power: suspend: Add suspend timeout handler
PATCH 1/1 power: suspend: Add suspend timeout handler
Thank you Greg's promptly reply.
let me try to explain detail in following.
Quoted 1 line · Joseph Jang
From: josephjang <[email protected]>
Please use your name as spelled out like you did above in the email header.
Sure, I will update the patch again like following.
From b3afca8f3ee1d587c188b830d94c683fe66cbfb3 Mon Sep 17 00:00:00 2001
From: Joseph Jang <[email protected]>
Date: Fri, 16 Oct 2020 16:31:38 +0800
Subject: [PATCH] power: suspend: Add suspend timeout handler
Add suspend timeout handler to prevent device stuck during suspend/ resume process. Suspend timeout handler will dump disk sleep task at first round timeout and trigger kernel panic at second round timeout. The default timer for each round is 30 seconds.
Quoted 9 lines
Add suspend timeout handler to prevent device stuck during suspend/ resume process. Suspend timeout handler will dump disk sleep task at first round timeout and trigger kernel panic at second round timeout. The default timer for each round is 30 seconds.
Note: Can use following command to simulate suspend hang for testing.
adb shell echo 1 >/sys/power/pm_hang
adb shell echo mem >/sys/power/stateSigned-off-by josephjang <[email protected]>
Need a blank line before the signed-off-by: and again, spell your name the same way.
Sure, I will update the patch again like following.
Note: Can use following command to simulate suspend hang for testing.
adb shell echo 1 > /sys/power/pm_hang
adb shell echo mem > /sys/power/state
---
Quoted 13 lines
--- include/linux/console.h | 1 + kernel/power/Kconfig | 9 +++ kernel/power/main.c | 66 ++++++++++++++++ kernel/power/suspend.c | 162 ++++++++++++++++++++++++++++++++++++++++ kernel/printk/printk.c | 5 ++ 5 files changed, 243 insertions(+)include/linux/console.h +0
-0
include/linux/console.h static inline void console_sysfs_notify(void) · line 195
Quoted 4 lines
include/linux/console.h +1
-0@@ -192,6 +192,7 @@ static inline void console_sysfs_notify(void) { }#endif extern bool console_suspend_enabled; +extern int is_console_suspended (void );
For global functions, how about:
bool console_is_suspended(void);
?
Agree, I will update like following.
extern bool console_suspend_enabled;
+extern int console_is_suspended(void);
Quoted 18 lines
include/linux/console.h +0
-0/* Suspend and resume console messages over PM events */ extern void suspend_console (void );kernel/power/Kconfig +8
-0@@ -207,6 +207,15 @@ config PM_SLEEP_DEBUG def_bool y depends on PM_DEBUG && PM_SLEEP +config PM_SLEEP_MONITOR + bool "Linux kernel suspend/resume process monitor" + depends on PM_SLEEP + help + This option will enable suspend/resume monitor to prevent device + stuck during suspend/resume process. Suspend timeout handler will + dump disk sleep task at first round timeout and trigger kernel panic + at second round timeout. The default timer for each round is 30
Quoted 1 line
seconds.
Ouch, are you sure you want to panic?
Yes, we would like to trigger kernel panic to reboot system and prevent system hang there.
Quoted 10 lines
kernel/power/Kconfig +1
-0+ config DPM_WATCHDOG bool "Device suspend/resume watchdog" depends on PM_DEBUG && PSTORE && EXPERTkernel/power/main.c +0
-0@@ -575,6 +575,69 @@ void __pm_pr_dbg(bool defer, const char *fmt, ...)static inline void pm_print_times_init (void ) {}
kernel/power/main.c void __pm_pr_dbg(bool defer, const char *fmt, ...) · line 585
Quoted 10 lines
kernel/power/main.c +8
-0#endif /* CONFIG_PM_SLEEP_DEBUG */ +#ifdef CONFIG_PM_SLEEP_MONITOR +/* If set, devices will stuck at suspend for verification */ +static bool pm_hang_enabled; + +static int pm_notify_test (struct notifier_block *nb, +unsigned long mode,void *_unused) +{ +pr_info ("Jump into infinite loop now\n" );
Why do you have debugging code still enabled?
Because we want to make sure system hang was cause by our test code here.
kernel/power/main.c void __pm_pr_dbg(bool defer, const char *fmt, ...) · line 590
Quoted 5 lines
kernel/power/main.c +5
-0+ +/* Suspend thread stuck at a loop forever */ +for (;;) + ; +
Don't busy spin, that will burn power.
Okay, I will add msleep() to prevent system burn power during testing. +static int pm_notify_test(struct notifier_block *nb,
+ unsigned long mode, void *_unused)
+{
+ pr_info("Jump into infinite loop now\n");
+
+ /* Suspend thread stuck at a loop forever */
+ for (;;)
+ msleep(100);
+
+ pr_info("Fail to stuck at loop\n");
+
+ return 0;
+}
kernel/power/main.c void __pm_pr_dbg(bool defer, const char *fmt, ...) · line 591
Quoted 1 line
kernel/power/main.c +1
-0+pr_info ("Fail to stuck at loop\n" );
And how can this happen?
I think we should not see this log unless the for loop has problem.
Sure, I add some description in the document like following.
