Re: [PATCH v3] printk: add halt_delay parameter for printk delay in halt phase

[Ingo Molnar <mingo@elte.hu>]

* Dave Young <hidave.darkstar@gmail.com> wrote:

> On Mon, Jun 8, 2009 at 4:28 PM, Andrew Morton<akpm@linux-foundation.org> wrote:
> > On Mon, 8 Jun 2009 10:14:39 +0200 Ingo Molnar <mingo@elte.hu> wrote:
> >
> >>
> >> * Dave Young <hidave.darkstar@gmail.com> wrote:
> >>
> >> > Add a halt_delay module parameter in printk.c used to read the
> >> > printk messages in halt/poweroff/restart phase, delay each printk
> >> > messages by halt_delay milliseconds. It is useful for debugging if
> >> > there's no other way to dump kernel messages that time.
> >> >
> >> > The halt_delay max value is 65535, default value is 0, change it
> >> > by:
> >> >
> >> > echo xxx > /sys/module/printk/parameters/halt_delay
> >> >
> >> > Signed-off-by: Dave Young <hidave.darkstar@gmail.com>
> >> > ---
> >> > Documentation/kernel-parameters.txt |    5 +++++
> >> > kernel/printk.c                     |   17 +++++++++++++++++
> >> > 2 files changed, 22 insertions(+)
> >> >
> >> > --- linux-2.6.orig/kernel/printk.c  2009-06-08 13:55:35.000000000 +0800
> >> > +++ linux-2.6/kernel/printk.c       2009-06-08 13:56:23.000000000 +0800
> >> > @@ -250,6 +250,22 @@ static inline void boot_delay_msec(void)
> >> >  }
> >> >  #endif
> >> >
> >> > +/* msecs delay after each halt/poweroff/restart phase printk,
> >> > + unsigned short is enough for delay in milliseconds */
> >> > +static unsigned short halt_delay;
> >> > +
> >> > +static inline void halt_delay_msec(void)
> >> > +{
> >> > +   if (unlikely(halt_delay == 0 || !(system_state == SYSTEM_HALT
> >> > +                           || system_state == SYSTEM_POWER_OFF
> >> > +                           || system_state == SYSTEM_RESTART)))
> >> > +           return;
> >>
> >> This is a tiny bit ugly (and goes into the vprintf path) but i can
> >> see no other way either - a system_state > SYSTEM_RUNNING check
> >> would needlessly include the suspend-to-disk state (which we dont
> >> want to include here).
> >>
> >> In theory we could turn system_state into a bitmask and have a
> >> print_delay_mask check instead of these flags ... but that is a far
> >> wider change and i'm not sure it's a net step forwards.
> >>
> >> I've applied your patch to tip:core/printk with small edits to the
> >> changelog - Linus & Andrew is Cc:ed, should they have any
> >> objections.
> >
> > Could we not put just a single delay in there, immediately prior to halting,
> > restarting or poweroffing?
> 
> But, then prink messages will still flush too fast for us to see 
> the detail.

Plus often there's a loop in architecture code that tries various 
methods of reboot. We dont know which one works - and any of them 
could produce warning messages. (and this happened a number of times 
in the past)

So this would mean having to find up to a hundred of 'reboot now' 
places in a two dozen architectures (and keeping them all maintained 
ongoing as well). Does not seem like a good choice to me.

	Ingo