[lwip-users] netif related quetions

[Bilahari Akkiraju]

I am getting more and more questions as I keep trying to understand the lwip 
code.

 And I am using lwip 1.2.0. I believe this is the latest version , am I right?. 
Hoping that I am right :

For now I have few netif related questions.


  I understand that ethernetif.c is to be tailored according to the user 
dependent implementation of how he glues his low level driver to lwip, on that 
I have these questions,

   1. Why is netif->state initialized two times, once in netif_add (netif.c) 
and again in ethernetif_init() (ethernetif.c)?. The second initialization is 
going to overide the first one anyway right ?. The question is the netif_add() 
calls ethernet_init(), so why do we have to assign netif->state two times both 
thes two functions ?. Cant it be initialized always only one time & at only one 
place ?. Whats the purpsoe behind doing it that way.
 
   2. Why isnt some function of format "err_t xxxxxxx(pbuf *, netif * )" 
declared or implemented in ethernetif.c, shouldnt this be passed as last 
argument when you call netif_add ( ) ?.  In ethernetif.c, in 
ethernetif_input(...) function, netif->input(p,netif) is called. What exactly 
should be called as input function ?. How should I pass the pbuf obtained from 
low_level_input() to network layer ?

Thanks in for sparing time to answer my questions

Bilahari
-----Original Message-----
From: address@hidden [mailto:address@hidden On Behalf Of address@hidden
Sent: Friday, May 25, 2007 4:45 AM
To: address@hidden
Subject: lwip-users Digest, Vol 45, Issue 33

Send lwip-users mailing list submissions to
        address@hidden

To subscribe or unsubscribe via the World Wide Web, visit
        http://lists.nongnu.org/mailman/listinfo/lwip-users
or, via email, send a message with subject or body 'help' to
        address@hidden

You can reach the person managing the list at
        address@hidden

When replying, please edit your Subject line so it is more specific than "Re: 
Contents of lwip-users digest..."


Today's Topics:

   1. RE: Xilinx lwIP raw-api beginner question (Pisano, Edward A)
   2. Problem with lwip_select under Nucleus (address@hidden)
   3. Re: Problem with lwip_select under Nucleus (Kieran Mansley)
   4. Basic Design Question (Spies, Dominik)
   5. RE: Basic Design Question (FreeRTOS.org Info)
   6. RE: Basic Design Question (Spies, Dominik)


----------------------------------------------------------------------

Message: 1
Date: Thu, 24 May 2007 16:32:39 -0000
From: "Pisano, Edward A" <address@hidden>
Subject: RE: [lwip-users] Xilinx lwIP raw-api beginner question
To: "Mailing list for lwIP users" <address@hidden>
Message-ID:
        <address@hidden>
        
Content-Type: text/plain;       charset="us-ascii"

Hi Erik,
The problem you describe sounds very similar to the problem I had when I first 
attempted to run the raw_api XiLinx example.  The solution was that the sample 
code assumed a polling design, whereas my MicroBlaze architecture had the EMAC 
configured for interrupts.  As such, I needed to also register the 
XEmac_IntrHandlerFifo function to be called when the ethernet controller 
notified EMAC that a message was incoming.  I put this call right before the 
call that registers the "mytimer_int_handler" like you have in your code.

   /*
    *Register XEmacHandler with interrupt controller
    *This is the FIFO-based interrupt handler
    */
   XIntc_RegisterHandler(XPAR_OPB_INTC_0_BASEADDR,
 
XPAR_OPB_INTC_0_OPB_ETHERNET_0_IP2INTC_IRPT_INTR,
                         (XInterruptHandler)XEmac_IntrHandlerFifo,
                         xemacif_ptr->instance_ptr);

Of course, you'll need to change the constants to the names used in your 
architecture.

Also, running lwip-debug for STDOUT messages has unpredictable effects on lwIP 
performance.  It has to do with how the UART is a synchronous device.  When 
strings are sent to it, processing is held up until the string output is 
complete.  This can results in lwIP not being able to service things as quickly 
as it should. I use STDOUT very sparingly when network traffic is underway.

Xemacif_input is still a polling function.  The Ethernet controller to the EMAC 
core may be interrupt driven.  But the EMAC to lwIP interface, via 
xemacif_input, is polled.  That's what your mytimer_int_handler helps to make 
happen. 

Regards, Ed (a.k.a.-Paisan)
-----Original Message-----
From: address@hidden
[mailto:address@hidden On Behalf Of Erik Jagre
Sent: Wednesday, May 23, 2007 11:57 PM
To: address@hidden
Subject: [lwip-users] Xilinx lwIP raw-api beginner question

Hi, I am a lwIP beginner and have some questions about getting a raw-api 
application up and running on a Xilinx Spartan-3E 1600E MicroBlaze Development 
Kit.

