[Rxtx] Port Locking / JVM Crashing

[Kharbanda, Pawan]

Trent,
    I found the culprit .... RXTX2.1.7 + JDK 1.5.0_06 + Redhat kernel
2.4.21-37.Elsmp + digi box is a bad combination. I upgraded one of my
machine(with 8 processors) to the new kernel (2.6.9-34.0.2.ELsmp) and
have absolutely no problem with JVM crashing or Port Locking even after
10000 operations (now I am trying 100,000 :). This is all with the
original RXTX 2.1.7 (final) api's without any of my fixes that I have
been trying for last 2-3 weeks. 

I will suggest that we update the INSTALL or Readme file accompnying the
RXTX saying the suggested kernel for Redhat Linux is 2.6.9-34.0.2.Elsmp
and use Digi PortServer II with the following configuration (Model:
PortServer TS 16 Rack  Firmware:   Version 82000684_S 05/22/2006 POST:
82000685_G 02/02/2006). This should helped everybody using the RXTX
API's with RedHat Linux atleast.

Regards
Pawan Kharbanda


   
-----Original Message-----
From: Trent Jarvi [mailto:tjarvi at qbang.org] 
Sent: Monday, August 07, 2006 6:02 PM
To: Kharbanda, Pawan
Cc: Trent Jarvi; RXTX Developers and Users
Subject: RE: Port Locking / JVM Crashing


Thanks Pawan

I thought for sure it was a kernel driver problem.  When I looked at
digi.com and saw it was probably not in the kernel, I started to think
even moreso.  The kernel authors like to make things hard for external
drivers.

But when I saw the error this morning, I thought it was a
synchronization issue.

The other thing I was thinking about was the undef's I mentioned.  rxtx
behaves differently for some kernel drivers that do not support
everything.  We have to do that to get the data we need.

I was trying to figure out what it may be but not having hardware makes
it interesting.

On Mon, 7 Aug 2006, Kharbanda, Pawan wrote:

> Trent,
>    I think I got the problem fixed after some code changes in the C
> (SerialImp.c) and Java(RXTXPort.java) code. I am now testing on the 
> staging environment (with 10 devices) with real time data and haven't 
> seen a JVM lock/ Port locking since last night. I am going to do 
> further stress testing and will send all my code changes and my 
> findings in next few days.
>
> I believe the problem was causing due to the digi drivers and how the 
> Linux kernel talked to it, so after making some code changes and 
> synchronization fixes, I got it to work. I am keeping my fingers 
> crossed
> :)
>
> Thanks for all your help. You have been a great help.
>
> Will keep you posted.
>
> Cheers
> ~pk
>
> PS : We are going to test my new RXTX changes with the 2.6 Red Hat 
> kernel (currently using 2.4.21-37.ELsmp) too.
>
> -----Original Message-----
> From: Trent Jarvi [mailto:tjarvi at qbang.org]
> Sent: Monday, August 07, 2006 5:08 PM
> To: Kharbanda, Pawan
> Subject: RE: Port Locking / JVM Crashing
>
>
>
> Hi Pawan
>
> As mentioned I did reproduce the error but after several hours of 
> cycles.
>
> I'm looking at RXTXPort.java and thinking about what may be happening.
> The error is in the C layer but it may be preventable in the java
code.
>
> lets say two closes are overlapping or an operation is trying to 
> happen while another port is closing.  The structure may be out of 
> sync by the time the code using the structure in another thread gets 
> to the event info struct.
>
> So I'm wondering if synchronization in RXTXPort() and close() may help

> you.
>
> A second thing that may help is you can put printf("got here\n"); in 
> each of the lines in initialize event info struct to verify that the 
> problem is in walking the linked list.  Thats my suspicion.
>
> So a properly timed close() while calling the constructor may be 
> catching rxtx off guard.  You can synchronize to test that and it may 
> do away with the problem.
>