Ethernet bug at Upstream-v3.15-rcx on Lager.

Ethernet bug at Upstream-v3.15-rcx on Lager.

[カオ ミン ヒェップ]

Hi Sergei

I am Hiep from JINZAI SOLUTION (a Japanese Company).
Now, I am testing a upstream-v3.15-rc2 on Lager for Renesas RSO.
I found a problem about Ethernet Driver so I would like to report to you.
The problem is losing data when sending data by UDP:

     - From Lager board to PC:
     BANDWIDTH from 10M to 70M are good,  but they suddenly  down 51.2 
Mbits/sec at 80M
     and 57.8 Mbits/sec at 90 M and 64.2 Mbits/sec at 100M.

     - From PC to Lager board:
    Some time when the message of "sh-eth r8a7790-ether eth0: Receive 
FIFO Overflow" occurs,
    some of datagrams will be lost as the logs show below:
     「$ iperf -c 172.16.1.36 -u -b 90M
     .....
     [  3] Server Report:
     [  3]  0.0-10.0 sec   108 MBytes  90.4 Mbits/sec   0.000 ms 
36/76923 (0.047%)
     .....
     # iperf -s -u
     [ ID] Interval       Transfer     Bandwidth        Jitter 
Lost/Total Datagrams
     sh-eth r8a7790-ether eth0: Receive FIFO Overflow
     [  3]  0.0-10.0 sec   108 MBytes  90.4 Mbits/sec   0.001 ms 
36/76923 (0.047%)

Please review it when you have time.

Bests Regards,
Cao Minh Hiep.

Re: Ethernet bug at Upstream-v3.15-rcx on Lager.

[Sergei Shtylyov]

Hello.

On 05/15/2014 10:28 AM, カオ　ミン　ヒェップ wrote:

> I am Hiep from JINZAI SOLUTION (a Japanese Company).
> Now, I am testing a upstream-v3.15-rc2 on Lager for Renesas RSO.
> I found a problem about Ethernet Driver so I would like to report to you.
> The problem is losing data when sending data by UDP:

>      - From Lager board to PC:
>      BANDWIDTH from 10M to 70M are good,  but they suddenly  down 51.2
> Mbits/sec at 80M
>      and 57.8 Mbits/sec at 90 M and 64.2 Mbits/sec at 100M.

    Do you mean 'iperf' tests here?

>      - From PC to Lager board:
>     Some time when the message of "sh-eth r8a7790-ether eth0: Receive FIFO
> Overflow" occurs,
>     some of datagrams will be lost as the logs show below:
>      「$ iperf -c 172.16.1.36 -u -b 90M
>      .....
>      [  3] Server Report:
>      [  3]  0.0-10.0 sec   108 MBytes  90.4 Mbits/sec   0.000 ms 36/76923
> (0.047%)
>      .....
>      # iperf -s -u
>      [ ID] Interval       Transfer     Bandwidth        Jitter Lost/Total
> Datagrams
>      sh-eth r8a7790-ether eth0: Receive FIFO Overflow
>      [  3]  0.0-10.0 sec   108 MBytes  90.4 Mbits/sec   0.001 ms 36/76923
> (0.047%)

> Please review it when you have time.

    Please try this patch:

http://marc.info/?l=linux-netdev&m=139602519809181

> Bests Regards,
> Cao Minh Hiep.

WBR, Sergei

Re: Ethernet bug at Upstream-v3.15-rcx on Lager.

[カオ ミン ヒェップ]

Hi Sergei

Thanks for your replying!

On 05/16/2014 02:20 AM, Sergei Shtylyov wrote:
> Hello.
>
> On 05/15/2014 10:28 AM, カオ　ミン　ヒェップ wrote:
>
>> I am Hiep from JINZAI SOLUTION (a Japanese Company).
>> Now, I am testing a upstream-v3.15-rc2 on Lager for Renesas RSO.
>> I found a problem about Ethernet Driver so I would like to report to 
>> you.
>> The problem is losing data when sending data by UDP:
>
>>      - From Lager board to PC:
>>      BANDWIDTH from 10M to 70M are good,  but they suddenly down 51.2
>> Mbits/sec at 80M
>>      and 57.8 Mbits/sec at 90 M and 64.2 Mbits/sec at 100M.
>
>    Do you mean 'iperf' tests here?
Yes, I used 'iperf' tool to test here.
And the results of test these test-cases on upstream-v3.15-rc5 as show 
below:
79.9 Mbits/sec at 80M, 81.5 Mbits/sec at 90 M and 75.1 Mbits/sec at 100M.

