[pci] pci/mrst: Detect loops when reading fixed BAR cap.

[pci] pci/mrst: Detect loops when reading fixed BAR cap.

[Ben Greear]

The mrst logic introduced in 2.6.33-rc8 in commit
a712ffbc199849364c46e9112b93b66de08e2c26 causes boot
to hang on at least this platform:

Intel E5405 CPU
System Information
	Manufacturer: Supermicro
	Product Name: X7DBU

The cause of the hang is that pci_cap points to itself
as the next capability, putting the fixed_bar_cap into
an endless loop.

This patch detects the loop, prints a warning, and
continues on with useful work.  This strategy was
suggested by Robert Hancock <hancockrwd@gmail.com>

This should be a candidate for 2.6.34.y as well.

Signed-off-by: Ben Greear <greearb@candelatech.com>
---
:100644 100644 1cdc02c... 9535ba9... M	arch/x86/pci/mrst.c
 arch/x86/pci/mrst.c |   11 +++++++++++
 1 files changed, 11 insertions(+), 0 deletions(-)

diff --git a/arch/x86/pci/mrst.c b/arch/x86/pci/mrst.c
index 1cdc02c..9535ba9 100644
--- a/arch/x86/pci/mrst.c
+++ b/arch/x86/pci/mrst.c
@@ -76,6 +76,17 @@ static int fixed_bar_cap(struct pci_bus *bus, unsigned int devfn)
 				return pos;
 		}
 
+		if ((pcie_cap >> 20) == 0)
+			break;
+		
+		if ((pcie_cap >> 20) <= pos) {
+			printk(KERN_WARNING "WARNING: mrst:  detected loop"
+			       " when searching for fixed BAR cap, previous"
+			       " position: 0x%x new position: 0x%x"
+			       " bus-number: %i  devfn: %i\n",
+			       pos, pcie_cap >> 20, bus->number, devfn);
+			break;
+		}
 		pos = pcie_cap >> 20;
 	}
 
-- 
1.6.2.5

Re: [pci] pci/mrst: Detect loops when reading fixed BAR cap.

[Bjorn Helgaas]

On Wednesday, July 14, 2010 11:02:03 am Ben Greear wrote:
> The mrst logic introduced in 2.6.33-rc8 in commit
> a712ffbc199849364c46e9112b93b66de08e2c26 causes boot
> to hang on at least this platform:
> 
> Intel E5405 CPU
> System Information
> 	Manufacturer: Supermicro
> 	Product Name: X7DBU
> 
> The cause of the hang is that pci_cap points to itself
> as the next capability, putting the fixed_bar_cap into
> an endless loop.
> 
> This patch detects the loop, prints a warning, and
> continues on with useful work.  This strategy was
> suggested by Robert Hancock <hancockrwd@gmail.com>
> 
> This should be a candidate for 2.6.34.y as well.
> 
> Signed-off-by: Ben Greear <greearb@candelatech.com>
> ---
> :100644 100644 1cdc02c... 9535ba9... M	arch/x86/pci/mrst.c
>  arch/x86/pci/mrst.c |   11 +++++++++++
>  1 files changed, 11 insertions(+), 0 deletions(-)
> 
> diff --git a/arch/x86/pci/mrst.c b/arch/x86/pci/mrst.c
> index 1cdc02c..9535ba9 100644
> --- a/arch/x86/pci/mrst.c
> +++ b/arch/x86/pci/mrst.c
> @@ -76,6 +76,17 @@ static int fixed_bar_cap(struct pci_bus *bus, unsigned int devfn)
>  				return pos;
>  		}
>  
> +		if ((pcie_cap >> 20) == 0)
> +			break;
> +		
> +		if ((pcie_cap >> 20) <= pos) {
> +			printk(KERN_WARNING "WARNING: mrst:  detected loop"
> +			       " when searching for fixed BAR cap, previous"
> +			       " position: 0x%x new position: 0x%x"
> +			       " bus-number: %i  devfn: %i\n",
> +			       pos, pcie_cap >> 20, bus->number, devfn);

Can you use dev_warn() here to print the device info in the
standard way?

Bjorn

Re: [pci] pci/mrst: Detect loops when reading fixed BAR cap.

[Ben Greear]

On 07/16/2010 10:43 AM, Bjorn Helgaas wrote:
> On Wednesday, July 14, 2010 11:02:03 am Ben Greear wrote:
>> The mrst logic introduced in 2.6.33-rc8 in commit
>> a712ffbc199849364c46e9112b93b66de08e2c26 causes boot
>> to hang on at least this platform:
>>
>> Intel E5405 CPU
>> System Information
>> 	Manufacturer: Supermicro
>> 	Product Name: X7DBU
>>
>> The cause of the hang is that pci_cap points to itself
>> as the next capability, putting the fixed_bar_cap into
>> an endless loop.
>>
>> This patch detects the loop, prints a warning, and
>> continues on with useful work.  This strategy was
>> suggested by Robert Hancock<hancockrwd@gmail.com>
>>
>> This should be a candidate for 2.6.34.y as well.
>>
>> Signed-off-by: Ben Greear<greearb@candelatech.com>
>> ---
>> :100644 100644 1cdc02c... 9535ba9... M	arch/x86/pci/mrst.c
>>   arch/x86/pci/mrst.c |   11 +++++++++++
>>   1 files changed, 11 insertions(+), 0 deletions(-)
>>
>> diff --git a/arch/x86/pci/mrst.c b/arch/x86/pci/mrst.c
>> index 1cdc02c..9535ba9 100644
>> --- a/arch/x86/pci/mrst.c
>> +++ b/arch/x86/pci/mrst.c
>> @@ -76,6 +76,17 @@ static int fixed_bar_cap(struct pci_bus *bus, unsigned int devfn)
>>   				return pos;
>>   		}
>>
>> +		if ((pcie_cap>>  20) == 0)
>> +			break;
>> +		
>> +		if ((pcie_cap>>  20)<= pos) {
>> +			printk(KERN_WARNING "WARNING: mrst:  detected loop"
>> +			       " when searching for fixed BAR cap, previous"
>> +			       " position: 0x%x new position: 0x%x"
>> +			       " bus-number: %i  devfn: %i\n",
>> +			       pos, pcie_cap>>  20, bus->number, devfn);
>
> Can you use dev_warn() here to print the device info in the
> standard way?

There seems to be a different way to do this.  Some of the folks who
understand this better are working on a patch (I tested something based
on HPA's suggestions and it fixed my problem at least.)

Thanks,
Ben

-- 
Ben Greear <greearb@candelatech.com>
Candela Technologies Inc  http://www.candelatech.com