Re: [PATCH] serial: 8250: add option to disable registration of legacy ISA ports

Greg Kroah-Hartman [off-list ref] writes:

On Sun, Jan 31, 2021 at 01:18:47PM +0000, Måns Rullgård wrote:

quoted

Greg Kroah-Hartman [off-list ref] writes:

quoted

On Thu, Jan 28, 2021 at 05:22:44PM +0000, Mans Rullgard wrote:

quoted

On systems that do not have the traditional PC ISA serial ports, the
8250 driver still creates non-functional device nodes.  This change
makes only ports that actually exist (PCI, DT, ...) get device nodes.

Signed-off-by: Mans Rullgard <redacted>
---
 drivers/tty/serial/8250/8250_core.c | 26 ++++++++++++++++++++------
 drivers/tty/serial/8250/Kconfig     |  5 +++++
 2 files changed, 25 insertions(+), 6 deletions(-)

diff --git a/drivers/tty/serial/8250/8250_core.c b/drivers/tty/serial/8250/8250_core.c
index cae61d1ebec5..49695dd3677c 100644
--- a/drivers/tty/serial/8250/8250_core.c
+++ b/drivers/tty/serial/8250/8250_core.c

@@ -555,6 +555,7 @@ static void __init serial8250_isa_init_ports(void)
 	}
 }
 
+#ifdef CONFIG_SERIAL_8250_ISA

This is just making a mess of the code. 

It was already a mess.

True, but don't make it a worse one please.

quoted

quoted

To do this right, pull the isa code out into a separate file and put
the #ifdef in a .h file, so we can properly maintain and support this
code over time.  This change as-is is not going to make that any
easier :(

I might put in that effort if there's a reasonable chance this change
will be accepted.  If it's going to be rejected regardless, I'd rather
not waste my time.

quoted

quoted

+config SERIAL_8250_ISA
+	bool "8250/16550 ISA device support" if EXPERT

So, no one will set this?

I followed the pattern of the existing SERIAL_8250_PNP option.  Was that
a mistake?  How would you prefer it?

I don't know, I'm just asking.

quoted

quoted

What userspace visable change will be caused by this? 

There won't be /dev/ttyS devices for ports that don't exist.

quoted

Will ports get renumbered?

Not if they had predictable numbers to begin with.

So that would be "yes"?  If so, obviously we couldn't take this, right?

On a Beaglebone Black based system with some of the UARTs enabled, those
keep their numbers such that e.g. ttyS0, ttyS1, and ttyS4 show up in
/dev while ttyS2 and ttyS3 do not since they don't correspond to any
(enabled) ports.

If any of the very many variants of this driver do not assign fixed
numbers, those would possibly be renumbered.  Should that be the case,
the numbering was never guaranteed to begin with, so I fail to see any
problem.

-- 
Måns Rullgård