Re: [PATCH] pci: x86-32/64 switch to pci_get API

Re: [PATCH] pci: x86-32/64 switch to pci_get API

[Muli Ben-Yehuda]

On Mon, Oct 16, 2006 at 04:51:32PM +0100, Alan Cox wrote:
> Use pci_get_bus_and_slot to find the router (and lock it for the kernel
> lifetime), also use pci_get_device_reverse() to walk the list finding
> calgary IOMMUs
> 
> Signed-off-by: Alan Cox <alan@redhat.com>
> 
> diff -u --new-file --recursive --exclude-from /usr/src/exclude linux.vanilla-2.6.19-rc1-mm1/arch/x86_64/kernel/pci-calgary.c linux-2.6.19-rc1-mm1/arch/x86_64/kernel/pci-calgary.c
> --- linux.vanilla-2.6.19-rc1-mm1/arch/x86_64/kernel/pci-calgary.c	2006-10-13 15:10:06.000000000 +0100
> +++ linux-2.6.19-rc1-mm1/arch/x86_64/kernel/pci-calgary.c	2006-10-13 17:14:40.000000000 +0100
> @@ -879,7 +879,7 @@
>  
>  error:
>  	do {
> -		dev = pci_find_device_reverse(PCI_VENDOR_ID_IBM,
> +		dev = pci_get_device_reverse(PCI_VENDOR_ID_IBM,
>  					      PCI_DEVICE_ID_IBM_CALGARY,
>  					      dev);
>  		if (!dev)

Looks good, this part is

Acked-By: Muli Ben-Yehuda <muli@il.ibm.com>

Cheers,
Muli

[PATCH] pci: x86-32/64 switch to pci_get API

[Alan Cox]

Use pci_get_bus_and_slot to find the router (and lock it for the kernel
lifetime), also use pci_get_device_reverse() to walk the list finding
calgary IOMMUs

Signed-off-by: Alan Cox <alan@redhat.com>

diff -u --new-file --recursive --exclude-from /usr/src/exclude linux.vanilla-2.6.19-rc1-mm1/arch/i386/pci/irq.c linux-2.6.19-rc1-mm1/arch/i386/pci/irq.c
--- linux.vanilla-2.6.19-rc1-mm1/arch/i386/pci/irq.c	2006-10-13 15:10:06.000000000 +0100
+++ linux-2.6.19-rc1-mm1/arch/i386/pci/irq.c	2006-10-13 17:14:40.000000000 +0100
@@ -758,7 +758,7 @@
 	DBG(KERN_DEBUG "PCI: Attempting to find IRQ router for %04x:%04x\n",
 	    rt->rtr_vendor, rt->rtr_device);
 
-	pirq_router_dev = pci_find_slot(rt->rtr_bus, rt->rtr_devfn);
+	pirq_router_dev = pci_get_bus_and_slot(rt->rtr_bus, rt->rtr_devfn);
 	if (!pirq_router_dev) {
 		DBG(KERN_DEBUG "PCI: Interrupt router not found at "
 			"%02x:%02x\n", rt->rtr_bus, rt->rtr_devfn);
@@ -778,6 +778,8 @@
 		pirq_router_dev->vendor,
 		pirq_router_dev->device,
 		pci_name(pirq_router_dev));
+		
+	/* The device remains referenced for the kernel lifetime */
 }
 
 static struct irq_info *pirq_get_info(struct pci_dev *dev)
diff -u --new-file --recursive --exclude-from /usr/src/exclude linux.vanilla-2.6.19-rc1-mm1/arch/x86_64/kernel/pci-calgary.c linux-2.6.19-rc1-mm1/arch/x86_64/kernel/pci-calgary.c
--- linux.vanilla-2.6.19-rc1-mm1/arch/x86_64/kernel/pci-calgary.c	2006-10-13 15:10:06.000000000 +0100
+++ linux-2.6.19-rc1-mm1/arch/x86_64/kernel/pci-calgary.c	2006-10-13 17:14:40.000000000 +0100
@@ -879,7 +879,7 @@
 
 error:
 	do {
-		dev = pci_find_device_reverse(PCI_VENDOR_ID_IBM,
+		dev = pci_get_device_reverse(PCI_VENDOR_ID_IBM,
 					      PCI_DEVICE_ID_IBM_CALGARY,
 					      dev);
 		if (!dev)

