Re: r8169 and out of space

Re: r8169 and out of space

[Alistair John Strachan]

Bugzilla for this report: http://bugzilla.kernel.org/show_bug.cgi?id=9468

On Thursday 10 January 2008 17:28:22 you wrote:
>   I saw your thread on LKML about the problem with r8169. Did you ever
> find a patch or solution?

No, we have not diagnosed the cause of the problem, beyond the swiotlb usage. 
I'm adding the r8169 maintainer, linux-net and linux-kernel to CC, to pass on 
your information, I hope you don't mind.

>   I have a Abit AB9 Pro motherboard with Intel P965 chipset, 4gb of
> memory, and two onboard r8169. The kernel is
> kernel-2.6.23.9-85.fc8.x86_64. I am using jumbo frames, the MTU set at
> 5924. The error is mentions 5946 bytes. 5946 - 5924 is 22. The same as
> your 7200 MTU subtracted from your bytes, 7222.
>
>   I can set my MTU to 7200, and I can ping with 60000 byte packets just
> fine, but tcp connections hang unless I set the MTU to 5924 or lower.
> The other side is a Nvidia based board with a forcedeth driver that has
> in the past taken 9000 just fine.

-- 
Cheers,
Alistair.

137/1 Warrender Park Road, Edinburgh, UK.

Re: r8169 and out of space

[Francois Romieu]

Alistair John Strachan <alistair@devzero.co.uk> :
[..]
> No, we have not diagnosed the cause of the problem, beyond the swiotlb usage. 
> I'm adding the r8169 maintainer, linux-net and linux-kernel to CC, to pass on 
> your information, I hope you don't mind.

swiotlb fragmentation perhaps ?

Switching the driver to be page based on Rx would help then.

[...]
> >   I have a Abit AB9 Pro motherboard with Intel P965 chipset, 4gb of
> > memory, and two onboard r8169.

Which r8169 model exactly ?

-- 
Ueimor

Re: r8169 and out of space

[Nathan Grennan]

Francois Romieu wrote:
> Alistair John Strachan <alistair@devzero.co.uk> :
> [..]
>   
>> No, we have not diagnosed the cause of the problem, beyond the swiotlb usage. 
>> I'm adding the r8169 maintainer, linux-net and linux-kernel to CC, to pass on 
>> your information, I hope you don't mind.
>>     
>
> swiotlb fragmentation perhaps ?
>
> Switching the driver to be page based on Rx would help then.
>
> [...]
>   
>>>   I have a Abit AB9 Pro motherboard with Intel P965 chipset, 4gb of
>>> memory, and two onboard r8169.
>>>       
>
> Which r8169 model exactly ?
>
>   
Here is some of the output from lspci -vv.

03:00.0 Ethernet controller: Realtek Semiconductor Co., Ltd. 
RTL8111/8168B PCI Express Gigabit Ethernet controller (rev 01)
        Subsystem: ABIT Computer Corp. Unknown device 240b
05:00.0 Ethernet controller: Realtek Semiconductor Co., Ltd. 
RTL8111/8168B PCI Express Gigabit Ethernet controller (rev 01)
        Subsystem: ABIT Computer Corp. Unknown device 240b


I think I fixed the the 5924 vs 7200 issue by upgrading to 
kernel-2.6.23.13-104.fc8.x86_64.

Re: r8169 and out of space

[Nathan Grennan]

Francois Romieu wrote:
> Alistair John Strachan <alistair@devzero.co.uk> :
> [..]
>   
>> No, we have not diagnosed the cause of the problem, beyond the swiotlb usage. 
>> I'm adding the r8169 maintainer, linux-net and linux-kernel to CC, to pass on 
>> your information, I hope you don't mind.
>>     
>
> swiotlb fragmentation perhaps ?
>
> Switching the driver to be page based on Rx would help then.
>
> [...]
>   
>>>   I have a Abit AB9 Pro motherboard with Intel P965 chipset, 4gb of
>>> memory, and two onboard r8169.
>>>       
>
> Which r8169 model exactly ?
>
>   
  I seem to have been wrong about the kernel update fixing my mtu issue. 
I am not 100% sure, but I think 7200 works if the forcedeth system is 
sending, but not if the r8169 system is sending.

  I tried a mtu of 3000, and basically the same errors, but I did 
noticed something interesting. Note the three different byte sizes.

DMA: Out of SW-IOMMU space for 2410 bytes at device 0000:03:00.0
DMA: Out of SW-IOMMU space for 3014 bytes at device 0000:03:00.0
DMA: Out of SW-IOMMU space for 3014 bytes at device 0000:03:00.0
DMA: Out of SW-IOMMU space for 3022 bytes at device 0000:03:00.0


  On a side note, leaving the mtu at 1500 seems to work without error, 
and I get about the same performance.