[PATCH 2.6.29-rc7 1/2] x86: fix iommu=nodac parameter handling

[PATCH 2.6.29-rc7 1/2] x86: fix iommu=nodac parameter handling

[Tejun Heo]

nodac should forbid dac not instead of enabling it.  Fix it.

Signed-off-by: Tejun Heo <tj@kernel.org>
---
 arch/x86/kernel/pci-dma.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/arch/x86/kernel/pci-dma.c b/arch/x86/kernel/pci-dma.c
index b254285..69c7a8d 100644
--- a/arch/x86/kernel/pci-dma.c
+++ b/arch/x86/kernel/pci-dma.c
@@ -197,7 +197,7 @@ static __init int iommu_setup(char *p)
 		if (!strncmp(p, "allowdac", 8))
 			forbid_dac = 0;
 		if (!strncmp(p, "nodac", 5))
-			forbid_dac = -1;
+			forbid_dac = 1;
 		if (!strncmp(p, "usedac", 6)) {
 			forbid_dac = -1;
 			return 1;

[PATCH 2.6.29-rc7 2/2] x86: disallow DAC for MCP51 PCI bridge

[Tejun Heo]

MCP51 corrupts DAC transfers.  Disallow it.  Reported by pgnet on
bnc#463829.

  https://bugzilla.novell.com/show_bug.cgi?id=463829

Signed-off-by: Tejun Heo <tj@kernel.org>
Reported-by: pgnet <pgnet.trash@gmail.com>
---
 arch/x86/kernel/pci-dma.c |   14 ++++++++++++++
 1 file changed, 14 insertions(+)

diff --git a/arch/x86/kernel/pci-dma.c b/arch/x86/kernel/pci-dma.c
index b254285..7b61798 100644
--- a/arch/x86/kernel/pci-dma.c
+++ b/arch/x86/kernel/pci-dma.c
@@ -296,4 +296,18 @@ static __devinit void via_no_dac(struct pci_dev *dev)
 	}
 }
 DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_VIA, PCI_ANY_ID, via_no_dac);
+
+/*
+ * MCP51 PCI bridge corrupts data for DAC.  Disable it.  Reported in
+ * bnc#463829.
+ */
+static __devinit void mcp51_no_dac(struct pci_dev *dev)
+{
+	if (forbid_dac == 0) {
+		printk(KERN_INFO
+		       "PCI: MCP51 PCI bridge detected. Disabling DAC.\n");
+		forbid_dac = 1;
+	}
+}
+DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_NVIDIA, 0x026f, mcp51_no_dac);
 #endif

Re: [PATCH 2.6.29-rc7 2/2] x86: disallow DAC for MCP51 PCI bridge

[Alan Cox]

On Wed, 04 Mar 2009 11:39:15 +0900
Tejun Heo <tj@kernel.org> wrote:

> MCP51 corrupts DAC transfers.  Disallow it.  Reported by pgnet on
> bnc#463829.

So you have a single piece of evidence as I read it - from a single
board that represents system with a very common chipset and no supporting
errata information from the vendor.

That bug could easily be a faulty motherboard or dodgy connector or a
specific hardware compatibility issue between two specific components. I
don't think that can be considered sufficient evidence for adding
sledgehammer sized workarounds that harm everyone elses performance and
cripples such systems for certain kinds of work (eg video)

Insufficient evidence and google searches and other bug searches don't
seem to be turning up any parallel cases so for the moment At the very
least someone needs to reproduce the hardware configuration and see the
same problem, and then repeat with a different DAC enabled card.

I don't believe this patch should go in ... yet. More information/testing
is needed.