Re: [PATCH] pci: x86-32/64 switch to pci_get API

[Andi Kleen]

> --- linux.vanilla-2.6.19-rc1-mm1/arch/x86_64/kernel/pci-calgary.c	2006-10-13 15:10:06.000000000 +0100
> +++ linux-2.6.19-rc1-mm1/arch/x86_64/kernel/pci-calgary.c	2006-10-13 17:14:40.000000000 +0100
> @@ -879,7 +879,7 @@
>  
>  error:
>  	do {
> -		dev = pci_find_device_reverse(PCI_VENDOR_ID_IBM,
> +		dev = pci_get_device_reverse(PCI_VENDOR_ID_IBM,
>  					      PCI_DEVICE_ID_IBM_CALGARY,
>  					      dev);

No put call anywhere?

-Andi

Re: [PATCH] pci: x86-32/64 switch to pci_get API

[Andi Kleen]

On Mon, Oct 16, 2006 at 05:45:13PM +0100, Alan Cox wrote:
> Ar Llu, 2006-10-16 am 18:07 +0200, ysgrifennodd Andi Kleen:
> > > --- linux.vanilla-2.6.19-rc1-mm1/arch/x86_64/kernel/pci-calgary.c	2006-10-13 15:10:06.000000000 +0100
> > > +++ linux-2.6.19-rc1-mm1/arch/x86_64/kernel/pci-calgary.c	2006-10-13 17:14:40.000000000 +0100
> > > @@ -879,7 +879,7 @@
> > >  
> > >  error:
> > >  	do {
> > > -		dev = pci_find_device_reverse(PCI_VENDOR_ID_IBM,
> > > +		dev = pci_get_device_reverse(PCI_VENDOR_ID_IBM,
> > >  					      PCI_DEVICE_ID_IBM_CALGARY,
> > >  					      dev);
> > 
> > No put call anywhere?
> 
> You can't hot unplug your MMU

Not sure about that. Calgary is afaik in the bridges and since Summit
has pluggable PCI cages and nodes i would assume the MMU instances are also
hot pluggables.

-Andi

Re: [PATCH] pci: x86-32/64 switch to pci_get API

[Alan Cox]

Ar Llu, 2006-10-16 am 18:07 +0200, ysgrifennodd Andi Kleen:
> > --- linux.vanilla-2.6.19-rc1-mm1/arch/x86_64/kernel/pci-calgary.c	2006-10-13 15:10:06.000000000 +0100
> > +++ linux-2.6.19-rc1-mm1/arch/x86_64/kernel/pci-calgary.c	2006-10-13 17:14:40.000000000 +0100
> > @@ -879,7 +879,7 @@
> >  
> >  error:
> >  	do {
> > -		dev = pci_find_device_reverse(PCI_VENDOR_ID_IBM,
> > +		dev = pci_get_device_reverse(PCI_VENDOR_ID_IBM,
> >  					      PCI_DEVICE_ID_IBM_CALGARY,
> >  					      dev);
> 
> No put call anywhere?

You can't hot unplug your MMU

Re: [PATCH] pci: x86-32/64 switch to pci_get API

[Alan Cox]

Ar Llu, 2006-10-16 am 18:24 +0200, ysgrifennodd Andi Kleen:
> > You can't hot unplug your MMU
> 
> Not sure about that. Calgary is afaik in the bridges and since Summit
> has pluggable PCI cages and nodes i would assume the MMU instances are also
> hot pluggables.

