Re: [patch] x86, serial: always probe for legacy COM ports

[Jeff Garzik <jeff@garzik.org>]

[-- Attachment #1: Type: text/plain, Size: 991 bytes --]

Bjorn Helgaas wrote:
> Andrew, if you drop
>     revert-x86-serial-convert-legacy-com-ports-to-platform-devices.patch
> and apply this, we should have the previous behavior.  I think the
> conversion to platform devices is still worthwhile.  Obviously, I intend
> this as 2.6.23 material.
> 
> 
> Always probe for serial ports at legacy addresses, even if we have PNPBIOS
> or ACPI that should tell us about those ports.
> 
> We have no reports yet of defective firmware that completely omits ports.
> However, Sébastien Dugué <sebastien.dugue@bull.net> reported that the
> NEC Express5800/120Lh with Phoenix BIOS 6.0.5N52, released 7/12/2005,
> reports COM2 first, then COM1 in the ACPI namespace.
> 
> 8250_pnp currently registers devices in namespace order.  On this machine,
> that makes ttyS0=COM2 and ttyS1=COM1, the reverse of what we want.
> 
> Signed-off-by: Bjorn Helgaas <bjorn.helgaas@hp.com>

This is still incomplete, as repeatedly stated.  Here is the email, again.

	Jeff




[-- Attachment #2: Attached Message --]
[-- Type: message/rfc822, Size: 2056 bytes --]

From: Jeff Garzik <jeff@garzik.org>
To: Russell King <rmk@arm.linux.org.uk>
Cc: akpm@linux-foundation.org,  mm-commits@vger.kernel.org,  ambx1@neo.rr.com,  bjorn.helgaas@hp.com,  lenb@kernel.org,  mjg59@srcf.ucam.org,  sebastien.dugue@bull.net
Subject: Re: + revert-x86-serial-convert-legacy-com-ports-to-platform-devices.patch added to -mm tree
Date: Tue, 24 Jul 2007 19:21:27 -0400
Message-ID: <46A68977.50800@garzik.org>

Russell King wrote:
> So we shouldn't even try to remove the legacy "legacy port" probing in
> the serial driver like other platforms have done, which this patch minus
> those bits I mentioned actually achieves?

Move [to a new driver / location in tree]?  Sure.  As long as there are 
no probe order issues, module naming issues (I note distinct lack of 
MODULE_ALIAS for example) or other transitions issues.

Remove?  Definitely not.

Even if we apply your suggested solution, updates are still needed:
* fix kernel-parameters.txt

* address transition issues related to module naming.  other transition 
issues?  has anybody even given this any thought?

* no apparent module unload support (regression or just a bug?)

* add docs for x86/x86-64 users (or point to docs in the source). 
Overall this was a highly "stealthy" change over to the new platform 
driver, without much notice at all.  Not optimum, for a popular device 
on Linux's most popular platform.

I have no complaints about platform_device direction, certainly.  But 
this particular commit was incomplete, poorly described, and broke 
decade+ behavior.

	Jeff