[PATCH 3/3] HACK: e1000: don't check for FLSWCTL.GLDONE when waiting for idle

[Uwe Kleine-König]

On Mon, Oct 09, 2017 at 11:15:09AM -0700, Andrey Smirnov wrote:
> On Mon, Oct 9, 2017 at 7:24 AM, Uwe Kleine-König
> <u.kleine-koenig at pengutronix.de> wrote:
> > On Mon, Oct 09, 2017 at 11:36:16AM +0200, Uwe Kleine-König wrote:
> >> I don't understand all the consequences of this patch yet, but this makes reading
> >> out the flash chip connected to an i210 work for me.
> >> ---
> >>  drivers/net/e1000/eeprom.c | 4 ++--
> >>  1 file changed, 2 insertions(+), 2 deletions(-)
> >>
> >> diff --git a/drivers/net/e1000/eeprom.c b/drivers/net/e1000/eeprom.c
> >> index 739bc17a519e..482a969f8d56 100644
> >> --- a/drivers/net/e1000/eeprom.c
> >> +++ b/drivers/net/e1000/eeprom.c
> >> @@ -709,8 +709,8 @@ static int e1000_flash_mode_wait_for_idle(struct e1000_hw *hw)
> >>        * execution by polling only FLSWCTL.DONE */
> >>
> >>       const int ret = e1000_poll_reg(hw, E1000_FLSWCTL,
> >> -                                    E1000_FLSWCTL_DONE | E1000_FLSWCTL_GLDONE,
> >> -                                    E1000_FLSWCTL_DONE | E1000_FLSWCTL_GLDONE,
> >> +                                    E1000_FLSWCTL_DONE,
> >> +                                    E1000_FLSWCTL_DONE,
> >
> > I tested a bit with and without this change at it seems as long as
> > nothing "strange" happens, testing for both FLSWCTL.DONE and
> > FLSWCTL.GLDONE (i.e. not applying my HACK patch) works fine.
> >
> > Still I think only testing for FLSWCTL.DONE is better because it works
> > well even if the state machine is in the middle of a read request and
> > then changing the command (which is always done after
> > e1000_flash_mode_wait_for_idle()) should work well.
> >
> > I'll resend with a better commit log once I tested this.
> >
> > Alexey: I didn't understand the comment above the patched line, maybe
> > I'm missing something?
> >
> 
> Assuming this question is for me: what I meant by that comment is that

Ah yes, sorry. It seems I have to participate in the cut'n'paste
seminar, too.

> all of the flash related operations (read, write, erase) already poll
> for E1000_FLSWCTL_DONE as their last step, so the only time that the
> state of that bit would be unknown would be right after reset, the
> first time any of those functions is executed.

Right after reset or when someone poked in the hardware using mw/md, or
if a previous transaction was not completed (nor sure this can happen
though).
 
> As for GLDONE bit, I tried to stick to algorithms described in
> "3.3.5.5 Software Flash Program Flow via the Flash-Mode Interface" and
> "3.3.5.6 Software Flash Read Flow via the Flash-Mode Interface" of Rev
> 2.8 of the datasheet and that's where that "requrement" is coming
> from.

3.3.5.5 and 3.3.5.6 in Rev 3.1 only check .DONE at the start and for the
last step tell (for 3.3.5.5): 

	FLSWCTL.GLDONE bit is set by hardware when the last byte
	programmed has been written.

So I'd not use that to check if I can start. Still more as there is:

	But software can stop the transaction in the middle as long as
	it got the DONE bit read as 1b.

> I haven't touched this particular HW/area in more than a year, so,
> since you actually work with it and can actually test things out I'll
> defer to you to judge if certain HW checks are needed or not. My code
> is by no means complete or exhaustively tested, so I have no problem
> believing that I got some of the parts wrong.

OK. So I will send a v2 that drops checking for GLDONE.

Thanks for your replies.

Uwe

-- 
Pengutronix e.K.                           | Uwe Kleine-König            |
Industrial Linux Solutions                 | http://www.pengutronix.de/  |