I have tried to run the xilinx reference design xapp663 (echo server) but 
without success. It seems like my design never takes care of incoming traffic. 
When I try to connect to my echo-server with telnet ARP requests are sent to 
the development board. But my design never replies.

I have tried to use the lwip-debug feature, but my program gets stuck during 
the call to netif_add (this does not happen without the lwip-debug feature 
activated). A print-out is made that confirms that the IP address has been set 
but after that the program stops and the program counter is set to 0x00. This 
makes me belive that my call to netif_add is not done in the right way. How is 
netif_add used correctly?
What preparations need to be done before calling it?

I am also curious about how xemacif_input(default_netif) works. Is this the 
call that takes care of all incoming traffic? Or is another call needed to take 
care of ARP requests?

My code is included in the zip-file.

Grateful for any suggestions or ideas.

Regards
Erik




------------------------------

Message: 2
Date: Fri, 25 May 2007 09:54:20 +0200
From: address@hidden
Subject: [lwip-users] Problem with lwip_select under Nucleus
To: address@hidden
Message-ID:
        <address@hidden>
Content-Type: text/plain; charset="iso-8859-1"

First of all I would like to thank you very much for your help.

Next, I would like to explain the scenario in which the problem takes 
place. I have found a standard Unix-sockets code in a book and I have 
implemented it for Win32 with no problem. Since the objective is to create 
a wrapper socket API in order to work with a few OSs (Win32, 
LWIP&Nucleus,...), I tried to do the same for LWIP. The structure of the 
application is as follows:

Server                   Client
======                   ======
socket
bind
listen
                         socket
                         connect <server>
accept 
send
                         recv
socket_delete


This daytime application works in Win32 (both server and client in Win32) 
with no problem. But in LWIP(server in Win32 and client in LWIP), before 
the server sends the packet, the client ends the program closing the 
connection. This way, the client finds nothing to receive, and the server 
has no connection for when he is going to execute the send statement, 
causing a send error.

At this point, I supposed that for default, sockets in LWIP were 
non-blocking (is this true???), and in consequence, the lwip_recv() worked 
as non blocking. That would be the reason for which it would not wait for 
the server to send, and the client execution would finish before the 
server send statement.
Introducing a one second sleep statement just before lwip_recv() all 
worked ok.

That´s why I tried to implement a blocking recv() by calling select just 
before recv(). But In spite of the 5 seconds timeval structure used as 
input to this select, it did not wait. I used this select looking at the 
readset in which the socket used to receive was set. At first I was afraid 
that there was some kind of problem with the semaphores, dealing with the 
port to Nucleus or something like that. But I have had a close look at 
some other select calls, and as long as there is not a connect before, 
they work perfectly waiting the timeval value etc.

I have gone into the select by debugging, and the execution changes in 
sys_sem_wait_timeout(in sys.c). This function returs a 1 when it does not 
wait, and a 0 if it does wait. Inside this function, the problem is in the 
sys_timeout function, as it does let the sswt_cb.timeflag with a 0 value 
when it does not wait, while it changes its value to 1 when it waits.

I cannot find an explanation to what happens, but actually, select does 
not wait when it is called after a connect. I would be very pleased if 
somebody could give me some advice.

Regards,

Borja.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: 
http://lists.gnu.org/pipermail/lwip-users/attachments/20070525/3a2d98e8/attachment.html

------------------------------

Message: 3
Date: Fri, 25 May 2007 09:14:55 +0100
From: Kieran Mansley <address@hidden>
Subject: Re: [lwip-users] Problem with lwip_select under Nucleus
To: Mailing list for lwIP users <address@hidden>
Message-ID: <address@hidden>
Content-Type: text/plain

On Fri, 2007-05-25 at 09:54 +0200, address@hidden wrote:

> At this point, I supposed that for default, sockets in LWIP were non-
> blocking (is this true???),

No.  By default socket operations are blocking.  I assume you haven't
specified MSG_DONTWAIT in the flags to the recv operation, or O_NONBLOCK
to the socket options.  These would both result in non-blocking sockets.

> I cannot find an explanation to what happens, but actually, select
> does not wait when it is called after a connect. I would be very
> pleased if somebody could give me some advice. 

It sounds like there's a problem with your port that is causing the
blocking operations in the sockets API to not block.  For some reason
lwIP is behaving as if there is data to read on that socket, when in
fact there is none.

Ignore select() for now as I think if we can solve the simpler recv-not-
blocking issue, that will be a good start.

Take a look at the lwip_recvfrom() function.  Can you check the
following when you call it by adding some extra debugging:
 - that sock->lastdata is NULL.
 - that buf returned by netconn_recv() is NULL.

If that is the case, take a look at netconn_recv().  This can return
NULL for all sorts of reasons.  Add debugging (e.g. a printf) to each
one and see which case is failing.

