lwIP

lwIP

[Frank J. Beckmann]

Hello,

are there plans for updating eCos' lwIP port to a more recent version?
-- 
Bye
Frank

Re: lwIP

[Simon Kallweit]

Frank J. Beckmann wrote:
> Hello,
>
> are there plans for updating eCos' lwIP port to a more recent version?

I was planning on doing that, if it's not already planned for the ecos 
3.0 release. I don't know if eCosCentric has a more up-to-date port. I'm 
currently working on a GSM/GPRS service for our STM32 based boards, 
which has to be quite memory efficient. I'm planning on using lwIP, and 
if there are features not currently available in the port, I guess I'm 
forced to do a port of a more up-to-date release.

Simon

Re: lwIP

[Simon Kallweit]

Simon Kallweit wrote:
> Frank J. Beckmann wrote:
>> Hello,
>>
>> are there plans for updating eCos' lwIP port to a more recent version?
>
> I was planning on doing that, if it's not already planned for the ecos 
> 3.0 release. I don't know if eCosCentric has a more up-to-date port. 
> I'm currently working on a GSM/GPRS service for our STM32 based 
> boards, which has to be quite memory efficient. I'm planning on using 
> lwIP, and if there are features not currently available in the port, I 
> guess I'm forced to do a port of a more up-to-date release.

I just had a quick look of the lwIP in the current ECOS version. It 
differs just very slightly from the actual lwIP 1.1.1 release, not many 
changes. I guess an update to a more recent lwIP version should be 
pretty simple and straight forward. I'll start working on it in around a 
week, have to get my GSM modem working first :)

Simon

Re: lwIP

[Jonathan Larmour]

Simon Kallweit wrote:
> Frank J. Beckmann wrote:
>> Hello,
>>
>> are there plans for updating eCos' lwIP port to a more recent version?
> 
> I was planning on doing that, if it's not already planned for the ecos
> 3.0 release. I don't know if eCosCentric has a more up-to-date port.

Sort of. Ours starts from the same original lwIP code base (1.1.0) but I
had to pretty much rewrite Jani's original eCos port as it had a number of
design flaws. It is essentially doomed under load.

But we don't have any plans to release our own port of it, so if you want
an update in the near future, it's not terribly likely to come from
eCosCentric I'm afraid.

Jifl
-- 
eCosCentric Limited      http://www.eCosCentric.com/     The eCos experts
Barnwell House, Barnwell Drive, Cambridge, UK.       Tel: +44 1223 245571
Registered in England and Wales: Reg No 4422071.
------["Si fractum non sit, noli id reficere"]------       Opinions==mine

Re: lwIP

[Jonathan Larmour]

Jonathan Larmour wrote:
> Simon Kallweit wrote:
>> Frank J. Beckmann wrote:
>>> Hello,
>>>
>>> are there plans for updating eCos' lwIP port to a more recent version?
>> I was planning on doing that, if it's not already planned for the ecos
>> 3.0 release. I don't know if eCosCentric has a more up-to-date port.
> 
> Sort of. Ours starts from the same original lwIP code base (1.1.0)

Ahem, 1.1.1 I mean.

Jifl
-- 
eCosCentric Limited      http://www.eCosCentric.com/     The eCos experts
Barnwell House, Barnwell Drive, Cambridge, UK.       Tel: +44 1223 245571
Registered in England and Wales: Reg No 4422071.
------["Si fractum non sit, noli id reficere"]------       Opinions==mine

Re: lwIP

[Frank Pagliughi]

Simon Kallweit wrote:
> Simon Kallweit wrote:
>> Frank J. Beckmann wrote:
>>> Hello,
>>>
>>> are there plans for updating eCos' lwIP port to a more recent version?
>>
>> I was planning on doing that, if it's not already planned for the 
>> ecos 3.0 release. I don't know if eCosCentric has a more up-to-date 
>> port. I'm currently working on a GSM/GPRS service for our STM32 based 
>> boards, which has to be quite memory efficient. I'm planning on using 
>> lwIP, and if there are features not currently available in the port, 
>> I guess I'm forced to do a port of a more up-to-date release.
>
> I just had a quick look of the lwIP in the current ECOS version. It 
> differs just very slightly from the actual lwIP 1.1.1 release, not 
> many changes. I guess an update to a more recent lwIP version should 
> be pretty simple and straight forward. I'll start working on it in 
> around a week, have to get my GSM modem working first :)
>
> Simon
>
>

Excellent!  If I'm not mistaken, the ppp code in the current eCos 
version is broken. An update would be welcome.

