[linux-cirrus] Re: SDRAM initialisation
- From: "Jesse Off" <joff@xxxxxxxxxxxxxxx>
- To: linux-cirrus@xxxxxxxxxxxxx
- Date: Thu, 10 Feb 2005 09:00:10 -0700
> There is an errata for the 9301 which includes a
> claim of instability with auto-precharge at SDRAM
> bus frequencies in excess of 50Mhz:
>
> http://www.cirrus.com/en/pubs/errata/er636c1.pdf
>
> Does this relate to the condition you describe?
No. Thats a completely different problem. The problem is in SDRAM
initialization, auto-precharge is a controller mode that has the potential
to increase SDRAM speed by saving an explicit SDRAM precharge cycle on the
bus.
There is a multi-step procedure you have to follow explicitly to properly
initialize SDRAM. One of those steps is a PRECHARGE-ALL operation which
will precharge (close) all open banks. The EP93xx datasheet says that by
having only the INIT and CKE bit set in the GICONFIG register will issue a
PRECHARGE-ALL bus cycle-- it doesn't. Way back before when we caught this
on the logic analyzer and devised our workaround, this was causing our POST
SDRAM test to fail randomly by clobbering the first 4-16 SDRAM reads/writes
out of reset. Certain SDRAM chips were more or less susceptible to not
being properly initialized. It is impossible to get a 100% up-to-spec SDRAM
initialization without a PRECHARGE-ALL bus-cycle.
//Jesse Off
Other related posts:
