[Xenomai] rt_e1000e driver issues

[Xenomai] rt_e1000e driver issues

[Stéphane ANCELOT]

Hi,

I am checking timing issues of realtime rtnet frames in my application.
Timing to send a frame is  convenient.
but I have some issues with the time to receive the frames.

Although the RxDelays are cleared, the frame is received almost 175us 
later than it should in comparison with a rt8169 driver (v2.6 version) 
and the same application.

Any idea where to look at first ?
Regards,
Steph

Re: [Xenomai] rt_e1000e driver issues

[Gilles Chanteperdrix]

On Mon, Aug 31, 2015 at 04:04:18PM +0200, Stéphane ANCELOT wrote:
> Hi,
> 
> I am checking timing issues of realtime rtnet frames in my application.
> Timing to send a frame is  convenient.
> but I have some issues with the time to receive the frames.
> 
> Although the RxDelays are cleared, the frame is received almost 175us later
> than it should in comparison with a rt8169 driver (v2.6 version) and the
> same application.
> 
> Any idea where to look at first ?

modinfo rt_e1000e

-- 
					    Gilles.
https://click-hack.org

Re: [Xenomai] rt_e1000e driver issues

[Gilles Chanteperdrix]

On Mon, Aug 31, 2015 at 08:30:03PM +0200, Gilles Chanteperdrix wrote:
> On Mon, Aug 31, 2015 at 04:04:18PM +0200, Stéphane ANCELOT wrote:
> > Hi,
> > 
> > I am checking timing issues of realtime rtnet frames in my application.
> > Timing to send a frame is  convenient.
> > but I have some issues with the time to receive the frames.
> > 
> > Although the RxDelays are cleared, the frame is received almost 175us later
> > than it should in comparison with a rt8169 driver (v2.6 version) and the
> > same application.
> > 
> > Any idea where to look at first ?
> 
> modinfo rt_e1000e

Actually, the module parameters are the same as the Linux driver,
so, they are documented in Documentation/networking/e1000e.txt in
the Linux kernel sources, here for instance:
https://git.xenomai.org/ipipe.git/tree/Documentation/networking/e1000e.txt?id=v4.2-rc1
(beware that the driver versions are not the same, so, there may be
some differences, like default values, or even some options, but
modinfo will tell you that).

Your problem is probably that you should disable interrupt
throttling passing InterruptThrottleRate=0.

-- 
					    Gilles.
https://click-hack.org

Re: [Xenomai] rt_e1000e driver issues

[Stéphane ANCELOT]

On 01/09/2015 12:20, Gilles Chanteperdrix wrote:
> On Mon, Aug 31, 2015 at 08:30:03PM +0200, Gilles Chanteperdrix wrote:
>> On Mon, Aug 31, 2015 at 04:04:18PM +0200, Stéphane ANCELOT wrote:
>>> Hi,
>>>
>>> I am checking timing issues of realtime rtnet frames in my application.
>>> Timing to send a frame is  convenient.
>>> but I have some issues with the time to receive the frames.
>>>
>>> Although the RxDelays are cleared, the frame is received almost 175us later
>>> than it should in comparison with a rt8169 driver (v2.6 version) and the
>>> same application.
>>>
>>> Any idea where to look at first ?
>> modinfo rt_e1000e
> Actually, the module parameters are the same as the Linux driver,
> so, they are documented in Documentation/networking/e1000e.txt in
> the Linux kernel sources, here for instance:
> https://git.xenomai.org/ipipe.git/tree/Documentation/networking/e1000e.txt?id=v4.2-rc1
> (beware that the driver versions are not the same, so, there may be
> some differences, like default values, or even some options, but
> modinfo will tell you that).
>
> Your problem is probably that you should disable interrupt
> throttling passing InterruptThrottleRate=0.
>
Hi,
I checked the timerdelays settings everything seems rigth by default.
However,  I think the main differences I mesure are related to the usage 
of EarlyReceive in r8169 driver. Thus spending less time in recv func .

I will soon check the same application using the igb driver

Regards
S.Ancelot

Re: [Xenomai] rt_e1000e driver issues

[Gilles Chanteperdrix]

On 09/02/2015 04:00 PM, Stéphane ANCELOT wrote:
> On 01/09/2015 12:20, Gilles Chanteperdrix wrote:
>> On Mon, Aug 31, 2015 at 08:30:03PM +0200, Gilles Chanteperdrix wrote:
>>> On Mon, Aug 31, 2015 at 04:04:18PM +0200, Stéphane ANCELOT wrote:
>>>> Hi,
>>>>
>>>> I am checking timing issues of realtime rtnet frames in my 
>>>> application.
>>>> Timing to send a frame is  convenient.
>>>> but I have some issues with the time to receive the frames.
>>>>
>>>> Although the RxDelays are cleared, the frame is received almost 
>>>> 175us later
>>>> than it should in comparison with a rt8169 driver (v2.6 version) 
>>>> and the
>>>> same application.
>>>>
>>>> Any idea where to look at first ?
>>> modinfo rt_e1000e
>> Actually, the module parameters are the same as the Linux driver,
>> so, they are documented in Documentation/networking/e1000e.txt in
>> the Linux kernel sources, here for instance:
>> https://git.xenomai.org/ipipe.git/tree/Documentation/networking/e1000e.txt?id=v4.2-rc1 
>>
>> (beware that the driver versions are not the same, so, there may be
>> some differences, like default values, or even some options, but
>> modinfo will tell you that).
>>
>> Your problem is probably that you should disable interrupt
>> throttling passing InterruptThrottleRate=0.
>>
> Hi,
> I checked the timerdelays settings everything seems rigth by default.
> However,  I think the main differences I mesure are related to the 
> usage of EarlyReceive in r8169 driver. Thus spending less time in recv 
> func .
>
> I will soon check the same application using the igb driver

I have made some round trip time measurement with two e1000e boards, and 
I am pretty sure
with InterruptThrottleRate=0 *RxDelay=0 the delays are in the 10s of 
microseconds order (on a
gigabit link).

--
                                                 Gilles Chanteperdrix