some systems some of the events signaled via the ACPI SCI while
suspended to idle should not cause the system to wake up.  In fact,
quite often they should just be discarded.

Arguably, systems should not resume entirely on such events, but in
order to decide which events really should cause the system to resume
and which are spurious, it is necessary to resume up to the point
when ACPI SCIs are actually handled and processed, which is after
executing dpm_resume_noirq() in the system resume path.

For this reasons, add a loop around freeze_enter() in which the
platforms can process events signaled via multiplexed IRQ lines
like the ACPI SCI and add suspend-to-idle hooks that can be
used for this purpose to struct platform_freeze_ops.

In the ACPI case, the ->wake hook is used for checking if the SCI
has triggered while suspended and deferring the interrupt-induced
system wakeup until the events signaled through it are actually
processed sufficiently to decide whether or not the system should
resume.  In turn, the ->sync hook allows all of the relevant event
queues to be flushed so as to prevent events from being missed due
to race conditions.

In addition to that, some ACPI code processing wakeup events needs
to be modified to use the "hard" version of wakeup triggers, so that
it will cause a system resume to happen on device-induced wakeup
events even if the "soft" mechanism to prevent the system from
suspending is not enabled (that also helps to catch device-induced
wakeup events occurring during suspend transitions in progress).

Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
drivers/acpi/battery.c
drivers/acpi/button.c
drivers/acpi/device_pm.c
drivers/acpi/sleep.c
drivers/base/power/main.c
drivers/base/power/wakeup.c
include/linux/suspend.h
kernel/power/process.c
kernel/power/suspend.c

index 4ef1e4624b2b4bcee2a59666e5929bd915fb40a2..83ab17e4a795e89cad8f55c75a48944ed773fb4a 100644 (file)
@@ -776,7 +776,7 @@ static int acpi_battery_update(struct acpi_battery *battery, bool resume)
        if ((battery->state & ACPI_BATTERY_STATE_CRITICAL) ||
            (test_bit(ACPI_BATTERY_ALARM_PRESENT, &battery->flags) &&
             (battery->capacity_now <= battery->alarm)))
-               pm_wakeup_event(&battery->device->dev, 0);
+               pm_wakeup_hard_event(&battery->device->dev);
 
        return result;
 }
index 668137e4a0697cf230b074e13cbb6dd255c7bf59..b7c2a06963d6fb79cb5f5aaa2576936d17d5088f 100644 (file)
@@ -216,7 +216,7 @@ static int acpi_lid_notify_state(struct acpi_device *device, int state)
        }
 
        if (state)
-               pm_wakeup_event(&device->dev, 0);
+               pm_wakeup_hard_event(&device->dev);
 
        ret = blocking_notifier_call_chain(&acpi_lid_notifier, state, device);
        if (ret == NOTIFY_DONE)
@@ -398,7 +398,7 @@ static void acpi_button_notify(struct acpi_device *device, u32 event)
                } else {
                        int keycode;
 
-                       pm_wakeup_event(&device->dev, 0);
+                       pm_wakeup_hard_event(&device->dev);
                        if (button->suspended)
                                break;
 
@@ -530,6 +530,7 @@ static int acpi_button_add(struct acpi_device *device)
                lid_device = device;
        }
 
+       device_init_wakeup(&device->dev, true);
        printk(KERN_INFO PREFIX "%s [%s]\n", name, acpi_device_bid(device));
        return 0;
 
index 993fd31394c854c99e5ce0c2af824f36c50b7a22..798d5003a039d876f275fc2d933be71cb7ebfbed 100644 (file)
@@ -24,6 +24,7 @@
 #include <linux/pm_qos.h>
 #include <linux/pm_domain.h>
 #include <linux/pm_runtime.h>
+#include <linux/suspend.h>
 
 #include "internal.h"
 
