With commit ("printk: Avoid scheduling irq_work on suspend") the
implementation of printk_get_console_flush_type() was modified to
avoid offloading when irq_work should be blocked during suspend.
Since printk uses the returned flush type to determine what
flushing methods are used, this was thought to be sufficient for
avoiding irq_work usage during the suspend phase.

However, vprintk_emit() implements a hack to support
printk_deferred(). In this hack, the returned flush type is
adjusted to make sure no legacy direct printing occurs when
printk_deferred() was used.

Because of this hack, the legacy offloading flushing method can
still be used, causing irq_work to be queued when it should not
be.

Adjust the vprintk_emit() hack to also consider
@console_irqwork_blocked so that legacy offloading will not be
chosen when irq_work should be blocked.

Link: https://lore.kernel.org/lkml/87fra90xv4.fsf@jogness.linutronix.de
Signed-off-by: John Ogness <john.ogness@linutronix.de>
---
 kernel/printk/printk.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/kernel/printk/printk.c b/kernel/printk/printk.c
index b1c0d35cf3caa..c27fc7fc64eb5 100644
--- a/kernel/printk/printk.c
+++ b/kernel/printk/printk.c
@@ -2393,7 +2393,7 @@ asmlinkage int vprintk_emit(int facility, int level,
 	/* If called from the scheduler, we can not call up(). */
 	if (level == LOGLEVEL_SCHED) {
 		level = LOGLEVEL_DEFAULT;
-		ft.legacy_offload |= ft.legacy_direct;
+		ft.legacy_offload |= ft.legacy_direct && !console_irqwork_blocked;
 		ft.legacy_direct = false;
 	}
 

base-commit: 26873e3e7f0cb26c45e6ad63656f9fe36b2aa31b
-- 
2.47.3

On Fri 2025-11-21 11:26:00, John Ogness wrote:
> With commit ("printk: Avoid scheduling irq_work on suspend") the
> implementation of printk_get_console_flush_type() was modified to
> avoid offloading when irq_work should be blocked during suspend.
> Since printk uses the returned flush type to determine what
> flushing methods are used, this was thought to be sufficient for
> avoiding irq_work usage during the suspend phase.
> 
> However, vprintk_emit() implements a hack to support
> printk_deferred(). In this hack, the returned flush type is
> adjusted to make sure no legacy direct printing occurs when
> printk_deferred() was used.
> 
> Because of this hack, the legacy offloading flushing method can
> still be used, causing irq_work to be queued when it should not
> be.
> 
> Adjust the vprintk_emit() hack to also consider
> @console_irqwork_blocked so that legacy offloading will not be
> chosen when irq_work should be blocked.
> 
> Link: https://lore.kernel.org/lkml/87fra90xv4.fsf@jogness.linutronix.de
> Signed-off-by: John Ogness <john.ogness@linutronix.de>

JFYI, the patch has been committed into printk/linux.git,
branch rework/suspend-fixes.

Best Regards,
Petr

On Fri 2025-11-21 11:26:00, John Ogness wrote:
> With commit ("printk: Avoid scheduling irq_work on suspend") the
> implementation of printk_get_console_flush_type() was modified to
> avoid offloading when irq_work should be blocked during suspend.
> Since printk uses the returned flush type to determine what
> flushing methods are used, this was thought to be sufficient for
> avoiding irq_work usage during the suspend phase.
> 
> However, vprintk_emit() implements a hack to support
> printk_deferred(). In this hack, the returned flush type is
> adjusted to make sure no legacy direct printing occurs when
> printk_deferred() was used.
> 
> Because of this hack, the legacy offloading flushing method can
> still be used, causing irq_work to be queued when it should not
> be.
> 
> Adjust the vprintk_emit() hack to also consider
> @console_irqwork_blocked so that legacy offloading will not be
> chosen when irq_work should be blocked.
> 
> Link: https://lore.kernel.org/lkml/87fra90xv4.fsf@jogness.linutronix.de
> Signed-off-by: John Ogness <john.ogness@linutronix.de>

Thanks for catching and fixing the problem:

Reviewed-by: Petr Mladek <pmladek@suse.com>

Best Regards,
Petr