Re: [RFC PATCH v1 08/25] printk: add ring buffer and kthread

[Sergey Senozhatsky]

On (03/06/19 23:22), John Ogness wrote:
> On 2019-03-06, Petr Mladek <pmladek@xxxxxxxx> wrote:
> >> _Both_ categories are important for the user, but their requirements
> >> are different:
> >> 
> >>    informational: non-disturbing
> >>    emergency:     reliable
> >
> > Isn't this already handled by the console_level?
> >
> > The informational messages can be reliably read via syslog, /dev/kmsg.
> > They are related to the normal works when the system works well.
> >
> > The emergency messages (errors, warnings) are printed in emergency
> > situations. They are printed as reliably as possible to the console
> > because the userspace might not be reliable enough.
> 
> I've never viewed console_level this way. _If_ console_level really is
> supposed to define the emergency/informational boundary, all
> informational messages are supposed to be handled by userspace, and
> console printing's main objective is reliability... then I would change
> my proposal such that:

OK, you guys are ahead of me.

FB folks want to have a per-console sysfs knob to dynamically adjust
loglevel on each console. The use case is to temporarily set loglevel
to, say, debug on fast consoles, gather some data/logs, set loglevel
back to less verbose afterwards.

Preserving the existing loglevel behaviour looks right to me.

> - if a console supports write_atomic(), _all_ console printing for that
>   console would use write_atomic()

Sounds right.
But Big-Konsole-Lock looks suspicious.

> - only consoles without write_atomic() will be printing via the
>   printk-kthread(s)
> 
> IMO, for consoles with write_atomic(), this would increase reliability
> over the current mainline implementation. It would also simplify
> write_atomic() implementations because they would no longer need to
> synchronize against write().

[..]
> For those consoles that cannot implement write_atomic() (vt and
> netconsole come to mind), or as a transition period until remaining
> console drivers have implemented write_atomic(), these would use the
> "fallback" of printing fully preemptively in their own kthread using
> write().

This sounds concerning. IMHO, netconsole is too important to rely
on preemptible printk and scheduler. Especially those netcons which
run in "report only when oops_in_progress" mode. Sometimes netconsole
is the fastest console available, but preemptible printk may turn it
into the slowest one.

	-ss