@@ -399,7 +400,7 @@ static void acpi_pm_notify_handler(acpi_handle handle, u32 val, void *not_used)
        mutex_lock(&acpi_pm_notifier_lock);
 
        if (adev->wakeup.flags.notifier_present) {
-               __pm_wakeup_event(adev->wakeup.ws, 0);
+               pm_wakeup_ws_event(adev->wakeup.ws, 0, true);
                if (adev->wakeup.context.work.func)
                        queue_pm_work(&adev->wakeup.context.work);
        }
index a4327af676fe81948cdb76c3b15786256929ffa8..e84005d642e6a8bcbf36f39f00c8bd3cf7c7b9e8 100644 (file)
@@ -662,14 +662,40 @@ static int acpi_freeze_prepare(void)
        acpi_os_wait_events_complete();
        if (acpi_sci_irq_valid())
                enable_irq_wake(acpi_sci_irq);
+
        return 0;
 }
 
+static void acpi_freeze_wake(void)
+{
+       /*
+        * If IRQD_WAKEUP_ARMED is not set for the SCI at this point, it means
+        * that the SCI has triggered while suspended, so cancel the wakeup in
+        * case it has not been a wakeup event (the GPEs will be checked later).
+        */
+       if (acpi_sci_irq_valid() &&
+           !irqd_is_wakeup_armed(irq_get_irq_data(acpi_sci_irq)))
+               pm_system_cancel_wakeup();
+}
+
+static void acpi_freeze_sync(void)
+{
+       /*
+        * Process all pending events in case there are any wakeup ones.
+        *
+        * The EC driver uses the system workqueue, so that one needs to be
+        * flushed too.
+        */
+       acpi_os_wait_events_complete();
+       flush_scheduled_work();
+}
+
 static void acpi_freeze_restore(void)
 {
        acpi_disable_wakeup_devices(ACPI_STATE_S0);
        if (acpi_sci_irq_valid())
                disable_irq_wake(acpi_sci_irq);
+
        acpi_enable_all_runtime_gpes();
 }
 
@@ -681,6 +707,8 @@ static void acpi_freeze_end(void)
 static const struct platform_freeze_ops acpi_freeze_ops = {
        .begin = acpi_freeze_begin,
        .prepare = acpi_freeze_prepare,
+       .wake = acpi_freeze_wake,
+       .sync = acpi_freeze_sync,
        .restore = acpi_freeze_restore,
        .end = acpi_freeze_end,
 };
index 9faee1c893e53c8dea6e14d472a73a8b7131bf96..e987a6f55d36747f79b470b0372a1abcff6390d7 100644 (file)
@@ -1091,11 +1091,6 @@ static int __device_suspend_noirq(struct device *dev, pm_message_t state, bool a
        if (async_error)
                goto Complete;
 
-       if (pm_wakeup_pending()) {
-               async_error = -EBUSY;
-               goto Complete;
-       }
-
        if (dev->power.syscore || dev->power.direct_complete)
                goto Complete;
 
index 84c41c98d6fce03788187cfba49d3730670f1dc6..f62082fdd6703e11c7576dc5db673dbc7dbba056 100644 (file)
@@ -28,8 +28,8 @@ bool events_check_enabled __read_mostly;
 /* First wakeup IRQ seen by the kernel in the last cycle. */
 unsigned int pm_wakeup_irq __read_mostly;
 
-/* If set and the system is suspending, terminate the suspend. */
-static bool pm_abort_suspend __read_mostly;
+/* If greater than 0 and the system is suspending, terminate the suspend. */
+static atomic_t pm_abort_suspend __read_mostly;
 
 /*
  * Combined counters of registered wakeup events and wakeup events in progress.
@@ -856,20 +856,26 @@ bool pm_wakeup_pending(void)
                pm_print_active_wakeup_sources();
        }
 
-       return ret || pm_abort_suspend;
+       return ret || atomic_read(&pm_abort_suspend) > 0;
 }
 
 void pm_system_wakeup(void)
 {
-       pm_abort_suspend = true;
+       atomic_inc(&pm_abort_suspend);
        freeze_wake();
 }
 EXPORT_SYMBOL_GPL(pm_system_wakeup);
 
