Re: [PATCH v2] reboot: Add timeout for device shutdown

[Khasnis Soumya]

On Wed, May 29, 2024 at 02:51:48PM +0200, Greg KH wrote:
> On Wed, May 29, 2024 at 11:00:49AM +0000, Soumya Khasnis wrote:
> > The device shutdown callbacks invoked during shutdown/reboot
> > are prone to errors depending on the device state or mishandling
> > by one or more driver.
> 
> Why not fix those drivers?  A release callback should not stall, and if
> it does, that's a bug that should be fixed there.
Yes, the drivers must be fixed. But currently there is no good mechanism
which detects such stalls and there by resulting in a system hang. This
serves as a fail-safe mechanism to prevent such stalls at reboot/shutdown.

> 
> Or use a watchdog and just reboot if that triggers at shutdown time.
The hard or software watchdog enabled by config_lockup_detector couldn’t
detect the cases when stalled on IO wait (wait_for_completion/io)

> 
> > In order to prevent a device hang in such
> > scenarios, we bail out after a timeout while dumping a meaningful
> > call trace of the shutdown callback which blocks the shutdown or
> > reboot process.
> 
> Dump it where?
Dump to kernel log and there by the console. Will update the commit message

> 
> 
> > 
> > Signed-off-by: Soumya Khasnis <soumya.khasnis@xxxxxxxx>
> > Signed-off-by: Srinavasa Nagaraju <Srinavasa.Nagaraju@xxxxxxxx>
> > ---
> >  drivers/base/Kconfig | 15 +++++++++++++++
> >  kernel/reboot.c      | 46 +++++++++++++++++++++++++++++++++++++++++++-
> >  2 files changed, 60 insertions(+), 1 deletion(-)
> > 
> > diff --git a/drivers/base/Kconfig b/drivers/base/Kconfig
> > index 2b8fd6bb7da0..d06e379b6281 100644
> > --- a/drivers/base/Kconfig
> > +++ b/drivers/base/Kconfig
> > @@ -243,3 +243,18 @@ config FW_DEVLINK_SYNC_STATE_TIMEOUT
> >  	  work on.
> >  
> >  endmenu
> > +
> > +config DEVICE_SHUTDOWN_TIMEOUT
> > +	bool "device shutdown timeout"
> > +	default n
> 
> That is the default, so no need for this.
Was mindful of the “slow” devices in the system with long IO wait times.
Will have to consider increasing DEVICE_SHUTDOWN_TIMEOUT_SEC and make this
a default y

> 
> 
> > +	help
> > +	   Enable timeout for device shutdown. Helps in case device shutdown
> > +	   is hung during shoutdonw and reboot.
> > +
> > +
> > +config DEVICE_SHUTDOWN_TIMEOUT_SEC
> > +	int "device shutdown timeout in seconds"
> > +	default 5
> > +	depends on DEVICE_SHUTDOWN_TIMEOUT
> > +	help
> > +	  sets time for device shutdown timeout in seconds
> 
> You need much more help text for all of these.
Will update the commit with more help text

> 
> And why are these in the drivers/base/Kconfig file?  It has nothing to
> do with "devices", or the driver core, it's all core kernel reboot
> logic.
This is handling only device shutdown, not complete reboot part,
so we want to keep it here.

> 
> 
> > diff --git a/kernel/reboot.c b/kernel/reboot.c
> > index 22c16e2564cc..8460bd24563b 100644
> > --- a/kernel/reboot.c
> > +++ b/kernel/reboot.c
> > @@ -18,7 +18,7 @@
> >  #include <linux/syscalls.h>
> >  #include <linux/syscore_ops.h>
> >  #include <linux/uaccess.h>
> > -
> > +#include <linux/sched/debug.h>
> 
> Why remove the blank line?
Will fix it.

> 
> >  /*
> >   * this indicates whether you can reboot with ctrl-alt-del: the default is yes
> >   */
> > @@ -48,6 +48,14 @@ int reboot_cpu;
> >  enum reboot_type reboot_type = BOOT_ACPI;
> >  int reboot_force;
> >  
> > +#ifdef CONFIG_DEVICE_SHUTDOWN_TIMEOUT
> > +struct device_shutdown_timeout {
> > +	struct timer_list timer;
> > +	struct task_struct *task;
> > +} devs_shutdown;
> > +#define SHUTDOWN_TIMEOUT CONFIG_DEVICE_SHUTDOWN_TIMEOUT_SEC
> > +#endif
> 
> #ifdefs should not be in .c files, please put this in a .h file where it
> belongs.  Same for the other #ifdefs.
Will fix it

> 
> 
> 
> > +
> >  struct sys_off_handler {
> >  	struct notifier_block nb;
> >  	int (*sys_off_cb)(struct sys_off_data *data);
> > @@ -88,12 +96,46 @@ void emergency_restart(void)
> >  }
> >  EXPORT_SYMBOL_GPL(emergency_restart);
> >  
> > +#ifdef CONFIG_DEVICE_SHUTDOWN_TIMEOUT
> > +static void device_shutdown_timeout_handler(struct timer_list *t)
> > +{
> > +	pr_emerg("**** device shutdown timeout ****\n");
> 
> What does this have to do with "devices"?  This is a whole-system issue,
> or really a "broken driver" issue.
Shutdown/reboot procedure involves quite a few routines. We are specifically
dealing with a broken driver or device malfunction.

> 
> > +	show_stack(devs_shutdown.task, NULL, KERN_EMERG);
> 
> How do you know this is the 'device shutdown' stack?  What is a "device
> shutdown"?
The timeout handler is invoked in the context of reboot/shutdown process.
We are interested in the device shutdown callback which was stuck and
preventing the reboot/shutdown. By device shutdown, we refer to device
class/bus/driver shutdown calls.

> 
> > +	if (system_state == SYSTEM_RESTART)
> > +		emergency_restart();
> > +	else
> > +		machine_power_off();
> > +}
> > +
> > +static void device_shutdown_timer_set(void)
> > +{
> > +	devs_shutdown.task = current;
> 
> It's just the normal shutdown stack/process, why call it a device?
We are specifically dealing with a broken driver or device malfunction,
So to indicate same, we want to name it as device_shutdown stack.

> 
> > +	timer_setup(&devs_shutdown.timer, device_shutdown_timeout_handler, 0);
> > +	devs_shutdown.timer.expires = jiffies + SHUTDOWN_TIMEOUT * HZ;
> > +	add_timer(&devs_shutdown.timer);
> > +}
> > +
> > +static void device_shutdown_timer_clr(void)
> > +{
> > +	del_timer(&devs_shutdown.timer);
> > +}
> > +#else
> > +static inline void device_shutdown_timer_set(void)
> > +{
> > +}
> > +static inline void device_shutdown_timer_clr(void)
> > +{
> > +}
> > +#endif
> > +
> >  void kernel_restart_prepare(char *cmd)
> >  {
> >  	blocking_notifier_call_chain(&reboot_notifier_list, SYS_RESTART, cmd);
> >  	system_state = SYSTEM_RESTART;
> >  	usermodehelper_disable();
> > +	device_shutdown_timer_set();
> >  	device_shutdown();
> > +	device_shutdown_timer_clr();
> 
> Why isn't all of this done in device_shutdown() if you think it is a
> device issue?  Why put it in reboot.c?
Will consider moving to device_shutdown()..

> 
> thanks,
> 
> greg k-h

Thanks and Regards,
Soumya.