[lwip-users] message api bug and throughput problem

[kuschel]

Hi all!

I have two problems:

  1.  Possible mailbox bug in the message api
  2.  Throughput problem



1. Possible mailbox bug in the message api
------------------------------------------

I believe I have found a bug in the message api. When a connection is closed by 
a remote host, sys_mbox_free() is called while just one mail is present in the 
box. sys_arch.txt says:

" - void sys_mbox_free(sys_mbox_t mbox)
  Deallocates a mailbox. If there are messages still present in the
  mailbox when the mailbox is deallocated, it is an indication of a
  programming error in lwIP and the developer should be notified. "


I have debugged this using the socket based chargen test program. The client is 
a Windows machine running a special testprogram that connects to port 19 and 
measures the throughput. The error happens when I close the Windows program.

Here is a list of mailbox calls affecting the mailbox conn->mbox that does the 
error:


  A new connection comes in:

    accept_function():  calls  conn->mbox = sys_mbox_new()


  chargen writes data using the socket function
  lwip_write() several times:

    do_write()      :  calls sys_mbox_post(conn->mbox)
    netconn_write() :  calls sys_mbox_fetch(conn->mbox)

    do_write()      :  calls sys_mbox_post(conn->mbox)
    netconn_write() :  calls sys_mbox_fetch(conn->mbox)

    do_write()      :  calls sys_mbox_post(conn->mbox)
    netconn_write() :  calls sys_mbox_fetch(conn->mbox)

    ... and so on ...


  The Windows host closes the connection. lwIP does now:

    do_write()      :  calls sys_mbox_post(conn->mbox)
    err_tcp()       :  calls sys_mbox_post(conn->mbox)
    netconn_write() :  calls sys_mbox_fetch(conn->mbox)


  The mailbox contains now the mail sent by err_tcp().

    netconn_delete():  calls sys_mbox_fetch(conn->mbox)
    do_delconn()    :  calls sys_mbox_post(conn->mbox)
    netconn_delete():  calls sys_mbox_free(conn->mbox)

As you can see, netconn_delete() gets the mail sent by err_tcp() just before 
do_delconn() sends the mail destined for netconn_delete(). The call to 
sys_mbox_free() fails now since the mailbox contains yet one mail.

Is there any reason why err_tcp() posts a mail to conn->mbox? Which function is 
normally waiting for this mail?



2. Throughput problem
---------------------

I am running lwIP on a PowerPC 440GP at 400MHz with uCOS-II. For our 
application I need a throughput of at least 35 MBit/s. The throughput I get is 
only 7.6 kB/s per connection. I can open multiple connections each running at 
7.6 kB/s while the PowerPC is still lazy at 3% CPU usage. Has anybody observed 
the same behaviour? I have tried to increase the tcp timer rate but without 
luck. I am using chargen.c from the lwIP contribution package for my test.


Help is welcome.

Dennis