Alan

Re: [PATCH 2.6.29-rc7 2/2] x86: disallow DAC for MCP51 PCI bridge

[Tejun Heo]

Alan Cox wrote:
> On Wed, 04 Mar 2009 11:39:15 +0900
> Tejun Heo <tj@kernel.org> wrote:
> 
>> MCP51 corrupts DAC transfers.  Disallow it.  Reported by pgnet on
>> bnc#463829.
> 
> So you have a single piece of evidence as I read it - from a single
> board that represents system with a very common chipset and no supporting
> errata information from the vendor.
> 
> That bug could easily be a faulty motherboard or dodgy connector or a
> specific hardware compatibility issue between two specific components. I
> don't think that can be considered sufficient evidence for adding
> sledgehammer sized workarounds that harm everyone elses performance and
> cripples such systems for certain kinds of work (eg video)
> 
> Insufficient evidence and google searches and other bug searches don't
> seem to be turning up any parallel cases so for the moment At the very
> least someone needs to reproduce the hardware configuration and see the
> same problem, and then repeat with a different DAC enabled card.
> 
> I don't believe this patch should go in ... yet. More information/testing
> is needed.

x86 is blacklisting DAC for everything VIA. :-)

* MCP51 is a very old chipset at this point (it's circa 2005).  Not
  many machine would be running with >4GB memory to begin with.

* Given the above and scarcity of DAC on most end user machines
  (nothing on MCP51 does DAC by default), lack of reports isn't too
  surprising.

* The board doesn't have a 64bit connector.  It can't be dodgy
  connector and sil24 is known to behave well with DAC.  The failure
  being specific to the particular machine doesn't seem likely.

I'll ping nvidia about it but I think your bar is too high.

Thanks.

-- 
tejun

Re: [PATCH 2.6.29-rc7 2/2] x86: disallow DAC for MCP51 PCI bridge

[Alan Cox]

> x86 is blacklisting DAC for everything VIA. :-)

I'd assume that comes from published docs given VIA actually publish
stuff if you ask nicely.
> 
> * MCP51 is a very old chipset at this point (it's circa 2005).  Not
>   many machine would be running with >4GB memory to begin with.
> 
> * Given the above and scarcity of DAC on most end user machines
>   (nothing on MCP51 does DAC by default), lack of reports isn't too
>   surprising.
> 
> * The board doesn't have a 64bit connector.  It can't be dodgy
>   connector and sil24 is known to behave well with DAC.  The failure
>   being specific to the particular machine doesn't seem likely.
> 
> I'll ping nvidia about it but I think your bar is too high.

I don't think wanting to see *two* examples is a high bar.

Re: [PATCH 2.6.29-rc7 2/2] x86: disallow DAC for MCP51 PCI bridge

[Tejun Heo]

Alan Cox wrote:
>> x86 is blacklisting DAC for everything VIA. :-)
> 
> I'd assume that comes from published docs given VIA actually publish
> stuff if you ask nicely.

Dunno the history but it's from the time before VIA begins publishing.
It's probably worthwhile to refine the blacklisting.

>> * MCP51 is a very old chipset at this point (it's circa 2005).  Not
>>   many machine would be running with >4GB memory to begin with.
>>
>> * Given the above and scarcity of DAC on most end user machines
>>   (nothing on MCP51 does DAC by default), lack of reports isn't too
>>   surprising.
>>
>> * The board doesn't have a 64bit connector.  It can't be dodgy
>>   connector and sil24 is known to behave well with DAC.  The failure
>>   being specific to the particular machine doesn't seem likely.
>>
>> I'll ping nvidia about it but I think your bar is too high.
> 
> I don't think wanting to see *two* examples is a high bar.

I don't think it's an easy combination.  Circa 2005 desktop machine
with >4GB ram + DAC capable controller running Linux with owner who
will report seemingly random data corruption which will end up in the
correct hands and given that non-working DAC isn't too surprising of
the machines of that time and isn't very likely caused by random
component or connection failure, I think we'll be better off just
blacklisting it.  That said, I admit I'm more trigger happy with
blacklists than some people (I think it's generally better to have
working systems than chasing performance on not-so-common cases).

Anyways, I pinged Peer Chen on the issue.  Let's see whether he can
confirm it.

Thanks.

-- 
tejun