-void pm_wakeup_clear(void)
+void pm_system_cancel_wakeup(void)
+{
+       atomic_dec(&pm_abort_suspend);
+}
+
+void pm_wakeup_clear(bool reset)
 {
-       pm_abort_suspend = false;
        pm_wakeup_irq = 0;
+       if (reset)
+               atomic_set(&pm_abort_suspend, 0);
 }
 
 void pm_system_irq_wakeup(unsigned int irq_number)
index d9718378a8bee0b327d08c2e80a6fd3b5490b967..0b1cf32edfd7ba1c456252124e23c68450d5bcc3 100644 (file)
@@ -189,6 +189,8 @@ struct platform_suspend_ops {
 struct platform_freeze_ops {
        int (*begin)(void);
        int (*prepare)(void);
+       void (*wake)(void);
+       void (*sync)(void);
        void (*restore)(void);
        void (*end)(void);
 };
@@ -428,7 +430,8 @@ extern unsigned int pm_wakeup_irq;
 
 extern bool pm_wakeup_pending(void);
 extern void pm_system_wakeup(void);
-extern void pm_wakeup_clear(void);
+extern void pm_system_cancel_wakeup(void);
+extern void pm_wakeup_clear(bool reset);
 extern void pm_system_irq_wakeup(unsigned int irq_number);
 extern bool pm_get_wakeup_count(unsigned int *count, bool block);
 extern bool pm_save_wakeup_count(unsigned int count);
@@ -478,7 +481,7 @@ static inline int unregister_pm_notifier(struct notifier_block *nb)
 
 static inline bool pm_wakeup_pending(void) { return false; }
 static inline void pm_system_wakeup(void) {}
-static inline void pm_wakeup_clear(void) {}
+static inline void pm_wakeup_clear(bool reset) {}
 static inline void pm_system_irq_wakeup(unsigned int irq_number) {}
 
 static inline void lock_system_sleep(void) {}
index c7209f060eeb7c8672cf8f07ba9c83ac6c9460ac..78672d324a6ef95394ad72a0b0ba29c7d1155d5d 100644 (file)
@@ -132,7 +132,7 @@ int freeze_processes(void)
        if (!pm_freezing)
                atomic_inc(&system_freezing_cnt);
 
-       pm_wakeup_clear();
+       pm_wakeup_clear(true);
        pr_info("Freezing user space processes ... ");
        pm_freezing = true;
        error = try_to_freeze_tasks(true);
index 15e6baef5c73f90b6817c0b1c4e871ea40e30318..c0248c74d6d4cef6dbf09f485f36686862c29094 100644 (file)
@@ -72,6 +72,8 @@ static void freeze_begin(void)
 
 static void freeze_enter(void)
 {
+       trace_suspend_resume(TPS("machine_suspend"), PM_SUSPEND_FREEZE, true);
+
        spin_lock_irq(&suspend_freeze_lock);
        if (pm_wakeup_pending())
                goto out;
@@ -98,6 +100,27 @@ static void freeze_enter(void)
  out:
        suspend_freeze_state = FREEZE_STATE_NONE;
        spin_unlock_irq(&suspend_freeze_lock);
+
+       trace_suspend_resume(TPS("machine_suspend"), PM_SUSPEND_FREEZE, false);
+}
+
+static void s2idle_loop(void)
+{
+       do {
+               freeze_enter();
+
+               if (freeze_ops && freeze_ops->wake)
+                       freeze_ops->wake();
+
+               dpm_resume_noirq(PMSG_RESUME);
+               if (freeze_ops && freeze_ops->sync)
+                       freeze_ops->sync();
+
+               if (pm_wakeup_pending())
+                       break;
+
+               pm_wakeup_clear(false);
+       } while (!dpm_suspend_noirq(PMSG_SUSPEND));
 }
 
 void freeze_wake(void)
@@ -371,10 +394,8 @@ static int suspend_enter(suspend_state_t state, bool *wakeup)
         * all the devices are suspended.
         */
        if (state == PM_SUSPEND_FREEZE) {
-               trace_suspend_resume(TPS("machine_suspend"), state, true);
-               freeze_enter();
-               trace_suspend_resume(TPS("machine_suspend"), state, false);
-               goto Platform_wake;
+               s2idle_loop();
+               goto Platform_early_resume;
        }
 
        error = disable_nonboot_cpus();