Frank

Re: lwIP

[John Eigelaar]

On Fri, 2008-11-21 at 14:49 -0500, Frank Pagliughi wrote:
> Simon Kallweit wrote:
> > Simon Kallweit wrote:
> >> Frank J. Beckmann wrote:
> >>> Hello,
> >>>
> >>> are there plans for updating eCos' lwIP port to a more recent version?
> >>
> >
> > I just had a quick look of the lwIP in the current ECOS version. It 
> > differs just very slightly from the actual lwIP 1.1.1 release, not 
> > many changes. I guess an update to a more recent lwIP version should 
> > be pretty simple and straight forward. I'll start working on it in 
> > around a week, have to get my GSM modem working first :)
> >
I have been trying on and off tp port lwip 1.3.0 to ecos. The current
version has a serious memory leak when under fire and also does not
honor the backlog parameter for lwip_listen and lwip_accept mechanisms.

In 1.3.0 the init part of the lwIP source has been greatly improved and
now works well to initialise the lwIP stack but is of course now
completely incompatible with the existing eCos port. The sys_arch API
has been update as well to counetr the aforementioned memory leaks.

I am also not sure what to do really with the IPv6 and IPv4 config
options for the lwIP stack.

I have compiled the test cases with the new 1.3.0 source but I  still
can not get any of them to run on the Linux synth target. Some of it is
actualy me learning the synth target as I go along as well as me not
really having any time to spend on this other than time set aside out of
curiosity.

John Eigelaar

Re: lwIP

[John Dallaway]

Hi John and Simon

John Eigelaar wrote:

> On Fri, 2008-11-21 at 14:49 -0500, Frank Pagliughi wrote:
>> Simon Kallweit wrote:
>>> Simon Kallweit wrote:
>>>> Frank J. Beckmann wrote:
>>>>> Hello,
>>>>>
>>>>> are there plans for updating eCos' lwIP port to a more recent version?
>>> I just had a quick look of the lwIP in the current ECOS version. It 
>>> differs just very slightly from the actual lwIP 1.1.1 release, not 
>>> many changes. I guess an update to a more recent lwIP version should 
>>> be pretty simple and straight forward. I'll start working on it in 
>>> around a week, have to get my GSM modem working first :)
>>>
> I have been trying on and off tp port lwip 1.3.0 to ecos. The current
> version has a serious memory leak when under fire and also does not
> honor the backlog parameter for lwip_listen and lwip_accept mechanisms.
> 
> In 1.3.0 the init part of the lwIP source has been greatly improved and
> now works well to initialise the lwIP stack but is of course now
> completely incompatible with the existing eCos port. The sys_arch API
> has been update as well to counetr the aforementioned memory leaks.
> 
> I am also not sure what to do really with the IPv6 and IPv4 config
> options for the lwIP stack.
> 
> I have compiled the test cases with the new 1.3.0 source but I  still
> can not get any of them to run on the Linux synth target. Some of it is
> actualy me learning the synth target as I go along as well as me not
> really having any time to spend on this other than time set aside out of
> curiosity.

I am keen to see lwIP updated in the eCos repository and can offer help
with the CDL if necessary.

It looks like good progress has been made with lwIP 1.3.0 on eCos
already. Could you drop a note to this list with the current status of
your work please and we can plan accordingly?

Given the changes in the lwIP initialisation code, can anyone comment on
the likely effort to port an eCos application based on the existing eCos
lwIP stack to lwIP 1.3.0?

lwIP is a good fit for eCos and I can see the use of this stack
dominating in the future. With this in mind, it would be great to
support as many features of the stack as possible (including IPv6).

