On Sun 2022-07-10 22:10:45, John Ogness wrote:
> On 2022-07-10, Andy Shevchenko <andy.shevchenko@xxxxxxxxx> wrote:
> >> Looking at freeze-5.19.0-rc1-bad.html, at 3431.221039 we see that
> >> suspend_console() was called. The additional 1-second delay you are
> >> referring to would be 3432.436187, where serial is
> >> suspended. pr_flush() would have been satisfied when the message at
> >> 3431.221039 was printed. So the question is, why is there still
> >> printing going on?
> >
> > It might be no_console_suspend hack. Are you, btw, aware of this ugly
> > hack in the kernel?
>
> I am aware of it. There are some cases where it actually works. But it
> is not being used here. The boot args are:
>
> BOOT_IMAGE=/boot/vmlinuz-5.19.0-rc1+ root=UUID=1dfec046-baf6-4f38-8b5e-a8f438a48038 ro rw quiet console=ttyS0,115200 console=tty0 i915.enable_psr=1 initcall_debug log_buf_len=32M quiet splash console=tty0 console=ttyS0,115200n8 vt.handoff=7
>
> I am curious if Todd sees this problem with 5.19-rc4 or later (the
> kthread printers were removed).
We removed the kthreads but not pr_flush(). The commit
3b604ca81202eea2a91 ("printk: add pr_flush()") is still there.
It seems that __pr_flush() does not check whether all consoles
are suspended. In this case the progress is not possible and
it has to wait the entire timeout.
Best Regards,
Petr