>
>>      - From PC to Lager board:
>>     Some time when the message of "sh-eth r8a7790-ether eth0: Receive 
>> FIFO
>> Overflow" occurs,
>>     some of datagrams will be lost as the logs show below:
>>      「$ iperf -c 172.16.1.36 -u -b 90M
>>      .....
>>      [  3] Server Report:
>>      [  3]  0.0-10.0 sec   108 MBytes  90.4 Mbits/sec   0.000 ms 
>> 36/76923
>> (0.047%)
>>      .....
>>      # iperf -s -u
>>      [ ID] Interval       Transfer     Bandwidth        Jitter 
>> Lost/Total
>> Datagrams
>>      sh-eth r8a7790-ether eth0: Receive FIFO Overflow
>>      [  3]  0.0-10.0 sec   108 MBytes  90.4 Mbits/sec   0.001 ms 
>> 36/76923
>> (0.047%)
>
>> Please review it when you have time.
>
>    Please try this patch:
>
> http://marc.info/?l=linux-netdev&m=139602519809181
>
I have tested this patch on v3.15.0-rc5-00077-g14186fe, and results are 
the message of
"sh-eth r8a7790-ether eth0: Receive FIFO Overflow" has disappeared and 
sending datagrams is good.
(Bandwidth 10M to 100M: Test with 5/5 times by script and by hand are good)

Bests Regards.
Hiep.

Re: Ethernet bug at Upstream-v3.15-rcx on Lager.

[Yoshihiro Shimoda]

Hi Hiep-san,

(2014/05/16 10:08), カオ　ミン　ヒェップ wrote:
>>
>>> Please review it when you have time.
>>
>>    Please try this patch:
>>
>> http://marc.info/?l=linux-netdev&m=139602519809181
>>
> I have tested this patch on v3.15.0-rc5-00077-g14186fe, and results are the message of
> "sh-eth r8a7790-ether eth0: Receive FIFO Overflow" has disappeared and sending datagrams is good.
> (Bandwidth 10M to 100M: Test with 5/5 times by script and by hand are good)
> 

It's nice!
So, I hope the patch is merged in upstream.
However, the patch's change log mentioned r8a7790 only.
The r8a7791's ethernet controller also has this bit.

To Ben,

So, would you update your patch and resubmit it?

Best regards,
Yoshihiro Shimoda

Re: Ethernet bug at Upstream-v3.15-rcx on Lager.

[Ben Dooks]

On 16/05/14 06:56, Yoshihiro Shimoda wrote:
> Hi Hiep-san,
> 
> (2014/05/16 10:08), カオ　ミン　ヒェップ wrote:
>>>
>>>> Please review it when you have time.
>>>
>>>    Please try this patch:
>>>
>>> http://marc.info/?l=linux-netdev&m=139602519809181
>>>
>> I have tested this patch on v3.15.0-rc5-00077-g14186fe, and results are the message of
>> "sh-eth r8a7790-ether eth0: Receive FIFO Overflow" has disappeared and sending datagrams is good.
>> (Bandwidth 10M to 100M: Test with 5/5 times by script and by hand are good)
>>
> 
> It's nice!
> So, I hope the patch is merged in upstream.
> However, the patch's change log mentioned r8a7790 only.
> The r8a7791's ethernet controller also has this bit.

The bit update covers both of them as the r8a7790 and r8a7791 both share
the same configuration structure.

I re-sent the patch but didn't find your email post until after I had
sent it, otherwise I would have mentioned your testing specifically. In
future it would be helpful if you could do the following

- Mention driver by name in the subject line
- Search the list for previous mentions and reply to that/those threads.


-- 
Ben Dooks				http://www.codethink.co.uk/
Senior Engineer				Codethink - Providing Genius

Re: Ethernet bug at Upstream-v3.15-rcx on Lager.

[Yoshihiro Shimoda]

Hi Ben,

(2014/05/30 23:04), Ben Dooks wrote:
> On 16/05/14 06:56, Yoshihiro Shimoda wrote:
>> Hi Hiep-san,
>>
>> (2014/05/16 10:08), カオ　ミン　ヒェップ wrote:
>>>>
>>>>> Please review it when you have time.
>>>>
>>>>    Please try this patch:
>>>>
>>>> http://marc.info/?l=linux-netdev&m=139602519809181
>>>>
>>> I have tested this patch on v3.15.0-rc5-00077-g14186fe, and results are the message of
>>> "sh-eth r8a7790-ether eth0: Receive FIFO Overflow" has disappeared and sending datagrams is good.
>>> (Bandwidth 10M to 100M: Test with 5/5 times by script and by hand are good)
>>>
>>
>> It's nice!
>> So, I hope the patch is merged in upstream.
>> However, the patch's change log mentioned r8a7790 only.
>> The r8a7791's ethernet controller also has this bit.
> 
> The bit update covers both of them as the r8a7790 and r8a7791 both share
> the same configuration structure.
> 
> I re-sent the patch but didn't find your email post until after I had
> sent it, otherwise I would have mentioned your testing specifically. In
> future it would be helpful if you could do the following
> 
> - Mention driver by name in the subject line
> - Search the list for previous mentions and reply to that/those threads.
> 

Thank you very much for the comment and your v2 patch!
I understood it.

Best regards,
Yoshihiro Shimoda