Help! isync, exceptions, dummy function problems with RTEMS (near-head) PowerPC/virtex BSP

Joel Sherrill joel.sherrill at oarcorp.com
Thu Aug 9 19:08:41 UTC 2007


Robert S. Grimes wrote:
> Hi,
>
> I wanted to heed Joel's call to test RTEMS 4.7.99.2, so I wanted to get
> my code into a "stable state" so I could compare the new against the
> code I grabbed from HEAD about four weeks ago. 
Thank you.  I really do want to get from 4.7->4.8 faster
than 4.5->4.6 or 4.6->4.7.

>  Everything has been
> going along quite well, until I added some seemingly simple code - no
> new tasks or other RTEMS objects, just some application level state
> management code.  So the problem I am having is almost certainly in my
> code, but I'm having trouble figuring out what it is, mainly because of
> how it dies; it just hangs, and when I break into gdb, I find it is
> stuck on the isync instruction at the very end of  _CPU_Context_switch
> in new-exceptions/cpu_asm.S.  There is, of course, no stack information,
> and I'm not sure how to proceed.  Call this the "isync Problem".
>
> Another problem I run into is and infinite "exception handler called for
> exception 0"; this message is streamed out the console as fast as the
> serial port allows.  Note that the register dump that is supposed to
> follow does not happen.  Call this the "Infinite Exception 0 Problem"
>
> Another problem I have seen is the message "exception handler called for
> exception 7"; this one _does_ print out the registers before it halts
> the system (in its infinite loop).  Call this (not surprisingly) the
> "Exception 7 Problem".
>
> All these problems seem to have nothing to do with anything you would
> expect.  Right now, without having _knowingly_ fixed anything, the
> Exception 7 Problem seems to be gone.  But I am still quite stuck!
> However, I have a relatively simple setup where I can selectively cause
> the Infinite Exception 0 Problem.  In this, everything works just fine.
> But if I add this function:
>
>     void printCurTaskId() {
>       rtems_id id;
>       rtems_task_ident(RTEMS_SELF, 0, &id);
>       printk("Current Task: $%08x\n", id);
>     }
>
>
> to the source file that contains my main initialization routine (you
> know, the one RTEMS calls after it initializes), I get the Infinite
> Exception 0 Problem.  Now, please note that nowhere in my code is this
> function actually called!  If I simply comment out this function,
> everything works just fine!  Argh!!!  As far as I can tell, nothing is
> changed of any significance in the map files (see follow-up email, if it gets
> approved by the moderator - if you're
> interested).  Call this one the "Dummy Function Problem" (named after
> me, it seems!)
>
> So, now some specific questions:
>
> 1. What could cause the isync Problem?  I have absolutely no clue on
> this one.
>
> 2. What are exception 0 and 7?  In the Xilinx PPC Reference Guide, they
> document the available exceptions, but they don't give them "ID"
> numbers; they do give, however, their "exception-vector offsets".  For
> example, the Program Interrupt exception has an offset of 0x0700, and I
> certainly would believe that that is exception 7; however, there is no
> exception-vector offset 0x0000, so I don't know what Exception 0 is.  So
> what is the mapping between the exception number reported by RTEMS and
> the documented exceptions...
>
> 3. Any ideas what could cause the Dummy Function Problem, or what to check?
>
> My hardware is fine, I have enough memory - maybe it's the phase of the
> moon???
>
>   
I suspect this is the classic problem that the BSP left
an interrupt source enabled that should be disabled.
You have an interrupt pending when the first context
switch enables interrupts and it doesn't have a handler.

Double check the BSP interrupt initialization.  It needs
to clear/disable all sources and let individual drivers
enable the ones they will handle.

--joel
> Thanks,
> -Bob
>
> _______________________________________________
> rtems-users mailing list
> rtems-users at rtems.com
> http://rtems.rtems.org/mailman/listinfo/rtems-users
>   




More information about the users mailing list