intel-agp: skip non-AGP devices

intel-agp: skip non-AGP devices

[Matt Domsch]

[-- Attachment #1: Type: text/plain, Size: 1558 bytes --]

Dave, 

On our PowerEdge 2600 system, which has an Intel E7501 Memroy
Controller Hub, the intel-agp probe code is reporting, at KERN_ERR no less:

agpgart: Unsupported Intel chipset (device id 254c)

Now, of course it says this, as this device does not present itself as
AGP-capable:

00:00.0 Host bridge: Intel Corp. E7501 Memory Controller Hub (rev 01)
	Control: I/O- Mem+ BusMaster+ SpecCycle- MemWINV- VGASnoop- ParErr+ Stepping- SERR- FastB2B-
	Status: Cap+ 66Mhz- UDF- FastB2B+ ParErr- DEVSEL=fast >TAbort- <TAbort- <MAbort- >SERR- <PERR-
	Latency: 0
	Capabilities: [40] #09 [1105]


agp_intel_probe()  calls pci_find_capability(PCI_CAP_ID_AGP)
but doesn't check the return value (should be zero in this case) prior
to moving into the switch.


The patch below checks for a valid cap_ptr prior to printing the
message, now at KERN_WARNING level (it's not really an error, is it?)

Thoughts?

Thanks,
Matt

-- 
Matt Domsch
Sr. Software Engineer, Lead Engineer
Dell Linux Solutions linux.dell.com & www.dell.com/linux
Linux on Dell mailing lists @ http://lists.us.dell.com

--- intel-agp.c.orig	Tue Jun  1 10:45:59 2004
+++ intel-agp.c	Tue Jun  1 11:02:56 2004
@@ -1382,8 +1382,9 @@ static int __devinit agp_intel_probe(str
 		name = "E7205";
 		break;
 	default:
-		printk(KERN_ERR PFX "Unsupported Intel chipset (device id: %04x)\n",
-			    pdev->device);
+		if (cap_ptr)
+			printk(KERN_WARNING PFX "Unsupported Intel chipset (device id: %04x)\n",
+			       pdev->device);
 		return -ENODEV;
 	};
 

[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: intel-agp: skip non-AGP devices

[Christoph Hellwig]

On Tue, Jun 01, 2004 at 11:04:57AM -0500, Matt Domsch wrote:
> Dave, 
> 
> On our PowerEdge 2600 system, which has an Intel E7501 Memroy
> Controller Hub, the intel-agp probe code is reporting, at KERN_ERR no less:
> 
> agpgart: Unsupported Intel chipset (device id 254c)
> 

...

> agp_intel_probe()  calls pci_find_capability(PCI_CAP_ID_AGP)
> but doesn't check the return value (should be zero in this case) prior
> to moving into the switch.
> 
> 
> The patch below checks for a valid cap_ptr prior to printing the
> message, now at KERN_WARNING level (it's not really an error, is it?)

The real problem is that agpgart doesn't properly fill in the pci_id
table but claims all devices and then does it's own probing internally.

This also breaks hotplug in a funny way.

Re: intel-agp: skip non-AGP devices

[Dave Jones]

On Tue, Jun 01, 2004 at 11:04:57AM -0500, Matt Domsch wrote:

 > On our PowerEdge 2600 system, which has an Intel E7501 Memroy
 > Controller Hub, the intel-agp probe code is reporting, at KERN_ERR no less:
 > 
 > agpgart: Unsupported Intel chipset (device id 254c)
 > 
 > Now, of course it says this, as this device does not present itself as
 > AGP-capable:
 > 
 > The patch below checks for a valid cap_ptr prior to printing the
 > message, now at KERN_WARNING level (it's not really an error, is it?)

indeed, patch applied.

Thanks,

		Dave

Re: intel-agp: skip non-AGP devices

[Dave Jones]

On Tue, Jun 01, 2004 at 05:20:58PM +0100, Christoph Hellwig wrote:

 > > The patch below checks for a valid cap_ptr prior to printing the
 > > message, now at KERN_WARNING level (it's not really an error, is it?)
 > 
 > The real problem is that agpgart doesn't properly fill in the pci_id
 > table but claims all devices and then does it's own probing internally.
 > This also breaks hotplug in a funny way.

This is fixed in agpgart-bk / -mm.  Andi went through all the drivers
adding their id's.  Should be going to Linus soon.

		Dave

Re: intel-agp: skip non-AGP devices

[Matt Domsch]

[-- Attachment #1: Type: text/plain, Size: 1061 bytes --]

On Tue, Jun 01, 2004 at 05:31:00PM +0100, Dave Jones wrote:
> On Tue, Jun 01, 2004 at 05:20:58PM +0100, Christoph Hellwig wrote:
> 
>  > > The patch below checks for a valid cap_ptr prior to printing the
>  > > message, now at KERN_WARNING level (it's not really an error, is it?)
>  > 
>  > The real problem is that agpgart doesn't properly fill in the pci_id
>  > table but claims all devices and then does it's own probing internally.
>  > This also breaks hotplug in a funny way.
> 
> This is fixed in agpgart-bk / -mm.  Andi went through all the drivers
> adding their id's.  Should be going to Linus soon.
> 
> 		Dave

FWIW, sworks-agp.c has the same issue in mainline yet today.
agpgart: Unsupported Serverworks chipset (device id: 0011) 
agpgart: Unsupported Serverworks chipset (device id: 0201) 

I'll look at -mm to verify it's fixed there.

