Re: [PATCH next v2 3/4] serial: 8250: Switch to nbcon console

[Andy Shevchenko <andriy.shevchenko@linux.intel.com>]

On Wed, Sep 18, 2024 at 04:35:06PM +0200, Petr Mladek wrote:
> On Wed 2024-09-18 17:01:27, Andy Shevchenko wrote:
> > On Wed, Sep 18, 2024 at 02:26:25PM +0200, Petr Mladek wrote:
> > > On Fri 2024-09-13 16:11:37, John Ogness wrote:
> > > > Implement the necessary callbacks to switch the 8250 console driver
> > > > to perform as an nbcon console.
> > > > 
> > > > Add implementations for the nbcon console callbacks (write_atomic,
> > > > write_thread, device_lock, device_unlock) and add CON_NBCON to the
> > > > initial flags.
> > > > 
> > > > All register access in the callbacks are within unsafe sections.
> > > > The write_thread() callback allows safe handover/takeover per byte.
> > > > The write_atomic() callback allows safe handover/takeover per
> > > > printk record and adds a preceding newline if it took over mid-line.
> > > > 
> > > > For the write_atomic() case, a new irq_work is used to defer modem
> > > > control since it may be a context that does not allow waking up
> > > > tasks.
> > > 
> > > It would be fair to mention that it does not longer support fifo in
> > > the 8250 driver. It basically reverted the commit 8f3631f0f6eb42e5
> > > ("serial/8250: Use fifo in 8250 console driver").
> > > 
> > > It is not usable in write_thread() because it would not allow
> > > a safe takeover between emitting particular characters.
> > > 
> > > It might still be used in write_atomic() but it is probably not
> > > worth it. This callback is used "only" in emergency and panic
> > > situations.
> > 
> > This is unfortunate. It will drop down the efficiency of printing.
> 
> The FIFO mode has been added by the commit 8f3631f0f6eb42e5
> ("serial/8250: Use fifo in 8250 console driver"). The interesting
> parts are:
> 
> <paste>
>   While investigating a bug in the RHEL kernel, I noticed that the serial
>   console throughput is way below the configured speed of 115200 bps in
>   a HP Proliant DL380 Gen9. I was expecting something above 10KB/s, but
>   I got 2.5KB/s.
> 
>   In another machine, I measured a throughput of 11.5KB/s, with the serial
>   controller taking between 80-90us to send each byte. That matches the
>   expected throughput for a configuration of 115200 bps.
> 
>   This patch changes the serial8250_console_write to use the 16550 fifo
>   if available. In my benchmarks I got around 25% improvement in the slow
>   machine, and no performance penalty in the fast machine.
> </paste>
> 
> I would translate it:
> 
> The FIFO mode helped with some buggy serial console. But it helped to gain
> only small portion of the expected speed. The commit message does not
> mention any gain with the normally working system.
> 
> It has been added in 2022. It was considered only because of a
> "broken" system. Nobody cared enough before.
> 
> > I think it should be done differently, i.e. the takeover the code
> > has to drop FIFO (IIRC it's easy to achieve by disabling it or so)
> > and switch to printing the panic/emergency message. But still at
> > some baud rate speeds draining the FIFO to the other end may be
> > not a bad idea as it takes a few dozens of microseconds.
> 
> Sure. it is doable. But I am not convinced that it is really worth it.

Fair enough. But perhaps Cc to the author to at least notify them about
this change?

-- 
With Best Regards,
Andy Shevchenko