On Monday 05 October 2009, Øyvind Harboe wrote:
> if (jtag_init_inner(cmd_ctx) == ERROR_OK)
What would you think about changing init_inner() so it
actually *fails* when significant errors are detected?
I still think that fix would have been sufficient to resolve
the problem Michal reported. His issue was that init_inner()
didn't report failure ... so the fallback "hard reset" logic
never got a chance to trigger. And it would have worked if
it had triggered... his proposed patch added the ability to
force that reset before calling init_inner().
Alternatively, using jtag_add_reset(0,0) to establish a
uniformly sane initial state should address the issue too.
(Since the root cause was the JTAG adapter *starting* in a
state with some resets asserted...) I think I'll try that
one out; I though I had observed the same error Michal was
reporting, when switching configs.
Right now in init_inner() we have:
/* examine chain first, as this could discover the real chain layout */
if (jtag_examine_chain() != ERROR_OK)
{
LOG_ERROR("Trying to use configured scan chain anyway...");
issue_setup = false;
}
... continuing after known brokenness. We might want to let
the "wrong version" cases succeed, and maybe some others, but
I don't see any valid reason to continue when the number of
TAPs observed doesn't even match the number configured, or
when TDO is stuck high (no device?).
When I run into this error it's because for example there's
only one TAP present but three are configured. Or the IDCODE
is wrong since I used the wrong config file. Or the board is
still in reset (Michal's original problem)...
if (jtag_validate_ircapture() != ERROR_OK)
{
LOG_WARNING("Errors during IR capture, continuing anyway...");
issue_setup = false;
}
... continuing after other known brokenness. Here, continuable
errors would be a lot harder to find; any error likely means we
don't even know the IR register sizes.
if (issue_setup)
jtag_notify_event(JTAG_TAP_EVENT_SETUP);
else
LOG_WARNING("Bypassing JTAG setup events due to errors");
return ERROR_OK;
... returning success regardless of any brokenness observed. So
the logic in the jtag_init() code that tried to recover from such
errors is a NOP, and never triggers.
And note, even after your patch these significant errors are STILL
completely ignored...
_______________________________________________
Openocd-development mailing list
Openocd-development@lists.berlios.de
https://lists.berlios.de/mailman/listinfo/openocd-development
