[PATCH] SERIAL: discover PNP ports before PCI, etc

[PATCH] SERIAL: discover PNP ports before PCI, etc

[Bjorn Helgaas]

PNP ports tend to be built-in, and discovering them after
PCI ports means the names of the built-in ports can change
if you add or remove PCI ports.

(And yes, we should look at getting rid of 8250_acpi.c
now that we have PNPACPI, but that's for another patch.)

Signed-off-by: Bjorn Helgaas <bjorn.helgaas@hp.com>

===== drivers/serial/Makefile 1.42 vs edited =====
--- 1.42/drivers/serial/Makefile	2005-02-03 07:42:40 -07:00
+++ edited/drivers/serial/Makefile	2005-02-09 12:58:23 -07:00
@@ -6,9 +6,9 @@
 
 serial-8250-y :=
 serial-8250-$(CONFIG_SERIAL_8250_ACPI) += 8250_acpi.o
+serial-8250-$(CONFIG_PNP) += 8250_pnp.o
 serial-8250-$(CONFIG_GSC) += 8250_gsc.o
 serial-8250-$(CONFIG_PCI) += 8250_pci.o
-serial-8250-$(CONFIG_PNP) += 8250_pnp.o
 serial-8250-$(CONFIG_HP300) += 8250_hp300.o
 
 obj-$(CONFIG_SERIAL_CORE) += serial_core.o

Re: [PATCH] SERIAL: discover PNP ports before PCI, etc

[Adam Belay]

On Wed, Feb 09, 2005 at 01:12:47PM -0700, Bjorn Helgaas wrote:
> PNP ports tend to be built-in, and discovering them after
> PCI ports means the names of the built-in ports can change
> if you add or remove PCI ports.
> 
> (And yes, we should look at getting rid of 8250_acpi.c
> now that we have PNPACPI, but that's for another patch.)
> 
> Signed-off-by: Bjorn Helgaas <bjorn.helgaas@hp.com>
>

I think this used to be the old behavior.  It was changed because some
of the resources reserved by pnp would also be reserved by pci, causing
pci to fail.  We should see if this is still the case.

Thanks,
Adam

Re: [PATCH] SERIAL: discover PNP ports before PCI, etc

[Bjorn Helgaas]

On Wed, 2005-02-09 at 21:16 -0500, Adam Belay wrote:
> On Wed, Feb 09, 2005 at 01:12:47PM -0700, Bjorn Helgaas wrote:
> > PNP ports tend to be built-in, and discovering them after
> > PCI ports means the names of the built-in ports can change
> > if you add or remove PCI ports.
> > 
> > (And yes, we should look at getting rid of 8250_acpi.c
> > now that we have PNPACPI, but that's for another patch.)
> > 
> > Signed-off-by: Bjorn Helgaas <bjorn.helgaas@hp.com>>
> 
> I think this used to be the old behavior.  It was changed because some
> of the resources reserved by pnp would also be reserved by pci, causing
> pci to fail.  We should see if this is still the case.

That must be before the 2.5 series ...  Everything in the BK
history (which admittedly only goes back to 7/21/2002) has
PCI before PNP.

Wouldn't it be a bug if PNP described a resource that can be
discovered by standard PCI discovery?

I think we should move PNP before PCI and fix whatever resource
problems crop up.

Re: [PATCH] SERIAL: discover PNP ports before PCI, etc

[Bjorn Helgaas]

On Thu, 2005-02-10 at 12:37 -0500, Adam Belay wrote:
> > Wouldn't it be a bug if PNP described a resource that can be
> > discovered by standard PCI discovery?
> 
> Maybe.  It's usually one of these:
> 
> pnp: 00:09: ioport range 0x4d0-0x4d1 has been reserved
> pnp: 00:09: ioport range 0xcf8-0xcff could not be reserved
> pnp: 00:0b: ioport range 0x800-0x87f has been reserved
> 
> The "could not be reserved" was touched by pci.  From memory, I think
> it's the pci configuration range.

Right.  0xcf8-0xcff is reserved in pci_direct_init(), which is
part of the arch PCI core, not part of a driver.  So I don't
think the 8250 PCI/PNP order should make a difference for this
particular range (the arch PCI init happens long before either
8250 PCI or PNP claim).

> > I think we should move PNP before PCI and fix whatever resource
> > problems crop up.
> 
> We could always add this change to -mm and see what effect it has.  Let me
> know if you still think this makes sense.

Yeah, I think that's Russell's plan, which I think is the right one.

Re: [PATCH] SERIAL: discover PNP ports before PCI, etc

[Russell King]

On Thu, Feb 10, 2005 at 10:49:22AM -0700, Bjorn Helgaas wrote:
> On Thu, 2005-02-10 at 12:37 -0500, Adam Belay wrote:
> > We could always add this change to -mm and see what effect it has.  Let me
> > know if you still think this makes sense.
> 
> Yeah, I think that's Russell's plan, which I think is the right one.

And I notice that Andrew released a new -mm with the change in.
Let the testing commence! 8)

-- 
Russell King
 Linux kernel    2.6 ARM Linux   - http://www.arm.linux.org.uk/
 maintainer of:  2.6 PCMCIA      - http://pcmcia.arm.linux.org.uk/
                 2.6 Serial core