Nanya/Micron 2Gb DRAM doesn't work

[wad]

When testing the latest 2Gb DD3 die from Nanya (NT5CB256M8GN-CG), which is also shipped as Micron's MT4J256M8HX-15E, memory errors prevented a current OFW from booting.

Further examination showed that our existing memory timing is not valid for those parts. Per JEDEC JESD-79-3D ( http://www.jedec.org/standards-documents/docs/jesd-79-3d ), 2Gb parts are allowed to have a longer tRFC than used by our current timing.

We haven't seen a problem with already qualified 2Gb parts, but those will probably be EOL'd soon in favor of newer dies.

I suggest that CForth's memory configuration code be changed to include conditional setting of register 0xd0000060 to 0x64660404 (instead of 0x646602C4) when configuring 1GB of memory.

This value meets the JEDEC requirements, and has been tested to work with the Nanya memory. It appears both Micron test boards had rework problems, and haven't been tested yet.

Before releasing this modification, we should test with all qualified 2Gb parts. It is expected, however, that even 1Gb parts should work with this more conservative timing.

[wad]

Just a clarification that 2Gb DDR3 parts are used in all XO-1.75 laptops which have 1GB of memory loaded.

[wmb@…]

Fixed by git commit ead02cb. Barely-tested so far. It seems to do the right thing on a 512 MiB machine. Test version at scp dev.laptop.org:~wmb/q4d16mk.rom

[Quozl]

Testing in progress, two 1GB and two 512MB machines.

diag-mode test /memory many

... and a counter.

[Quozl]

Passed 730 cycles on 1GB, and 1475 cycles on 512MB, looks good.

Tested q4d16mk with os14, it doesn't boot, but q4d16 boots fine, perhaps something in q4d16mk? Tested q4d16je which is using the same git commit ead02cbe, no symptom. Will test runin with that over the weekend.

[Quozl]

Is in Q4D17.

[Quozl]

Runin testing of Q4D17 and 11.3.1 is in progress.

Runin testing of Q4D17 and 12.1.0 os14 is too unstable to give me confidence with this change, so it is ongoing.

(Pulling this ticket out of the 11.3.1 milestone since that release has been just completed, and the firmware is loaded before the software download by manufacturing.)

[Quozl]

Runin test of Q4D17 and 11.3.1 passed 24-hours fine.

[Quozl]

Q4D18 memtest

• in progress on SKU200X3, overnight, 486 cycles.
• in progress on SKU201, overnight, 555 cycles.
• in progress on four units by Chia-Hsiu, (one Micro, three Nanya), overnight passed,

Runin test of Q4D17 and 11.3.1 passed 48-hours fine.

Chia-Hsiu plans to put four units in runin test over weekend.

[Quozl]

Q4D18 memtest

• completed SKU201, 1184 cycles, over 24-hours,
• completed SKU200X3, over 24-hours, but evidence lost, need to retest.

[Quozl]

Runin test of Q4D17 and 11.3.1 passed 72-hours fine, completed.

[greenfeld]

Is this ticket safe to close?

[Quozl]

Thinks so. I've not seen problems yet with it.