Skip to contentSkip to conversation

x86-64 build fix

PATCH In review Jeff Garzik 6 participants 11 messages Jun 22, 2006 – Jun 23, 2006 Agpgart driver

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 x86-64 build fix

As of last night, I still needed the Kconfig patch below to successfully build allmodconfig on x86-64. I believe Andrew has the patch with a proper description and attribution, so I would prefer that he send it...

Jeff

arch/x86_64/Kconfig +2 -3
index 408d44a..7d3bc5a 100644
@@ -389,6 +389,7 @@ config GART_IOMMU
 	bool "K8 GART IOMMU support"
 	default y
 	select SWIOTLB
+	select AGP
 	depends on PCI
 	help
 	  Support for hardware IOMMU in AMD's Opteron/Athlon64 Processors
@@ -401,11 +402,9 @@ config GART_IOMMU
   	  northbridge and a software emulation used on other systems without
 	  hardware IOMMU.  If unsure, say Y.
 
-# need this always enabled with GART_IOMMU for the VIA workaround
+# need this always selected by GART_IOMMU for the VIA workaround
 config SWIOTLB
 	bool
-	default y
-	depends on GART_IOMMU
 
 config X86_MCE
 	bool "Machine check support" if EMBEDDED
drivers/char/agp/Kconfig +1 -2
index 7c88c06..46685a5 100644
@@ -1,7 +1,6 @@
 config AGP
-	tristate "/dev/agpgart (AGP Support)" if !GART_IOMMU
+	tristate "/dev/agpgart (AGP Support)"
 	depends on ALPHA || IA64 || PPC || X86
-	default y if GART_IOMMU
 	---help---
 	  AGP (Accelerated Graphics Port) is a bus system mainly used to
 	  connect graphics cards to the rest of the system.
#1 lore raw
Quoted 5 lines · Jeff Garzik

As of last night, I still needed the Kconfig patch below to successfully build allmodconfig on x86-64. I believe Andrew has the patch with a proper description and attribution, so I would prefer that he send it...

It was transferred from the -mm-only stuff and into the x86_64 tree, so Andi owns it now.

I'll steal it back and will send it to Linus.

#2 lore raw
Quoted 11 lines · Andrew Morton
Quoted 5 lines · Jeff Garzik

As of last night, I still needed the Kconfig patch below to successfully build allmodconfig on x86-64. I believe Andrew has the patch with a proper description and attribution, so I would prefer that he send it...

It was transferred from the -mm-only stuff and into the x86_64 tree, so Andi owns it now.

I'll steal it back and will send it to Linus.

It's in my pile to eventually send, but so far nobody was able to explain to me why it is suddenly needed at all, so I wasn't feeling very comfortable with it.

-Andi

#3 lore raw
Quoted 9 lines · Andi Kleen
Quoted 7 lines · Andrew Morton
Quoted 5 lines · Jeff Garzik

As of last night, I still needed the Kconfig patch below to successfully build allmodconfig on x86-64. I believe Andrew has the patch with a proper description and attribution, so I would prefer that he send it...

Quoted 9 lines
Quoted 4 lines

It was transferred from the -mm-only stuff and into the x86_64 tree, so Andi owns it now.

I'll steal it back and will send it to Linus.

It's in my pile to eventually send, but so far nobody was able to explain to me why it is suddenly needed at all, so I wasn't feeling very comfortable with it.

urgh, we worked all this out several weeks ago. Forget.

Looking at the patch, I guess the problem is the

config AGP
default y if GART_IOMMU

which doesn't dtrt if you already have AGP=m.

The patch does this, instead:

config GART_IOMMU
select AGP

which gives us AGP=y if GART_IOMMU=y.

I don't think Jeff has sent us an example .config, but I hit this a few times too, before we fixed it. I think this was all triggered by a Kconfig change in the AGP tree, so you wouldn't have seen it, but -mm did.

#4 lore raw
Quoted 3 lines · Andrew Morton

I don't think Jeff has sent us an example .config, but I hit this a few times too, before we fixed it. I think this was all triggered by a Kconfig change in the AGP tree, so you wouldn't have seen it, but -mm did.

