On Fri 2025-11-28 03:43:42, Chris Down wrote:
> As part of the per-console loglevel implementation, printk_delay()
> introduces a call to suppress_message_printing_everywhere() within
> printk_delay(). This check acquires the console SRCU lock and iterates
> through all registered consoles to determine if the message implies a
> delay based on per-console loglevels.
>
> Since printk_delay() is called in the hotpath of vprintk_emit(), this
> introduces measurable overhead (locking and pointer chasing) for every
> printk message, even on systems where no delay is configured.
>
> Optimise this by short-circuiting the function. If neither a boot delay
> nor a runtime printk_delay_msec is configured, return immediately. This
> avoids the expensive console list iteration in the common case where
> delays are disabled.
>
> Signed-off-by: Chris Down <chris@chrisdown.name>
> ---
> kernel/printk/printk.c | 25 ++++++++++++++++++-------
> 1 file changed, 18 insertions(+), 7 deletions(-)
>
> diff --git a/kernel/printk/printk.c b/kernel/printk/printk.c
> index 0d7f3dac8177..13e17d892ec9 100644
> --- a/kernel/printk/printk.c
> +++ b/kernel/printk/printk.c
> @@ -2233,20 +2233,31 @@ int printk_delay_msec __read_mostly;
>
> static inline void printk_delay(int level)
> {
> + bool boot_delay_active = false;
> + int m;
> +
> +#ifdef CONFIG_BOOT_PRINTK_DELAY
> + boot_delay_active = boot_delay && system_state < SYSTEM_RUNNING;
> +#endif
The same code is used also in boot_delay_msec(). It can be removed
there. But still, it would be better to create a helper function
and avoid the #ifdef in the middle of the function.
> + if (!boot_delay_active && !READ_ONCE(printk_delay_msec))
There is another READ_ONCE(printk_delay_msec) below. It would
be better to read the value at the beginning of the function
and use it everywhere.
> + return;
> +
> /* If the message is forced (e.g. panic), we must delay */
> if (!is_printk_force_console() &&
> suppress_message_printing_everywhere(level))
> return;
>
> - boot_delay_msec();
> + if (boot_delay_active)
> + boot_delay_msec();
>
> - if (unlikely(printk_delay_msec)) {
> - int m = printk_delay_msec;
> + m = READ_ONCE(printk_delay_msec);
> + if (!m)
> + return;
>
> - while (m--) {
> - mdelay(1);
> - touch_nmi_watchdog();
> - }
> + while (m--) {
> + mdelay(1);
> + touch_nmi_watchdog();
> }
> }
I played with the code. I am going to send my variant as a reply to
the 1st patch where I proposed 2 alternative solutions, see
https://lore.kernel.org/all/aThQ5mZNIijc1G94@pathway
Best Regards,
Petr