With that information we should be able to work out what's wrong.

Kieran







------------------------------

Message: 4
Date: Fri, 25 May 2007 10:41:40 +0200
From: "Spies, Dominik" <address@hidden>
Subject: [lwip-users] Basic Design Question
To: "Mailing list for lwIP users" <address@hidden>
Message-ID:
        <address@hidden>
Content-Type: text/plain; charset="us-ascii"

Hi!

I have some basic questions about the software design with the lwIP
stack.
Currently I'm using a port of FreeRTOS with lwIP. This port uses lwIP
with NO_SYS true and calling it in one single task in a infinite loop.

Every cycle ethernetif_handlepackets is called, and after that depending
on some timing things the tmr functions.

So, I do not think this is a good solution. My first idea is:

Two Tasks, on handling the incoming packets, blocked via queue. An ISR
handles incoming packets and write it to the queue.

The second tasks executes all tmr functions.

Because a preemptive scheduler is used, the lwip stack stuff has to be
protected via mutexes.

Two tasks for recieving/tmr processing is overhead. So how could I do
that in one task? I can either wait for a specified time (tmr stuff) OR
for an event (incoming packet) but not both (This is by design of
FreeRTOS I think). Suggestions?

Or is it in my case generally better to use the socket api, especially
regarding to RAM consumption? I did'nt spent any time on considering to
use it because rawapi.txt says it uses much more ram, and I do not have
unlimited amount (40kb total) of ram. I do not know how much more ram
the socket api needs, but maybe it is not much more or even less than
with my "two tasks and queue... " solution.

I hope this is not to basic for this list, but I'm no expert in realtime
kernels or embedded systems design, so I hope some of you experts can
spend some time helping a beginner ;)

Thanks.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: 
http://lists.gnu.org/pipermail/lwip-users/attachments/20070525/eed1fa16/attachment.html

------------------------------

Message: 5
Date: Fri, 25 May 2007 10:27:21 +0100
From: "FreeRTOS.org Info" <address@hidden>
Subject: RE: [lwip-users] Basic Design Question
To: "'Mailing list for lwIP users'" <address@hidden>
Message-ID: <address@hidden>
Content-Type: text/plain;       charset="us-ascii"

> I have some basic questions about the software design with 
> the lwIP stack.
> Currently I'm using a port of FreeRTOS with lwIP. This port 
> uses lwIP with NO_SYS true and calling it in one single task 
> in a infinite loop.


I would be interested others comments on this too - but also be aware that
there are some nice lwIP demos for FreeRTOS.org already that do not use
NO_SYS.  You can use these as a reference.  The most recent in the AVR32
demo.  There is also an ARM7 demo.  These are available in the download.

Regards,
Richard.

+ http://www.FreeRTOS.org
A free real time kernel for 8, 16 and 32bit systems.

+ http://www.SafeRTOS.com
An IEC 61508 certified real time kernel for safety related systems.









------------------------------

Message: 6
Date: Fri, 25 May 2007 13:44:32 +0200
From: "Spies, Dominik" <address@hidden>
Subject: RE: [lwip-users] Basic Design Question
To: "Mailing list for lwIP users" <address@hidden>
Message-ID:
        <address@hidden>
Content-Type: text/plain;       charset="us-ascii"

Richard,

This didn't answer my questions, but thanks for your hint on that, the
AVR32 example is nice. One question regarding the socket api:
Where is the timeout, icmp and other stuff processed?

Dominik



-----Original Message-----
From: address@hidden
[mailto:address@hidden On
Behalf Of FreeRTOS.org Info
Sent: Freitag, 25. Mai 2007 11:27
To: 'Mailing list for lwIP users'
Subject: RE: [lwip-users] Basic Design Question

> I have some basic questions about the software design with 
> the lwIP stack.
> Currently I'm using a port of FreeRTOS with lwIP. This port 
> uses lwIP with NO_SYS true and calling it in one single task 
> in a infinite loop.


I would be interested others comments on this too - but also be aware
that
there are some nice lwIP demos for FreeRTOS.org already that do not use
NO_SYS.  You can use these as a reference.  The most recent in the AVR32
demo.  There is also an ARM7 demo.  These are available in the download.

Regards,
Richard.

+ http://www.FreeRTOS.org
A free real time kernel for 8, 16 and 32bit systems.

+ http://www.SafeRTOS.com
An IEC 61508 certified real time kernel for safety related systems.







_______________________________________________
lwip-users mailing list
address@hidden
http://lists.nongnu.org/mailman/listinfo/lwip-users




------------------------------

_______________________________________________
lwip-users mailing list
address@hidden
http://lists.nongnu.org/mailman/listinfo/lwip-users

End of lwip-users Digest, Vol 45, Issue 33
******************************************