John Dallaway

Re: lwIP

[Simon Kallweit]

John Dallaway wrote:
> Hi John and Simon
>
> John Eigelaar wrote:
>
>   
>> On Fri, 2008-11-21 at 14:49 -0500, Frank Pagliughi wrote:
>>     
>>> Simon Kallweit wrote:
>>>       
>>>> Simon Kallweit wrote:
>>>>         
>>>>> Frank J. Beckmann wrote:
>>>>>           
>>>>>> Hello,
>>>>>>
>>>>>> are there plans for updating eCos' lwIP port to a more recent version?
>>>>>>             
>>>> I just had a quick look of the lwIP in the current ECOS version. It 
>>>> differs just very slightly from the actual lwIP 1.1.1 release, not 
>>>> many changes. I guess an update to a more recent lwIP version should 
>>>> be pretty simple and straight forward. I'll start working on it in 
>>>> around a week, have to get my GSM modem working first :)
>>>>
>>>>         
>> I have been trying on and off tp port lwip 1.3.0 to ecos. The current
>> version has a serious memory leak when under fire and also does not
>> honor the backlog parameter for lwip_listen and lwip_accept mechanisms.
>>
>> In 1.3.0 the init part of the lwIP source has been greatly improved and
>> now works well to initialise the lwIP stack but is of course now
>> completely incompatible with the existing eCos port. The sys_arch API
>> has been update as well to counetr the aforementioned memory leaks.
>>
>> I am also not sure what to do really with the IPv6 and IPv4 config
>> options for the lwIP stack.
>>
>> I have compiled the test cases with the new 1.3.0 source but I  still
>> can not get any of them to run on the Linux synth target. Some of it is
>> actualy me learning the synth target as I go along as well as me not
>> really having any time to spend on this other than time set aside out of
>> curiosity.
>>     
>
> I am keen to see lwIP updated in the eCos repository and can offer help
> with the CDL if necessary.
>
> It looks like good progress has been made with lwIP 1.3.0 on eCos
> already. Could you drop a note to this list with the current status of
> your work please and we can plan accordingly?
>
> Given the changes in the lwIP initialisation code, can anyone comment on
> the likely effort to port an eCos application based on the existing eCos
> lwIP stack to lwIP 1.3.0?
>
> lwIP is a good fit for eCos and I can see the use of this stack
> dominating in the future. With this in mind, it would be great to
> support as many features of the stack as possible (including IPv6).
>
> John Dallaway
>   

I just wanted to let you know that I have done an lwip 1.3.0 port for 
our current project too. I haven't committed anything yet as the port is 
not yet finished. I have re-written parts of the CDL, rewritten the 
ethernet driver interface and have a working port using the RAW 
interfaces. The sequential APIs are not yet supported in my port as I 
currently don't need them. I have also done some work on PPP, as we're 
using lwip to run a GPRS modem connection. I also use ethernet via the 
synth target for testing though. PPP is really in quite a horrible state 
and unmaintained in lwip, there is still lots to improve. Unfortunately 
I didn't really had the time for anything serious. I just got it working 
for our particular needs, which basically is just another hack ;)

Anyone interested in my code may have a look at 
http://git.inthemill.ch/?p=ecos.git;a=summary

I think it would be great to update the official lwip port in ecos and 
I'm willing to help where I can.

Simon

Re: lwIP

[John Eigelaar]

On Mon, 2009-04-06 at 11:44 +0200, Simon Kallweit wrote:

> I just wanted to let you know that I have done an lwip 1.3.0 port for 
> our current project too. I haven't committed anything yet as the port is 
> not yet finished. I have re-written parts of the CDL, rewritten the 
> ethernet driver interface and have a working port using the RAW 
> interfaces. The sequential APIs are not yet supported in my port as I 
> currently don't need them. I have also done some work on PPP, as we're 
> using lwip to run a GPRS modem connection. I also use ethernet via the 
> synth target for testing though. PPP is really in quite a horrible state 
> and unmaintained in lwip, there is still lots to improve. Unfortunately 
> I didn't really had the time for anything serious. I just got it working 
> for our particular needs, which basically is just another hack ;)
> 
> Anyone interested in my code may have a look at 
> http://git.inthemill.ch/?p=ecos.git;a=summary
> 
> I think it would be great to update the official lwip port in ecos and 
> I'm willing to help where I can.
> 
> Simon
> 

I will be more than willing to carry on from Simon's port in order to
get the sequential part going, as he has made much better progress than
I have. We almost exclusively use the stack in sequential mode on the
AT91 SAM7 mcu, so test environments should not be a problem.

Maybe we can get this port in some time soon then.