Documentation/ABI/testing/sysfs-power +15 -0
@@ -420,3 +420,18 @@ Description: disables it. Reads from the file return the current value. The default is "1" if the build-time "SUSPEND_SKIP_SYNC" config flag is unset, or "0" otherwise. + +What: /sys/power/pm_hang +Date: Oct 2020 +Contact: Joseph Jang <[email protected]> +Description: + The /sys/power/pm_hang file controls the variable pm_hang_enabled + which controls whether you need to simulate system hang during + suspend. + + Writing a "1" to this file will set the pm_hang_enabled to "1" and + register a pm_notify function pm_notify_test() to simulate system + hang during suspend. + Writing a "0" to this file will set the pm_hang_enabled to "0" and + unregister a pm_notify function pm_notify_test(). + Reads from this file return the current value of pm_hang_enabled.
Quoted 9 lines
kernel/power/main.c +1
-0+#endif #endif #ifdef CONFIG_FREEZER &pm_freeze_timeout_attr.attr,kernel/power/suspend.c +0
-0
kernel/power/suspend.c line 38
Quoted 9 lines
kernel/power/suspend.c +6
-0@@ -30,6 +30,12 @@#include <trace/events/power.h> #include <linux/compiler.h> #include <linux/moduleparam.h> +#ifdef CONFIG_PM_SLEEP_MONITOR +#include <linux/sched/debug.h> +#include <linux/kthread.h> +#include <linux/sched.h> +#include <uapi/linux/sched/types.h> +#endif
Don't #ifdef include files.
And why the uapi file?
I refer to linux-stable/kernel/rcu/tree.c source code.
In order to prevent build error for MAX_RT_PRIO, I need to include this header file <uapi/linux/sched/types.h>.
Quoted 15 lines
kernel/power/suspend.c +9
-0#include "power.h" @@ -61,6 +67,133 @@ static DECLARE_SWAIT_QUEUE_HEAD(s2idle_wait_head);enum s2idle_states __read_mostly s2idle_state;static DEFINE_RAW_SPINLOCK (s2idle_lock); +#ifdef CONFIG_PM_SLEEP_MONITOR +/* Suspend monitor thread toggle reason */ +enum toggle_reason { +TOGGLE_NONE , +TOGGLE_START , +TOGGLE_STOP , +}; + +#define SUSPEND_TIMER_TIMEOUT_MS 30000
kernel/power/suspend.c static DECLARE_SWAIT_QUEUE_HEAD(s2idle_wait_head); · line 88
Quoted 10 lines
kernel/power/suspend.c +10
-0+static struct task_struct *ksuspend_mon_tsk; +static DECLARE_WAIT_QUEUE_HEAD (power_suspend_waitqueue); +static enum toggle_reason suspend_mon_toggle; +static DEFINE_MUTEX (suspend_mon_lock); + +static void start_suspend_mon (void ) +{ +mutex_lock (&suspend_mon_lock); + suspend_mon_toggle =TOGGLE_START ; +mutex_unlock (&suspend_mon_lock);
Why do you need a lock for a single integer?
Since we have two kernel thread may reference variable "suspend_mon_toggle", one is suspend thread, another is suspend_monitor_kthread. We use it to start/stop suspend monitor.
Yes, in order to notify user this kernel panic was triggered by suspend_timeout(), I add debug code to prevent confused.
kernel/power/suspend.c static DECLARE_SWAIT_QUEUE_HEAD(s2idle_wait_head); · line 91
Quoted 3 lines
kernel/power/suspend.c +3
-0+ +/* Trigger a NULL pointer dereference */ + *null_pointer ='a' ;
Are you sure this will work on all platforms? We do have a panic function if you really want to do that.
I agree to change to use panic() like following. +static void suspend_timeout(int timeout_count)
+{
+ pr_info("Suspend monitor timeout (timer is %d seconds)\n",
+ (SUSPEND_TIMER_TIMEOUT_MS/1000));
+
+ show_state_filter(TASK_UNINTERRUPTIBLE);
+
+ if (timeout_count < 2)
+ return;
+
+ if (console_is_suspended())
+ resume_console();
+
+ /* Trigger a NULL pointer dereference */
+ panic("suspend timeout and panic");
+}
kernel/power/suspend.c static DECLARE_SWAIT_QUEUE_HEAD(s2idle_wait_head); · line 101
Quoted 10 lines
kernel/power/suspend.c +10
-0+ +/* Should not reach here */ +pr_err ("Trigger panic failed!\n" ); +} + +static int suspend_monitor_kthread (void *arg) +{ +long err; +struct sched_param param = {.sched_priority + =MAX_RT_PRIO -1 };
Ick, no, call the scheduler functions properly, don't do this "by hand" ever.
I refer to kernel/sched/core.c and they use MAX_RT_PRIO as thread priority directly. If you have other suggestion, I am willing to try.
kernel/power/suspend.c static DECLARE_SWAIT_QUEUE_HEAD(s2idle_wait_head); · line 105
Quoted 4 lines
kernel/power/suspend.c +4
-0+static int timeout_count; +static long timeout; + +pr_info ("Init ksuspend_mon thread\n" );
Again, debugging code :(
In order to make sure ksuspend_mon thread has been initialize, I add log here. If you think that is not necessary, I can remove it.
Sure, I got it, could you help to review if we can use IS_ENABLED() like following?
+ if (IS_ENABLED(CONFIG_PM_SLEEP_MONITOR))
+ stop_suspend_mon();
+
Thank you,
Joseph.