Thanks,
Matt

-- 
Matt Domsch
Sr. Software Engineer, Lead Engineer
Dell Linux Solutions linux.dell.com & www.dell.com/linux
Linux on Dell mailing lists @ http://lists.us.dell.com

[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: intel-agp: skip non-AGP devices

[Dave Jones]

On Tue, Jun 08, 2004 at 11:00:27AM -0500, Matt Domsch wrote:

 > > This is fixed in agpgart-bk / -mm.  Andi went through all the drivers
 > > adding their id's.  Should be going to Linus soon.
 > 
 > FWIW, sworks-agp.c has the same issue in mainline yet today.
 > agpgart: Unsupported Serverworks chipset (device id: 0011) 
 > agpgart: Unsupported Serverworks chipset (device id: 0201) 
 > 
 > I'll look at -mm to verify it's fixed there.

Andi did all of the drivers iirc, and these changes are now
in mainline btw.

		Dave

Re: intel-agp: skip non-AGP devices

[Matt Domsch]

[-- Attachment #1: Type: text/plain, Size: 2344 bytes --]

On Tue, Jun 08, 2004 at 11:00:27AM -0500, Matt Domsch wrote:
> On Tue, Jun 01, 2004 at 05:31:00PM +0100, Dave Jones wrote:
> > On Tue, Jun 01, 2004 at 05:20:58PM +0100, Christoph Hellwig wrote:
> > 
> >  > > The patch below checks for a valid cap_ptr prior to printing the
> >  > > message, now at KERN_WARNING level (it's not really an error, is it?)
> >  > 
> >  > The real problem is that agpgart doesn't properly fill in the pci_id
> >  > table but claims all devices and then does it's own probing internally.
> >  > This also breaks hotplug in a funny way.
> > 
> > This is fixed in agpgart-bk / -mm.  Andi went through all the drivers
> > adding their id's.  Should be going to Linus soon.
> > 
> > 		Dave
> 
> FWIW, sworks-agp.c has the same issue in mainline yet today.
> agpgart: Unsupported Serverworks chipset (device id: 0011) 
> agpgart: Unsupported Serverworks chipset (device id: 0201) 
> 
> I'll look at -mm to verify it's fixed there.

agpgart-bk and -mm didn't add proper PCI ID lists to sworks-agp.c (yet
I assume).  Patch below does the same for this as I submitted for
Intel previously.  It only prints a warning now if the device is AGP
capable but unrecognized.

Thanks,
Matt

-- 
Matt Domsch
Sr. Software Engineer, Lead Engineer
Dell Linux Solutions linux.dell.com & www.dell.com/linux
Linux on Dell mailing lists @ http://lists.us.dell.com

===== drivers/char/agp/sworks-agp.c 1.41 vs edited =====
--- 1.41/drivers/char/agp/sworks-agp.c	Tue Jun  1 02:50:11 2004
+++ edited/drivers/char/agp/sworks-agp.c	Tue Jun  8 11:13:38 2004
@@ -447,6 +447,7 @@
 	struct agp_bridge_data *bridge;
 	struct pci_dev *bridge_dev;
 	u32 temp, temp2;
+	u8 cap_ptr = 0;
 
 	/* Everything is on func 1 here so we are hardcoding function one */
 	bridge_dev = pci_find_slot((unsigned int)pdev->bus->number,
@@ -457,6 +458,8 @@
 		return -ENODEV;
 	}
 
+	cap_ptr = pci_find_capability(pdev, PCI_CAP_ID_AGP);
+
 	switch (pdev->device) {
 	case 0x0006:
 		/* ServerWorks CNB20HE
@@ -470,8 +473,9 @@
 		break;
 
 	default:
-		printk(KERN_ERR PFX "Unsupported Serverworks chipset "
-				"(device id: %04x)\n", pdev->device);
+		if (cap_ptr)
+			printk(KERN_WARNING PFX "Unsupported Serverworks chipset "
+			       "(device id: %04x)\n", pdev->device);
 		return -ENODEV;
 	}
 

[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: intel-agp: skip non-AGP devices

[Dave Jones]

On Tue, Jun 08, 2004 at 11:17:45AM -0500, Matt Domsch wrote:

 > agpgart-bk and -mm didn't add proper PCI ID lists to sworks-agp.c (yet
 > I assume).  Patch below does the same for this as I submitted for
 > Intel previously.  It only prints a warning now if the device is AGP
 > capable but unrecognized.

Cool, thanks Matt. I'll look over the other drivers to double check.

		Dave

Re: intel-agp: skip non-AGP devices

[Dave Jones]

On Tue, Jun 08, 2004 at 11:17:45AM -0500, Matt Domsch wrote:

 > agpgart-bk and -mm didn't add proper PCI ID lists to sworks-agp.c (yet
 > I assume).  Patch below does the same for this as I submitted for
 > Intel previously.  It only prints a warning now if the device is AGP
 > capable but unrecognized.

Ok, the other drivers do sensible things like checking for cap_ptr,
and -ENODEV'ing if we don't find it. Too bad there are some wacky
Intel/Sworks chipsets out there that don't have the AGP header,
yet have AGP.

		Dave