[PATCH] printk: Fix panic log flush to serial console during kdump in PREEMPT_RT kernels

[Ryo Takakura]

Hi cuiguoqi!

I'm not an expert on the subject, but hope this helps.

On Thu,  7 Aug 2025 19:22:47 +0800, cuiguoqi wrote:
>When a system running a real-time (PREEMPT_RT) kernel panics and triggers kdump,
>the critical log messages (e.g., panic reason, stack traces) may fail to appear
>on the serial console.
>
>When kdump cannot be used properly, serial console logs are crucial,
>whether for diagnosing kdump issues or troubleshooting the underlying problem.

The console not being flushed in the case of kexec should be expected
as described [0]. Its about prioritizing kexec over serial output.

>This issue arises due to synchronization or deferred flushing of the printk buffer
>in real-time contexts, where preemptible console locks or delayed workqueues prevent
>timely log output before kexec transitions to the crash kernel.
>
> /**
>  * kexec_image_info - For debugging output.
>@@ -176,6 +177,9 @@ void machine_kexec(struct kimage *kimage)
> 
> 	pr_info("Bye!\n");
> 
>+	if (IS_ENABLED(CONFIG_PREEMPT_RT) && in_kexec_crash)
>+		console_flush_on_panic(CONSOLE_FLUSH_PENDING);
>+

Calling console_flush_on_panic() while trying to kexec will
reduce its chance of success.

>diff --git a/kernel/panic.c b/kernel/panic.c
>index 72fcbb5..e0ad0df 100644
>--- a/kernel/panic.c
>+++ b/kernel/panic.c
>@@ -437,6 +437,8 @@ void vpanic(const char *fmt, va_list args)
> 	 */
> 	kgdb_panic(buf);
> 
>+	printk_legacy_allow_panic_sync();
>+
> 	/*
> 	 * If we have crashed and we have a crash kernel loaded let it handle
> 	 * everything else.
>@@ -450,8 +452,6 @@ void vpanic(const char *fmt, va_list args)
> 
> 	panic_other_cpus_shutdown(_crash_kexec_post_notifiers);
> 
>-	printk_legacy_allow_panic_sync();
>-

The ordering here should be kept where we don't want CPUs other than
the one panicked flushing legacy consoles.

Sincerely,
Ryo Takakura

[0] https://lore.kernel.org/lkml/847cagmjsx.fsf@jogness.linutronix.de/