If so Linux doesn't currently support that and the patch keeps things as
they are except for using hotplug safe APIs (and since I want to
exterminate pci_find_device* shortly thats preferable)

Alan

Re: [PATCH] pci: x86-32/64 switch to pci_get API

[Muli Ben-Yehuda]

On Mon, Oct 16, 2006 at 06:07:59PM +0200, Andi Kleen wrote:
> > --- linux.vanilla-2.6.19-rc1-mm1/arch/x86_64/kernel/pci-calgary.c	2006-10-13 15:10:06.000000000 +0100
> > +++ linux-2.6.19-rc1-mm1/arch/x86_64/kernel/pci-calgary.c	2006-10-13 17:14:40.000000000 +0100
> > @@ -879,7 +879,7 @@
> >  
> >  error:
> >  	do {
> > -		dev = pci_find_device_reverse(PCI_VENDOR_ID_IBM,
> > +		dev = pci_get_device_reverse(PCI_VENDOR_ID_IBM,
> >  					      PCI_DEVICE_ID_IBM_CALGARY,
> >  					      dev);
> 
> No put call anywhere?

The put is implicit in the call, since for each dev returned from
pci_get_device_reverse() we then call it with the same dev as the last
('from') parameter. We take a reference to dev elsewhere, just before
we start using it.

In any case these bridges can't be hot-plugged, so as long as the ref
count remains positive, we're fine.

Cheers,
Muli

Re: [PATCH] pci: x86-32/64 switch to pci_get API

[Andi Kleen]

On Mon, Oct 16, 2006 at 06:05:40PM +0100, Alan Cox wrote:
> Ar Llu, 2006-10-16 am 18:24 +0200, ysgrifennodd Andi Kleen:
> > > You can't hot unplug your MMU
> > 
> > Not sure about that. Calgary is afaik in the bridges and since Summit
> > has pluggable PCI cages and nodes i would assume the MMU instances are also
> > hot pluggables.
> 
> If so Linux doesn't currently support that and the patch keeps things as
> they are except for using hotplug safe APIs (and since I want to
> exterminate pci_find_device* shortly thats preferable)
> 

Ok i applied the patch to -rc2, but it results in 

arch/x86_64/pci/built-in.o: In function `pcibios_irq_init':
irq.c:(.init.text+0xc7e): undefined reference to `pci_get_bus_and_slot'

That function is also nowhere to be found:

% gid pci_get_bus_and_slot
%

So dropped again.

-Andi

Re: [PATCH] pci: x86-32/64 switch to pci_get API

[Muli Ben-Yehuda]

On Mon, Oct 16, 2006 at 09:01:15PM +0200, Andi Kleen wrote:
> On Mon, Oct 16, 2006 at 06:05:40PM +0100, Alan Cox wrote:
> > Ar Llu, 2006-10-16 am 18:24 +0200, ysgrifennodd Andi Kleen:
> > > > You can't hot unplug your MMU
> > > 
> > > Not sure about that. Calgary is afaik in the bridges and since Summit
> > > has pluggable PCI cages and nodes i would assume the MMU instances are also
> > > hot pluggables.
> > 
> > If so Linux doesn't currently support that and the patch keeps things as
> > they are except for using hotplug safe APIs (and since I want to
> > exterminate pci_find_device* shortly thats preferable)
> > 
> 
> Ok i applied the patch to -rc2, but it results in 
> 
> arch/x86_64/pci/built-in.o: In function `pcibios_irq_init':
> irq.c:(.init.text+0xc7e): undefined reference to `pci_get_bus_and_slot'
> 
> That function is also nowhere to be found:
> 
> % gid pci_get_bus_and_slot
> %
> 
> So dropped again.

Alain submitted the patch to add that function roughly at the same
time. See
http://marc.theaimsgroup.com/?l=linux-kernel&m=116101265428620&w=2

Cheers,
Muli

Re: [PATCH] pci: x86-32/64 switch to pci_get API

[Alan Cox]

Ar Llu, 2006-10-16 am 21:01 +0200, ysgrifennodd Andi Kleen:
> Ok i applied the patch to -rc2, but it results in 
> 
> arch/x86_64/pci/built-in.o: In function `pcibios_irq_init':
> irq.c:(.init.text+0xc7e): undefined reference to `pci_get_bus_and_slot'
> 
> That function is also nowhere to be found:

I was just cc'ing you on a set of patches for -mm which included this
one and the other neccessary PCI changes. Sorry for any confusion.