[U-Boot] Numonyx NOR support in u-boot

[Andrew Dyer]

On Tue, Jun 29, 2010 at 2:28 PM, Todd Fischer <todd.fischer at ridgerun.com> wrote:
> Hi,
>
> We are seeing NOR write failures on 6 of the 10 boards (DM6446 ARM SoC
> with Numonyx JS28F256M29EWL NOR flash chip).  Four of the boards seem to
> work without error.  All 10 boards were made on the same manufacturing
> line at the same time.
>
> Initially, we couldn't properly read the  QRY response from the chip.
> We had to use this patch
>
> http://www.mail-archive.com/u-boot@lists.denx.de/msg26614.html
>
> We are getting 'flash not erased' error or timeout error.  We have tried
> monitoring D7 (busy) and monitoring D6 (activity toggle) to detect
> erase / program command completion / timeout.  We have tried changing
> the timing of the control signals to the NOR chip, but need to use a
> logic state analyzer to verify the timing is changed as expected.
>
> Example error output:
>
>> cp 0x81700000 0x02000000 0x400000
> Copy to Flash... Flash not Erased
> Timeout writing to Flash
>
>> cp 0x81700000 0x02000000
> 0x400000
> Copy to Flash... Flash write timeout at address 2000000 data
> 129f34
>
> We have tried cp, cp.b, cp.w and get the same results.
>
> With 4 boards working correctly, the code must be close.
>
> Any suggestions (beyond looking at the signals with a Logic State
> Analyzer)?
>

You didn't say what version of u-boot.  A few years ago when doing a
DM6446 design, we used a modified 1.2.x that TI was distributing with
the DM6446 eval boards (customer insisted that we start from the TI
distribution), and as I recall it had a bad implementation of the
timer routines.  If I remember right they were running the timer at
something like 3.84 MHz.  For a while now, the u-boot standard has
been that CONFIG_HZ should be 1000.  In the case where it's not, some
of the timeouts in the common CFI flash code will be way too short and
will occasionally fail.  IIRC, TI had their own individual flash.c
where the timeouts were done differently, and we only found it when we
tried to use the CFI flash code.  Another timer problem that will
cause similar issues is if the timer ignores the base argument to
get_timer().