Re: [PATCH v4] fs/pstore: provide panic data even in suspend

[Petr Mladek <pmladek@suse.com>]

On Mon 2015-08-17 16:51:39, check.kernel@gmail.com wrote:
> From: Dongdong Yang <yangdongdong@xiaomi.com>
> 
> If system restart after panic, this patch also enables
> panic and oops messages which in suspend context to be
> logged into ramoops console buffer where it can be read
> back at some later point.
> 
> Signed-off-by: Dongdong Yang <yangdongdong@xiaomi.com>
> Signed-off-by: Linghua Gu <gulinghua@xiaomi.com>
> Signed-off-by: Hui Du <duhui@xiaomi.com>
> ---
>  fs/pstore/ram.c        | 16 ++++++++++++++++
>  include/linux/printk.h |  6 ++++++
>  kernel/printk/printk.c |  6 ++++++
>  3 files changed, 28 insertions(+)
> 
> diff --git a/fs/pstore/ram.c b/fs/pstore/ram.c
> index 6c26c4d..3d981a1 100644
> --- a/fs/pstore/ram.c
> +++ b/fs/pstore/ram.c
> @@ -642,8 +642,23 @@ static void ramoops_register_dummy(void)
>  	}
>  }
>  
> +static int ramoops_console_notify(struct notifier_block *this,
> +		unsigned long event, void *ptr)
> +{
> +	pr_emerg("ramoops unlock console ...\n");
> +	emergency_unlock_console();
> +
> +	return 0;
> +}
> +
> +static struct notifier_block ramoop_nb = {
> +	.notifier_call = ramoops_console_notify,
> +	.priority = INT_MAX,
> +};
> +
>  static int __init ramoops_init(void)
>  {
> +	atomic_notifier_chain_register(&panic_notifier_list, &ramoop_nb);
>  	ramoops_register_dummy();
>  	return platform_driver_register(&ramoops_driver);
>  }
> @@ -654,6 +669,7 @@ static void __exit ramoops_exit(void)
>  	platform_driver_unregister(&ramoops_driver);
>  	platform_device_unregister(dummy);
>  	kfree(dummy_data);
> +	atomic_notifier_chain_unregister(&panic_notifier_list, &ramoop_nb);
>  }
>  module_exit(ramoops_exit);
>  
> diff --git a/include/linux/printk.h b/include/linux/printk.h
> index a6298b2..bd043ed 100644
> --- a/include/linux/printk.h
> +++ b/include/linux/printk.h
> @@ -230,6 +230,12 @@ static inline void show_regs_print_info(const char *log_lvl)
>  }
>  #endif
>  
> +/*
> + * Enables printk() before emergency_restart
> + * if we go into suspend states and run into oops.
> + */
> +void emergency_unlock_console(void);
> +
>  extern asmlinkage void dump_stack(void) __cold;
>  
>  #ifndef pr_fmt
> diff --git a/kernel/printk/printk.c b/kernel/printk/printk.c
> index cf8c242..abb8af4 100644
> --- a/kernel/printk/printk.c
> +++ b/kernel/printk/printk.c
> @@ -2107,6 +2107,12 @@ void resume_console(void)
>  	console_unlock();
>  }
>  
> +void emergency_unlock_console(void)
> +{
> +	if (oops_in_progress && panic_timeout != 0 && console_suspended)
> +		resume_console();

IMHO, you need to put "console_sem" down if you want to test
"console_suspended". And you must keep it down until you resume
the console to avoid a race => we could not call resume_console()
in the current form.

Also I am a bit nervous that this operation is not symmetric
and the console will stay resumed. Please look at bust_spinlocks(),
how this function is used in panic(), and see how the
"oops_in_progress" is handled.

To be honest, I do not understand this code in details enough
and I do not have time to investigate it at the moment.
I just use a common sense to produce this feedback. Others
might give you a better feedback. It is possible that this
should get handled a more generic way and not by a ramoops
specific handler.

BTW: Have you tested this with more consoles attached? I wonder
if other consoles are still usable in this situation and if
they will not block the output to ramoops.

Best Regards,
Petr


> +}
> +
>  /**
>   * console_cpu_notify - print deferred console messages after CPU hotplug
>   * @self: notifier struct
> -- 
> 2.5.0
> 
> 
> v4: 
> - Move declaration emergency_unlock_console to printk.h from pstore_ram.h
> - Update sign-off information.
> 
> Thanks,
> Dongdong Yang