On Fri 2025-09-26 14:49:09, Petr Mladek wrote:
> This patchset should solve problem which was being discussed
> at https://lore.kernel.org/all/aNFR45fL2L4PavNc@pathway.suse.cz
>
> __nbcon_atomic_flush_pending_con() preserves the nbcon console
> ownership all the time when flushing pending messages. It might
> take a long time with slow serial consoles.
>
> It might trigger a hardlockup report on another CPU which is
> busy waiting for the nbcon console ownership, for example,
> in nbcon_reacquire_nobuf() or __uart_port_nbcon_acquire().
>
> The problem is solved by the 3rd patch. It releases the console
> context ownership after each record.
>
> The 3rd patch alone would increase the risk of takeovers and repeated
> lines. It is prevented by the 1st patch which blocks the printk kthread
> when any CPU is in an emergency context.
>
> The 2nd patch allows to block the printk kthread also in panic.
> It is not important. It is just an obvious update of the check
> for emergency contexts.
>
> Note: The patchset applies against current Linus' tree (v6.17-rc7).
>
> The 2nd patch would need an update after the consolisation of
> the panic state API gets merged via -mm tree,
> see https://lore.kernel.org/r/20250825022947.1596226-2-wangjinchao600@gmail.com
>
> Petr Mladek (3):
> printk/nbcon: Block printk kthreads when any CPU is in an emergency
> context
> printk/nbcon/panic: Allow printk kthread to sleep when the system is
> in panic
> printk/nbcon: Release nbcon consoles ownership in atomic flush after
> each emitted record
>
> kernel/printk/internal.h | 1 +
> kernel/printk/nbcon.c | 43 +++++++++++++++++++++++++++++++++++-----
> kernel/printk/printk.c | 2 +-
> 3 files changed, 40 insertions(+), 6 deletions(-)
JFYI, the patchset has been comitted into printk/linux.git,
branch rework/atomic-flush-hardlockup[1].
It is queued for 6.19.
Note that I did the following modifications:
+ Added changes into the 1st patch proposed by John[2], namely:
+ initialize nbcon_cpu_emergency_cnt and make it static.
+ call nbcon_kthreads_wake() only when printk_get_console_flush_type()
sets ft.nbcon_offload.
+ Rebased 2nd patch on top of 6.18-rc1 (panic_in_progress() moved to
linux/panic.h).
[1] https://git.kernel.org/pub/scm/linux/kernel/git/printk/linux.git/log/?h=rework/atomic-flush-hardlockup
[2] https://lore.kernel.org/all/841pnti8k2.fsf@jogness.linutronix.de/
Best Regards,
Petr
PS: I thought about sending v2. But v1 already got enough Acks and
I added the requested changes by cut&paste.