Regards 
John

Re: lwIP

[John Dallaway]

Hi John and Simon

On Mon, 2009-04-06 at 11:44 +0200, Simon Kallweit wrote:

> Anyone interested in my code may have a look at 
> http://git.inthemill.ch/?p=ecos.git;a=summary
>
> I think it would be great to update the official lwip port in ecos and 
> I'm willing to help where I can.

and John Eigelaar replied:

> I will be more than willing to carry on from Simon's port in order to
> get the sequential part going, as he has made much better progress than
> I have. We almost exclusively use the stack in sequential mode on the
> AT91 SAM7 mcu, so test environments should not be a problem.
> 
> Maybe we can get this port in some time soon then.

This looks like an excellent opportunity to demonstrate the benefits of
our free open source development model.

In a perfect world, it would be great to see:

a) Support for all functionality of lwIP 1.3.0 (including ethernet, PPP,
   IPv4, IPv6, raw API, sequential API, BSD socket API)
b) Compatibility with eCos applications using the existing eCos lwIP
   1.1.1 package if possible (API and existing CDL option names
   preserved)
d) Absolutely minimal mangling of the lwIP sources (it should be easy to
   upgrade lwIP again in the future)
e) All lwIP configuration points accessible via CDL options
f) DocBook (.sgml) documentation which describes any eCos-specific
   aspects of the stack and otherwise points to generic lwIP
   documentation
g) Any fixes necessary in the lwIP sources contributed up to the master
   lwIP project for inclusion in the next generic release
h) eCos test cases for all functionality

Of the above, IPv6 is probably the least likely to get covered. Note
that IPv6 requires multicast support (CYGINT_IO_ETH_MULTICAST) in the
eCos ethernet driver. Is there anyone with an IPv6 network to hand who
could put some effort into the IPv6 side? Perhaps on i386 PC hardware
with the RTL8139 or i82559 ethernet drivers which already support
multicasting?

Perhaps the best way forward is for John Eigelaar to take Simon
Kallweit's package, work on the sequential API initially and post his
revised package for review and further work by others. Simon, is the
lwIP package in your repository ready for handoff?

Many thanks to both of you

John Dallaway

Re: lwIP

[Simon Kallweit]

John Dallaway wrote:
>
> This looks like an excellent opportunity to demonstrate the benefits of
> our free open source development model.
>
> In a perfect world, it would be great to see:
>
> a) Support for all functionality of lwIP 1.3.0 (including ethernet, PPP,
>    IPv4, IPv6, raw API, sequential API, BSD socket API)
>   

Well, I think we might probably work with the CVS version as the 1.3.1 
release is coming along. The only problems I see is with PPP. In the 
current state it's very hard to support both the sequential (threaded) 
and raw (non-threaded) modes. Might be a good opportunity to clean it up 
and also commit it to the lwIP project.

> b) Compatibility with eCos applications using the existing eCos lwIP
>    1.1.1 package if possible (API and existing CDL option names
>    preserved)
>   

Hmm, I didn't really focus on that. I agree that CDL names should be 
identical as far as possible. Otherwise I'm not sure (initialization 
etc.) if this really matters as long as it's documented well.

> d) Absolutely minimal mangling of the lwIP sources (it should be easy to
>    upgrade lwIP again in the future)
>   

Mangling is very minimal for these exact reasons, except the PPP sources 
which were changed quite a bit.

> e) All lwIP configuration points accessible via CDL options
>   

Pretty much done I think.

> f) DocBook (.sgml) documentation which describes any eCos-specific
>    aspects of the stack and otherwise points to generic lwIP
>    documentation
>   

I agree. No documentation in my current port.

> g) Any fixes necessary in the lwIP sources contributed up to the master
>    lwIP project for inclusion in the next generic release
>   

I agree. Again, PPP might be the biggest part here.

> h) eCos test cases for all functionality
>   

I have already ported some tests.

> Perhaps the best way forward is for John Eigelaar to take Simon
> Kallweit's package, work on the sequential API initially and post his
> revised package for review and further work by others. Simon, is the
> lwIP package in your repository ready for handoff?
>   

I think we should first discuss my port a little and see if it needs to 
be changed to fit the needs of others.

The sources in my git repository are what I currently use in my project. 
Anybody willing to work on it may just branch my tree.

