Zach Welch wrote:
> On Sat, 2009-04-18 at 06:31 -0500, Dick Hollenbeck wrote: 
>   
>> 1)
>> I would like to see tap_set_state() called for ALL state changes, not 
>> just the end points of of a multi-state path move.  This way when 
>> logging is compiled ON, we can see what each driver is doing (more 
>> exactly, what it thinks it is doing) to the taps.  The importance of 
>> this comes into play when programming FPGAs and CPLDs and 
>> troubleshooting that process.
>>     
>
> It sounds like this could be accomplished as an independent patch, yes?
>   

Well, since this needs to be done for each driver, I think it could be a 
matter of getting a reference implementation in place, then the other 
drivers would have to sprinkle in the calls to tap_set_state() as 
needed.  But there may be some jtag***() functions where this can be 
done in a common way, like the path move, etc.  It will have to be 
looked into.


>   
>> 2)
>> Have a look at ft2232.c which I can get to seg fault and also to get 
>> into a mode from which it does not recover simply by disconnecting the 
>> cable.  It is not robust.
>>     
>
> Have you posted a backtrace for the crash?  I just looked back and was
> unable to find one since Feb.
>   
I never posted it.  When the SVF and XSVF functions call the jtag 
routines, they can push those underlying jtag routines in directions 
that they don't normally travel.  There was a case where if a bitcount 
exceeded some number, then ft2232.c would seg fault.  I know where the 
bug is, and have been planning on dealing with it.  That limit is not 
normally hit with the bulk of the code in OpenOCD.  A hint as to the fix 
to the problem is in my add_clocks function ft2232.c.  The overflow 
condition has to be checked for *within* the loop, basically.  I'm not 
staring at the code now, but that's what I remember.


Dick

_______________________________________________
Openocd-development mailing list
[email protected]
https://lists.berlios.de/mailman/listinfo/openocd-development

Reply via email to