Re: Support for serial console on legacy free machines

From: Eric W. Biederman (
Date: Tue Jul 31 2001 - 11:00:12 EST

"Stuart MacDonald" <> writes:

> From: "Khalid Aziz" <>
> > AFAIK, you can not have console on a PCI serial port at this time. I
> > looked at it few months back and found out that PCI initialization
> > happens much too late for a serial console. It would take quite a bit of
> That's very odd. That implies that serial consoles don't use the serial
> driver at all then, as the pci serial port setup is done at the same
> time as the regular serial port setups.
> If a serial console is using serial.c, the pci serial ports will also
> be available.
> Hm, looking through the driver quick, I find some interesting things:
> A) Serial console support is mutually exclusive with the serial driver
> being a module.
> B) The serial console will not share its irq. Other ports with the same
> irq are set to polled mode. This may impact performance. It also suggests
> that using the console on a pci board isn't a good idea, as pci will
> share the irq to other devices.
> C) serial.c contains a completely separate serial console driver,
> complete with its own init routine. Which meshes with the current
> suggestion that the "serial driver" isn't used, and pci init happens
> too late.
> > work to get serial console working on PCI cards. PA-Linux faced the same
> > problem but they were able to get around it by using the firmware calls
> > to do console I/O. If serial console were working on PCI serial cards,
> > you wouldn't need ACPI to use it.
> It seems like pci consoles won't work, now that I think about it.

It depends. The pci iniitalization is straight forward. It should
be relatively simple to build a pci serial driver that hard codes the
pci card. I have done similiar things because for some debugging I
have done the current serial console is initialized too late. So I
hacked a hardcoded driver into printk.c

> The
> console driver gets an index, which I'm going to assume works thusly:
> lilo console=ttyS1 ends up passing 1 as the index. That index is used
> to pick a serial port out of the array of serial ports that the driver
> knows about. If console init happens early, and serial driver init happens
> late (it would be dependent on pci init) then only hard coded ports
> would work. Those are defined in asm/serial.h, and for i386 include the
> standard ports, and a number of isa ports from various board manufacturers.
> Using one of our pci ports would require knowledge of its io address,
> which wouldn't be available until the pci subsystem had inited. Perhaps
> that could be changed to allow pci based consoles?

I would say the simple solution would be to add a configure option to
do an early PCI init on the card, and to hard code it's ports.
> Elsethread someone mentions a stripped down usb console driver; that's
> analogous to the serial console driver contained in serial.c. And if
> serial can do it...

If it isn't too complex.

To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to
More majordomo info at
Please read the FAQ at

This archive was generated by hypermail 2b29 : Tue Jul 31 2001 - 21:00:51 EST