Linux MIPS PCI resource sanity check

Linux MIPS PCI resource sanity check

[Michael Buesch]

Hi,

There's a sanity check in pcibios_enable_resources() that looks like this:

	r = &dev->resource[idx];
	if (!r->start && r->end) {
		printk(KERN_ERR "PCI: Device %s not available because of resource collisions\n", pci_name(dev));
		return -EINVAL;
	}

What is this check actually doing? It triggers for me on a BCM4318 device
which is behind a BCM4710 PCI bridge.
r->start is 0 and r->end is 0x1FFF when this triggers.
If I simply comment out that check the device is detected correctly
and seems to initialize just fine.

-- 
Greetings Michael.

Re: Linux MIPS PCI resource sanity check

[Sergei Shtylyov]

Michael Buesch wrote:

> There's a sanity check in pcibios_enable_resources() that looks like this:

> 	r = &dev->resource[idx];
> 	if (!r->start && r->end) {
> 		printk(KERN_ERR "PCI: Device %s not available because of resource collisions\n", pci_name(dev));
> 		return -EINVAL;
> 	}
> 
> What is this check actually doing?

   It makes sure that a PCI resource is allocated (base of 0 means that it's 
unallocated due to previously detected resource conlict (or some other reason).

> It triggers for me on a BCM4318 device which is behind a BCM4710 PCI bridge.
> r->start is 0 and r->end is 0x1FFF when this triggers.
> If I simply comment out that check the device is detected correctly
> and seems to initialize just fine.

    No, that failnig resource should be relocated.

WBR, Sergei

Re: Linux MIPS PCI resource sanity check

[Michael Buesch]

On Saturday 16 February 2008 11:49:56 Sergei Shtylyov wrote:
> Michael Buesch wrote:
> 
> > There's a sanity check in pcibios_enable_resources() that looks like this:
> 
> > 	r = &dev->resource[idx];
> > 	if (!r->start && r->end) {
> > 		printk(KERN_ERR "PCI: Device %s not available because of resource collisions\n", pci_name(dev));
> > 		return -EINVAL;
> > 	}
> > 
> > What is this check actually doing?
> 
>    It makes sure that a PCI resource is allocated (base of 0 means that it's 
> unallocated due to previously detected resource conlict (or some other reason).

So well, that's what the error message already told me. ;)
But where is the actual bug? I mean, this tells me there's some collision
for this MMIO resource. Still, I can access the MMIO space just fine.

> > It triggers for me on a BCM4318 device which is behind a BCM4710 PCI bridge.
> > r->start is 0 and r->end is 0x1FFF when this triggers.
> > If I simply comment out that check the device is detected correctly
> > and seems to initialize just fine.
> 
>     No, that failnig resource should be relocated.

How?

-- 
Greetings Michael.

Re: Linux MIPS PCI resource sanity check

[Andrew Sharp]

On Sat, 16 Feb 2008 13:49:56 +0300 Sergei Shtylyov
<sshtylyov@ru.mvista.com> wrote:

> Michael Buesch wrote:
> 
> > There's a sanity check in pcibios_enable_resources() that looks
> > like this:
> 
> > 	r = &dev->resource[idx];
> > 	if (!r->start && r->end) {
> > 		printk(KERN_ERR "PCI: Device %s not available
> > because of resource collisions\n", pci_name(dev)); return -EINVAL;
> > 	}
> > 
> > What is this check actually doing?
> 
>    It makes sure that a PCI resource is allocated (base of 0 means
> that it's unallocated due to previously detected resource conlict (or
> some other reason).


Actually, IIRC, resources are based on what the device requested, so a
device behind a bridge could request a resource starting at 0.  I had
to change this for a system as well.  I changed it to

if (!r->start && !r->end) {

because I couldn't see anything in the code that made r->start == 0 an
improper thing.  Not to mention I couldn't access the device any other
way.  Both being 0 is definitelty bogus.

> > It triggers for me on a BCM4318 device which is behind a BCM4710
> > PCI bridge. r->start is 0 and r->end is 0x1FFF when this triggers.
> > If I simply comment out that check the device is detected correctly
> > and seems to initialize just fine.
> 
>     No, that failnig resource should be relocated.
> 
> WBR, Sergei
> 

Re: Linux MIPS PCI resource sanity check

[Michael Buesch]

On Sunday 17 February 2008 00:35:30 Andrew Sharp wrote:
> Actually, IIRC, resources are based on what the device requested, so a
> device behind a bridge could request a resource starting at 0.  I had
> to change this for a system as well.  I changed it to
> 
> if (!r->start && !r->end) {
> 
> because I couldn't see anything in the code that made r->start == 0 an
> improper thing.  Not to mention I couldn't access the device any other
> way.  Both being 0 is definitelty bogus.

I think what's happening for me is the following:
I have a PCI bridge and behind that bridge is one device.
This has a fixed location and fixed size memory window (hardwired).

register_pci_controller() requires me to pass some io_resource
and mem_resource in the controller struct. So I pass the memory window
which is assigned to the controller and the devices behind it.
Later I fixup the bases and sizes for each resource in the
pcibios_plat_dev_init() routine.

So, well. I still don't know where the mips PCI subsystem would
detect this resource conflict and what that means to me.
If I simply rip out the check everything works fine, as I fixup
the addresses and sizes later anyway. (I fixup more stuff like the
IRQ routing an so on, too).

The code is in drivers/ssb/driver_pcicore.c

-- 
Greetings Michael.

Re: Linux MIPS PCI resource sanity check

[Sergei Shtylyov]

Andrew Sharp wrote:

>>>There's a sanity check in pcibios_enable_resources() that looks
>>>like this:

>>>	r = &dev->resource[idx];
>>>	if (!r->start && r->end) {
>>>		printk(KERN_ERR "PCI: Device %s not available
>>>because of resource collisions\n", pci_name(dev)); return -EINVAL;
>>>	}

>>>What is this check actually doing?

>>   It makes sure that a PCI resource is allocated (base of 0 means
>>that it's unallocated due to previously detected resource conlict (or
>>some other reason).

> Actually, IIRC, resources are based on what the device requested, so a
> device behind a bridge could request a resource starting at 0.  I had

    Zero value in BAR was considered unallocated resource in the PCI 2.2 spec...

> to change this for a system as well.  I changed it to

> if (!r->start && !r->end) {

    r->end can't be 0.

WBR, Sergei

Re: Linux MIPS PCI resource sanity check

[Ralf Baechle]

On Sat, Feb 16, 2008 at 01:49:56PM +0300, Sergei Shtylyov wrote:

>   It makes sure that a PCI resource is allocated (base of 0 means that it's 
> unallocated due to previously detected resource conlict (or some other 
> reason).
>
>> It triggers for me on a BCM4318 device which is behind a BCM4710 PCI bridge.
>> r->start is 0 and r->end is 0x1FFF when this triggers.
>> If I simply comment out that check the device is detected correctly
>> and seems to initialize just fine.
>
>    No, that failnig resource should be relocated.

The resources were assigned during the PCI bus scan so at least with the
current implementation it's too late to reassign resources.  I get the
feeling this is an indication of a problem elsewhere.

  Ralf

Re: Linux MIPS PCI resource sanity check

[Ralf Baechle]

On Sat, Feb 16, 2008 at 11:39:10AM +0100, Michael Buesch wrote:

Can you take a look at ed6d14f9760857c745206c978b80352fc09cfd19 which fixed
a somewhat similar problem for i386, does that seem to be related to your
problem?

The i386 fix makes sense so I'm almost decieded to apply to the MIPS code
even if it should turn out not to make a difference for you.

  Ralf