Not guilty your honour.
See my mail..
Subject: Re: intelfb: enable on x86_64

It got busted due to INTEL_FB becoming available on x86-64, and it doing select AGP which if INTEL_FB was =m turned CONFIG_AGP=y into CONFIG_AGP=m for who-knows-why reason. This busts the IOMMU code which expects the AGP code to be =y or =n

Dave

Signature · 1 line
http://www.codemonkey.org.uk
#5 lore raw

Hi,

Quoted 6 lines · Dave Jones
Quoted 3 lines · Andrew Morton

I don't think Jeff has sent us an example .config, but I hit this a few times too, before we fixed it. I think this was all triggered by a Kconfig change in the AGP tree, so you wouldn't have seen it, but -mm did.

Quoted 8 lines

Not guilty your honour.
See my mail..
Subject: Re: intelfb: enable on x86_64

It got busted due to INTEL_FB becoming available on x86-64, and it doing select AGP which if INTEL_FB was =m turned CONFIG_AGP=y into CONFIG_AGP=m for who-knows-why reason. This busts the IOMMU code which expects the AGP code to be =y or =n

The select overrides the default (it actually overrides pretty much everything, that's why one should be careful with it).

bye, Roman

#6 lore raw
Quoted 14 lines · On Fri, 23 Jun 2006 01:28:01 +0200 (CEST) Roman Zippel wrote:

Hi,

Quoted 10 lines · Dave Jones
Quoted 3 lines · Andrew Morton

I don't think Jeff has sent us an example .config, but I hit this a few times too, before we fixed it. I think this was all triggered by a Kconfig change in the AGP tree, so you wouldn't have seen it, but -mm did.

Not guilty your honour.
See my mail..
Subject: Re: intelfb: enable on x86_64

Quoted 7 lines
Quoted 4 lines

It got busted due to INTEL_FB becoming available on x86-64, and it doing select AGP which if INTEL_FB was =m turned CONFIG_AGP=y into CONFIG_AGP=m for who-knows-why reason. This busts the IOMMU code which expects the AGP code to be =y or =n

The select overrides the default (it actually overrides pretty much everything, that's why one should be careful with it).

Just checking: so the "select AGP"
sets CONFIG_AGP to the same level or value (m or y) as the option where it is being used?

---
~Randy

#7 lore raw

Hi,

Quoted 6 lines · Randy.Dunlap
Quoted 2 lines

The select overrides the default (it actually overrides pretty much everything, that's why one should be careful with it).

Just checking: so the "select AGP"
sets CONFIG_AGP to the same level or value (m or y) as the option where it is being used?

To be exact it defines the minimum value, so if it's selected as m, the user might still be able to choose between m or y.

bye, Roman

#8 lore raw
In reply to Andrew Morton
Quoted 3 lines · Andrew Morton

I don't think Jeff has sent us an example .config, but I hit this a few times too, before we fixed it. I think this was all triggered by a Kconfig change in the AGP tree, so you wouldn't have seen it, but -mm did.

'make allmodconfig' on x86-64. You can create this .config yourself :)

I think the tree suffers [sometimes due to me :(] when 'allyesconfig' and 'allmodconfig' stop working.

Jeff

#9 lore raw
Quoted 9 lines · Jeff Garzik
Quoted 3 lines · Andrew Morton

I don't think Jeff has sent us an example .config, but I hit this a few times too, before we fixed it. I think this was all triggered by a Kconfig change in the AGP tree, so you wouldn't have seen it, but -mm did.

'make allmodconfig' on x86-64. You can create this .config yourself :)

I think the tree suffers [sometimes due to me :(] when 'allyesconfig' and 'allmodconfig' stop working.

Yes, but they work in 2.6.17 right?
And nothing should have changed since 2.6.17 so far.

But apparently something did. What was it?

-Andi

#10 lore raw
Quoted 6 lines · Andi Kleen
Quoted 5 lines · Jeff Garzik
Quoted 3 lines · Andrew Morton

I don't think Jeff has sent us an example .config, but I hit this a few times too, before we fixed it. I think this was all triggered by a Kconfig change in the AGP tree, so you wouldn't have seen it, but -mm did.

Quoted 6 lines
Quoted 4 lines

'make allmodconfig' on x86-64. You can create this .config yourself :)

I think the tree suffers [sometimes due to me :(] when 'allyesconfig' and 'allmodconfig' stop working.

Yes, but they work in 2.6.17 right?

worked for me, it only broke as of .17-git2 or so.

Quoted 2 lines

And nothing should have changed since 2.6.17 so far. But apparently something did. What was it?

(for the 3rd time today) INTEL_FB now builds on x86-64, and selects AGP, so if INTEL_FB=m, AGP becomes =m too, which breaks the select AGP which GART_IOMMU wants.

Dave

Signature · 1 line
http://www.codemonkey.org.uk