I have been giving this a lot of thought, and I would like to suggest a compromise :)
I dont have the code with me right now so some of some function names ´might be a bit off but the idea should be clear. > One problem about jtag_add_end_state() is that it's effectively > setting a global variable. > > This makes it hard to tell, reading some code, what is expected > to happen after a JTAG scan. > > Some other calling code invoked jtag_add_end_state() with some > end state. True, but for some parts of the code, the job is to send a certain sequence through the DR scan chain without worring about end states, while some other section handles the logic of rest states, if we must pass idle a certain number of times, or not, etc. So it happens that this is a feature, example in the arm_adi low level scan functions, the end state is really a "dont care, unless top level has a strong opinion". And there are several layers betwen the top layer and the generation of the scan structures. In the worst case, every call to memap_read/write_memory functions must supply an end state, and this must propagate all the way down to the scan_inout funtions. > Other than running the debugger, I don't know how to figure out > what the end state is. We can call get_end_state() > Wouldn't it be better if the code had to pass in the end state > as an argument to the scan operations? Not always, TAP_INVALID is really TAP_DONTCARE > That way there would be some clue locally in the code as to what state > the JTAG state machine moves to after a scan.... > > Is there a fundamental reason(other than that we need > stability on svn head :-) why the following couldn't work: > > - remove jtag_add_endstate() And my alternative suggestion is: - ONLY call jtag_add_endstate() from target layers, never inside jtag drivers. - end_state is the resting state for the next complete scan, nor the end of the next state_move generated within jtag drivers. - add an state_move_end_state state to the state_move commands, jtag_state_move(state_move_end_state) effectively changing tho following pattern seen in some jtag drivers today saved_end_state = get_end_state() add_end_state(end_of_scan_section ) state_move() /* to end of scan section */ set_end_state(saved_end_state) .. .. state_move() /* to "real" end state */ TO state_move(end_of_scan_section) .. .. state_move(get_end_state()) This will not change the api calling the jtag layer, but it will clean up some of the jtag implementation code. Best regards Magnus _______________________________________________ Openocd-development mailing list Openocd-development@lists.berlios.de https://lists.berlios.de/mailman/listinfo/openocd-development
