atl1c: DMA: Out of SW-IOMMU space

atl1c: DMA: Out of SW-IOMMU space

[Andre Heider]

Hi,

my "AR8151 v2.0 Gigabit Ethernet (rev c0)" becomes unusable on v3.8
after some usage.

It's easily reproducible with "dd if=/dev/zero of=<nfsshare>/test
bs=1M count=2048. Point of failure varies, but somewhat after >1.5gb
of data written to the nfs server dmesg gets spammed with "DMA: Out of
SW-IOMMU space for x bytes at device 0000:06:00.0" (where x varies)
and all sorts of follow up problems start occurring.
Apparently I'm not the only one affected:
https://www.google.com/search?q=atl1c+%22DMA%3A+Out+of+SW-IOMMU+space%22

The problem doesn't exist on v3.8-rc7.

The only atl1c commit in v3.8-rc7..v3.8 is:

commit ac574804d412047e52de4dd887615cc88f58aeb0
Author: Huang, Xiong <xiong@qca.qualcomm.com>
Date:   Mon Feb 11 14:44:40 2013 +0000

    atl1c: add error checking for pci_map_single functions

    it is reported that code hit DMA-API errors on 3.8-rc6+,
    (see https://bugzilla.redhat.com/show_bug.cgi?id=908436, and
         https://bugzilla.redhat.com/show_bug.cgi?id=908550)

    this patch just adds error handler for
        pci_map_single and skb_frag_dma_map.

    Signed-off-by: xiong <xiong@qca.qualcomm.com>
    Signed-off-by: David S. Miller <davem@davemloft.net>

Reverting that on top of v3.8 seems to fix the problem in my little
testing, so it sounds like that fix introduced a iommu leak?

Thanks,
Andre

RE: atl1c: DMA: Out of SW-IOMMU space

[Huang, Xiong]

There is another patch follows commit ac574804d412047e52de4dd887615cc88f58aeb0, would you have a try ?

Thanks
Xiong 

> -----Original Message-----
> From: Andre Heider [mailto:a.heider@gmail.com]
> Sent: Sunday, March 03, 2013 1:29 AM
> To: netdev@vger.kernel.org; Huang, Xiong
> Subject: atl1c: DMA: Out of SW-IOMMU space
> 
> The problem doesn't exist on v3.8-rc7.
> 
> The only atl1c commit in v3.8-rc7..v3.8 is:
> 
> commit ac574804d412047e52de4dd887615cc88f58aeb0
> Author: Huang, Xiong <xiong@qca.qualcomm.com>
> Date:   Mon Feb 11 14:44:40 2013 +0000
> 
>     atl1c: add error checking for pci_map_single functions
> 
>     it is reported that code hit DMA-API errors on 3.8-rc6+,
>     (see https://bugzilla.redhat.com/show_bug.cgi?id=908436, and
>          https://bugzilla.redhat.com/show_bug.cgi?id=908550)
> 
>     this patch just adds error handler for
>         pci_map_single and skb_frag_dma_map.
> 
>     Signed-off-by: xiong <xiong@qca.qualcomm.com>
>     Signed-off-by: David S. Miller <davem@davemloft.net>
> 
> Reverting that on top of v3.8 seems to fix the problem in my little testing, so
> it sounds like that fix introduced a iommu leak?
> 
> Thanks,
> Andre

RE: atl1c: DMA: Out of SW-IOMMU space

[Eric Dumazet]

On Sun, 2013-03-03 at 01:48 +0000, Huang, Xiong wrote:
> There is another patch follows commit ac574804d412047e52de4dd887615cc88f58aeb0, would you have a try ?


Do you mean the following ?


BTW, f1f220ea1dda078 doesnt exist


commit 7cb08d7f3a5ea6131f4f243c2080530ac41cb293
Author: Huang, Xiong <xiong@qca.qualcomm.com>
Date:   Tue Feb 19 07:23:09 2013 +0000

    atl1c: restore buffer state
    
    in the previous commit : f1f220ea1dda078, the BUSY state of buffer is wrongly
    deleted. this patch just restore it.
    
    Signed-off-by: xiong <xiong@qca.qualcomm.com>
    Signed-off-by: David S. Miller <davem@davemloft.net>

Re: atl1c: DMA: Out of SW-IOMMU space

[Andre Heider]

On Sun, Mar 3, 2013 at 3:52 AM, Eric Dumazet <eric.dumazet@gmail.com> wrote:
> On Sun, 2013-03-03 at 01:48 +0000, Huang, Xiong wrote:
>> There is another patch follows commit ac574804d412047e52de4dd887615cc88f58aeb0, would you have a try ?
>
>
> Do you mean the following ?
>
> commit 7cb08d7f3a5ea6131f4f243c2080530ac41cb293
> Author: Huang, Xiong <xiong@qca.qualcomm.com>
> Date:   Tue Feb 19 07:23:09 2013 +0000
>
>     atl1c: restore buffer state
>
>     in the previous commit : f1f220ea1dda078, the BUSY state of buffer is wrongly
>     deleted. this patch just restore it.
>
>     Signed-off-by: xiong <xiong@qca.qualcomm.com>
>     Signed-off-by: David S. Miller <davem@davemloft.net>

Yeah, that's the only relevant commit I could find too, and indeed
that patch fixes the problem, thanks!

Can we please apply this to 3.8.y ? atl1c is broken on 3.8 and 3.8.1,
<= 3.7 is fine.

Thanks,
Andre

Re: atl1c: DMA: Out of SW-IOMMU space

[David Miller]

From: Andre Heider <a.heider@gmail.com>
Date: Sun, 3 Mar 2013 04:21:52 +0100

> On Sun, Mar 3, 2013 at 3:52 AM, Eric Dumazet <eric.dumazet@gmail.com> wrote:
>> On Sun, 2013-03-03 at 01:48 +0000, Huang, Xiong wrote:
>>> There is another patch follows commit ac574804d412047e52de4dd887615cc88f58aeb0, would you have a try ?
>>
>>
>> Do you mean the following ?
>>
>> commit 7cb08d7f3a5ea6131f4f243c2080530ac41cb293
>> Author: Huang, Xiong <xiong@qca.qualcomm.com>
>> Date:   Tue Feb 19 07:23:09 2013 +0000
>>
>>     atl1c: restore buffer state
>>
>>     in the previous commit : f1f220ea1dda078, the BUSY state of buffer is wrongly
>>     deleted. this patch just restore it.
>>
>>     Signed-off-by: xiong <xiong@qca.qualcomm.com>
>>     Signed-off-by: David S. Miller <davem@davemloft.net>
> 
> Yeah, that's the only relevant commit I could find too, and indeed
> that patch fixes the problem, thanks!
> 
> Can we please apply this to 3.8.y ? atl1c is broken on 3.8 and 3.8.1,
> <= 3.7 is fine.

I've queued this up for -stable submission, thanks.