Re: [avrdude-dev] chip_erase and cycle_count cleanup

[Jan-Hinnerk Reichert]

On Monday 01 December 2003 04:06, Brian Dean wrote:
> On Sun, Nov 30, 2003 at 10:53:52PM +0100, Jan-Hinnerk Reichert 
wrote:
> > However, I'm still unsure about a few things.
> > - I have added a check, so that cycle count isn't tracked, if it
> > can't be read before erase. Is this the right thing to do?
>
> Sounds reasonable.  If it cannot be read, it cannot be updated.
>
> > - Why is cycle count reset at 0xffff?
>
> Good question.  I think this is a bug.  I think it should be
> compared against 0xffffffff.  A long time ago the cycle count was
> 16 bits, but was later changed to 32 bits.  This code was progably
> missed.

There was another goodie in avr_get_cycle(). If the count was 
0xffffffff it was set to 0xffff. So, every check against 0xffffffff 
was not doing anything.

> The idea here is that if '-y' is specified, but we are using a
> blank chip, i.e., one where -Y has never been used, just assume
> that 0xffffffff means 0 for the cycle count instead of forcing
> someone to initialize the cycle count to 0 first.

Okay, I now do the following: avr_get_cycle_count() returns 0 in 
cycles if 0xffffffff is read.

The only downside is that -y does not display the cycle count if it is 
not set yet and no erase is performed.

I also changed the output logic a bit:
- main.c outputs the cycle count, if it is not set anywhere else.
- on all other occasions cycle count is output, if and only if it is 
set.

And I changed avr_get_cycle_count() and avr_set_cycle_count() to use a 
loop. I'm not sure, if this increases or decreases readability.

Patch still works fine for avr910.

> At least that is what I think.  I've been tracking down the origin
> of that code via CVS.  Funny how we come full circle - there
> actually used to be an avr_chip_erase() function in avr.c and that
> is where the cycle count code eventually ended up (moved from
> main.c).  Later, the avr_chip_erase() function became the
> pgm->erase() function for each programmer, and at that time there
> were only 2 programmers: ppi and stk500.

;-)

/Jan-Hinnerk

cycle_count2.diff
Description: Text Data