The board does boot, but only at DDR667. (I tried each option up to DDR1600)


On Tue, May 21, 2013 at 7:19 PM, David Hubbard <david.c.hubbard+coreboot@gmail.com> wrote:
> > If DDR3-667 works but AGESA fails at DDR3-1600, is it possible to go back
> > to DDR3-667 or try an intermediate speed, say DDR3-1333 ?
>
> Maybe there is an option in AGESA where you can limit the maximum
> frequency.
>
> $ more src/mainboard/asus/f2a85-m/buildOpts.c
> […]
> #define BLDCFG_MEMORY_BUS_FREQUENCY_LIMIT         DDR1866_FREQUENCY
> […]
> #define BLDCFG_MEMORY_CLOCK_SELECT                DDR1600_FREQUENCY
> […]

This is a great suggestion.

I have spent a few days studying the AGESA code to find out what frequencies it tries. I concluded that the built-in AGESA autodetection is very limited. It may be there is a bug in the AGESA autodetection because this is not the same code path generally used in vendor BIOSes. The error messages about missing tables are suspicious. I believe it means AMD has the motherboard OEM (i.e. Asus) provide parameter tables measured with a scope that help the timings match the production board.

Specifically I added IDS_HDT_CONSOLE (MEM_FLOW, "MemPPSCFlow enter " __FILE__ ":%d\n", __LINE__); and IDS_HDT_CONSOLE (MEM_FLOW, "MemPPSCFlow exit " __FILE__ ":%d\n", __LINE__); to the function MemPPSCFlow() in src/vendorcode/amd/agesa/f15tn/Proc/Mem/Ps/mp.c



I attempted setting BLDCFG_MEMORY_BUS_FREQUENCY_LIMIT and BLDCFG_MEMORY_CLOCK_SELECT:
diff --git a/src/mainboard/asus/f2a85-m/buildOpts.c b/src/mainboard/asus/f2a85-m/buildOpts.c
index 7f893f9..b9346b7 100644
--- a/src/mainboard/asus/f2a85-m/buildOpts.c
+++ b/src/mainboard/asus/f2a85-m/buildOpts.c
@@ -102,7 +102,7 @@
 
 #define BLDCFG_AMD_PLATFORM_TYPE                  AMD_PLATFORM_MOBILE
 
-#define BLDCFG_MEMORY_BUS_FREQUENCY_LIMIT         DDR1866_FREQUENCY
+#define BLDCFG_MEMORY_BUS_FREQUENCY_LIMIT         DDR1333_FREQUENCY
 #define BLDCFG_MEMORY_MODE_UNGANGED               TRUE
 #define BLDCFG_MEMORY_QUAD_RANK_CAPABLE           TRUE
 #define BLDCFG_MEMORY_QUADRANK_TYPE               QUADRANK_UNBUFFERED
@@ -116,8 +116,8 @@
 #define BLDCFG_POWER_DOWN_MODE                    POWER_DOWN_BY_CHIP_SELECT
 #define BLDCFG_ONLINE_SPARE                       FALSE
 #define BLDCFG_BANK_SWIZZLE                       TRUE
-#define BLDCFG_TIMING_MODE_SELECT                 TIMING_MODE_AUTO
-#define BLDCFG_MEMORY_CLOCK_SELECT                DDR1600_FREQUENCY
+#define BLDCFG_TIMING_MODE_SELECT                 TIMING_MODE_LIMITED
+#define BLDCFG_MEMORY_CLOCK_SELECT                DDR1333_FREQUENCY
 #define BLDCFG_DQS_TRAINING_CONTROL               TRUE
 #define BLDCFG_IGNORE_SPD_CHECKSUM                FALSE
 #define BLDCFG_USE_BURST_MODE                     FALSE

I then ran make clean; make and tried booting coreboot, but hit the same error (ASSERTION FAILED mmExcludeDimm.c line 26). I'm going to keep trying lower speeds.

I also attemped configuring coreboot with 1.65V though my memory is only rated to 1.60V. I checked with the vendor BIOS that the memory is at least bootable at 1.65V. Then I tried booting coreboot at 1.65V (without the previous memory bus limit), but it stopped at the same assertion.


> Could you try to find out with `bios_extract` for example what AGESA
> version the vendor BIOS uses.

I have not had time to run bios_extract but I did identify that the BIOS file supplied on Asus' website for the F2A85-M/CSM is identical to the data on the chip after stripping the first 2048 bytes. For example:

dd if=f2a85-m-asus-5103-2012.09.10.v2.10.1208.cap of=f2a85-flashrom.bin bs=2048 skip=1


> PS: David, Google Mail changed the compose interface and they send HTML
> message in addition to text by default. Could you change that to just
> plain text please [3][4]?

I think this is now plain-text only, please let me know if gmail sends it as HTML. Sorry!