Strange bug indeed ! Is this bug present on all CPU revision (D0 and D1, I don't know if E0 is available yet) ? JP Francois On Wed, 2005-02-09 at 17:10, Jesse Off wrote: > The multiple loads of the mode register look strange to me too. You should > only need to load the mode register once per chip, so e.g. if you have 4 > chip selects being used, you have to emit 4 load mode register operations. > > Note that there is an as-of-yet undocumented Cirrus chip errata we at > Technologic Systems found back in October. It turns out that the SDRAM > initialization operation for emitting the "PRECHARGE-ALL" does not actually > emit a PRECHARGE-ALL on the sdram bus. I would urge everybody designing > this chip in to contact their rep about this as it seems they really want to > keep this bug quiet. There is a kludgy software workaround we figured out > they should be able to give you. > > //Jesse Off > > > > Hi, > > > > I am looking at the SDRAM initialisation in U-boot. > > For the EDB9312 board, here is how the Mode register > > is set : > > mov r1, #0x0008800 > > ldr r0, [r1] > > > > ldr r1, =0x0400000 > > ldr r0, [r1] > > > > ldr r1, =0x0808800 > > ldr r0, [r1] > > > > ldr r1, =0x0c08800 > > ldr r0, [r1] > > > > According to the user guide, when MRS is set, bit 23-10 of the read > > address are sent in the mode register of the SDRAM. > > What is the use of doing multiple read, since each read will erase the > > previous one ? > > > > According to the user guide only one mode is supported : > > it's CAS=3, Burst length = 4, sequential mode, use burst write, > > which should translate in the following command : > > > > mov r1, #0x000c800 > > ldr r0, [r1] > > > > (See user manual p 459, under table 12.7) > > > > I am probably missing something here, so any light is welcome. > > > > Regards, > > > > JP Francois > > > > > > > > > >