On Wed 2024-03-13 15:50:52, Sreenath Vijayan wrote:
> Add a generic function console_replay_all() for replaying
> the kernel log on consoles, in any context. It would allow
> viewing the logs on an unresponsive terminal via sysrq.
>
> Reuse the existing code from console_flush_on_panic() for
> resetting the sequence numbers, by introducing a new helper
> function __console_rewind_all(). It is safe to be called
> under console_lock().
>
> Try to acquire lock on the console subsystem without waiting.
> If successful, reset the sequence number to oldest available
> record on all consoles and call console_unlock() which will
> automatically flush the messages to the consoles.
>
> --- a/kernel/printk/printk.c
> +++ b/kernel/printk/printk.c
> @@ -4259,6 +4271,23 @@ void kmsg_dump_rewind(struct kmsg_dump_iter *iter)
> }
> EXPORT_SYMBOL_GPL(kmsg_dump_rewind);
>
> +/**
> + * console_replay_all - replay kernel log on consoles
It should rather be called "console_try_replay_all()" to make it clear
that it is just the best effort.
> + *
> + * Try to obtain lock on console subsystem and replay all
> + * available records in printk buffer on the consoles.
> + * Does nothing if lock is not obtained.
> + *
> + * Context: Any context.
This should be:
* Context: Any, except for NMI
Basically only lockless code is safe in NMI which is not the case here.
> + */
> +void console_replay_all(void)
> +{
> + if (console_trylock()) {
> + __console_rewind_all();
> + /* Consoles are flushed as part of console_unlock(). */
> + console_unlock();
> + }
> +}
> #endif
Otherwise, it looks good. With the two changes:
Reviewed-by: Petr Mladek <pmladek@xxxxxxxx>
I am sorry for the late review. I have been snowed under tasks.
Also I had healthy problems.
I have seen a mail that Greg has queued the patch in tty-next.
I am not sure if it still can be fixed. It will be perfectly fine
to change this by a followup patch.
Best Regards,
Petr
