Re: [PATCH printk] printk: flush consoles before checking progress

[Petr Mladek <pmladek@suse.com>]

On Fri 2023-09-29 13:38:33, John Ogness wrote:
> Finally, in case pr_flush() needs to poll-wait (for example,
> because there has been a handover to another CPU), reduce the
> polling interval from 100ms to 1ms. In the bug report it was
> mentioned that 100ms is unnecessarily long.
> 
> Reported-by: Todd Brandt <todd.e.brandt@intel.com>
> Closes: https://bugzilla.kernel.org/show_bug.cgi?id=217955
> Fixes: 9e70a5e109a4 ("printk: Add per-console suspended state")
> Signed-off-by: John Ogness <john.ogness@linutronix.de>
> ---
>  kernel/printk/printk.c | 23 ++++++++++++-----------
>  1 file changed, 12 insertions(+), 11 deletions(-)
> 
> diff --git a/kernel/printk/printk.c b/kernel/printk/printk.c
> index 7e0b4dd02398..cf7841fc6eef 100644
> --- a/kernel/printk/printk.c
> +++ b/kernel/printk/printk.c
> @@ -3775,16 +3781,11 @@ static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progre
>  		if (diff == 0 || remaining == 0)
>  			break;
>  
> -		if (remaining < 0) {
> -			/* no timeout limit */
> -			msleep(100);
> -		} else if (remaining < 100) {
> -			msleep(remaining);
> -			remaining = 0;
> -		} else {
> -			msleep(100);
> -			remaining -= 100;
> -		}
> +		msleep(1);

I was about to push this patch and ran checkpatch.pl. It warned about

WARNING: msleep < 20ms can sleep for up to 20ms; see Documentation/timers/timers-howto.rst
#73: FILE: kernel/printk/printk.c:3782:
+               msleep(1);

And indeed, Documentation/timers/timers-howto.rst says that msleep()
might sleep longer that expected for <20ms delays. I guess that
it is somehow related to jiffies, HZ, and load on the system.

I think that we need to count jiffies here. Something like:

diff --git a/kernel/printk/printk.c b/kernel/printk/printk.c
index db81b68d7f14..6ea500d95fd9 100644
--- a/kernel/printk/printk.c
+++ b/kernel/printk/printk.c
@@ -3723,7 +3723,8 @@ late_initcall(printk_late_init);
 /* If @con is specified, only wait for that console. Otherwise wait for all. */
 static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progress)
 {
-	int remaining = timeout_ms;
+	unsigned long timeout_jiffies = msecs_to_jiffies(timeout_ms);
+	unsigned_long timeout_end = jiffies + timeout_jiffies;
 	struct console *c;
 	u64 last_diff = 0;
 	u64 printk_seq;
@@ -3772,24 +3773,19 @@ static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progre
 		console_srcu_read_unlock(cookie);
 
 		if (diff != last_diff && reset_on_progress)
-			remaining = timeout_ms;
+			timeout_end = jiffies + timeout_jiffies;
 
 		console_unlock();
 
 		/* Note: @diff is 0 if there are no usable consoles. */
-		if (diff == 0 || remaining == 0)
+		if (diff == 0)
 			break;
 
-		if (remaining < 0) {
-			/* no timeout limit */
-			msleep(100);
-		} else if (remaining < 100) {
-			msleep(remaining);
-			remaining = 0;
-		} else {
-			msleep(100);
-			remaining -= 100;
-		}
+		/* Negative timeout means an infinite wait. */
+		if (timeout_ms >= 0 && time_after_eq(jiffies, timeout_end))
+			break;
+
+		msleep(2000 / HZ);
 
 		last_diff = diff;
 	}

Note that I used 2000 / HZ to make sure that it will be >= 1.

And we should do this in a separate patch. It seems that sleeping
is a bigger magic than I expected.

Best Regards,
Petr