Simon

Re: lwIP

[John Dallaway]

Hi Simon

Simon Kallweit wrote:

>> In a perfect world, it would be great to see:
>>
>> a) Support for all functionality of lwIP 1.3.0 (including ethernet, PPP,
>>    IPv4, IPv6, raw API, sequential API, BSD socket API)
> 
> Well, I think we might probably work with the CVS version as the 1.3.1
> release is coming along. The only problems I see is with PPP. In the
> current state it's very hard to support both the sequential (threaded)
> and raw (non-threaded) modes. Might be a good opportunity to clean it up
> and also commit it to the lwIP project.

I agree.

>> b) Compatibility with eCos applications using the existing eCos lwIP
>>    1.1.1 package if possible (API and existing CDL option names
>>    preserved)
> 
> Hmm, I didn't really focus on that. I agree that CDL names should be
> identical as far as possible. Otherwise I'm not sure (initialization
> etc.) if this really matters as long as it's documented well.

If it's just a case of making a different API call to initialise the
stack then I agree this is no problem.

>> d) Absolutely minimal mangling of the lwIP sources (it should be easy to
>>    upgrade lwIP again in the future)
> 
> Mangling is very minimal for these exact reasons, except the PPP sources
> which were changed quite a bit.

Has the PPP support in the current lwIP code regressed relative to lwIP
1.1.1? If there have been serious regressions, we will need to consider
how to manage the transition for users of the existing eCos lwIP package.

If you consider your own fixes for PPP to be just a hack then we might
also consider focussing on ethernet for now and tacking PPP as a
separate activity.

>> h) eCos test cases for all functionality
> 
> I have already ported some tests.

Excellent.

>> Perhaps the best way forward is for John Eigelaar to take Simon
>> Kallweit's package, work on the sequential API initially and post his
>> revised package for review and further work by others. Simon, is the
>> lwIP package in your repository ready for handoff?
> 
> I think we should first discuss my port a little and see if it needs to
> be changed to fit the needs of others.

Sure. What aspects do you think need discussion?

> The sources in my git repository are what I currently use in my project.
> Anybody willing to work on it may just branch my tree.

No problem, so long as other contributors are comfortable with git.

John Dallaway

Re: lwIP

[Simon Kallweit]

John Dallaway wrote:
>>> b) Compatibility with eCos applications using the existing eCos lwIP
>>>    1.1.1 package if possible (API and existing CDL option names
>>>    preserved)
>>>       
>> Hmm, I didn't really focus on that. I agree that CDL names should be
>> identical as far as possible. Otherwise I'm not sure (initialization
>> etc.) if this really matters as long as it's documented well.
>>     
>
> If it's just a case of making a different API call to initialise the
> stack then I agree this is no problem.
>   

Most of the API comes from lwIP anyway. What will change (at least in my 
port) are a few CDL names as well as the API for initialization.

>>> d) Absolutely minimal mangling of the lwIP sources (it should be easy to
>>>    upgrade lwIP again in the future)
>>>       
>> Mangling is very minimal for these exact reasons, except the PPP sources
>> which were changed quite a bit.
>>     
>
> Has the PPP support in the current lwIP code regressed relative to lwIP
> 1.1.1? If there have been serious regressions, we will need to consider
> how to manage the transition for users of the existing eCos lwIP package.
>
> If you consider your own fixes for PPP to be just a hack then we might
> also consider focussing on ethernet for now and tacking PPP as a
> separate activity.
>   

I don't know if PPP code in lwIP has regressed. The code in my port 
certainly has, as it is tuned to run in a polled, single-threaded 
environment. I think it's a good idea to focus on ethernet support 
first. Fixing PPP is going to be quite some work if we want to do it 
right (in lwIP as well as in eCos). Also adding proper support for 
polling will need some changes in the lwIP code.

>>> Perhaps the best way forward is for John Eigelaar to take Simon
>>> Kallweit's package, work on the sequential API initially and post his
>>> revised package for review and further work by others. Simon, is the
>>> lwIP package in your repository ready for handoff?
>>>       
>> I think we should first discuss my port a little and see if it needs to
>> be changed to fit the needs of others.
>>     
>
> Sure. What aspects do you think need discussion?
>   

Well basically I think the code that I wrote so far :) I'll post a 
little summary of my thoughts and implementation later this day when I 
have some free time.

Simon