I've been working on bringing up Linuxbios on a board which is based closely on the EPIA-M and I think I may have found a bug in the DDR SDRAM init code in src/northbridge/via/vt8623/raminit.inc
The VIA Northbridge data sheet says that as part of the DDR init sequence you need to read from specific addresses on the bank but I think the registers to do these reads are not being set up properly which can cause the RAM test to fail.
I adjusted a number of register settings to be the same as a sample Award Bios but it made no difference so I took a closer look at the DDR and found that %ecx was being set up with the memory location to read from instead of %esi.
The diff for my fix is as follows:
Index: src/northbridge/via/vt8623/raminit.inc =================================================================== RCS file: /cvsroot/freebios/freebios/src/northbridge/via/vt8623/raminit.inc,v retrieving revision 1.5 diff -r1.5 raminit.inc 85c85 < movl $0x2000, %ecx ---
movl $0x2000, %esi /* IAS changed from ecx to esi */
89c89 < movl $0x800, %ecx ---
movl $0x800, %esi /* IAS changed from ecx to esi */
119c119 < movl $0x350, %ecx ---
movl $0x350, %esi /* IAS changed from ecx to esi */
What is odd is that the original code works fine on the Via reference board (same design as EPIA-M) but when I try to run it on my customer board it fails until I make these changes.
Does anyone with more knowledge than me of the Via chipsets have any ideas about this?
Cheers
Ian