Hi,
ED sent me a 4G board too, so here are results with mine: DLL disable: MLO hangs (5+ tries). All further tests done with DLL on (default).
DDR3_1.35V+ARM_1.5GHz - boots without failure or errors (3 tries), seems to be stable while idle (~5minutes at least). Hangs with memtester in 1-2 minutes (screen started displaying repeating garbage suggesting RAM reads are failing
further memterster tests (running on both cores): DDR3_1.5V+ARM_1.5GHz - thermal shutdown in ~1 minute [1] DDR3_1.5V+ARM_1.3GHz - thermal shutdown DDR3_1.5V+ARM_1.2GHz - pass (90+ minutes stable) DDR3_1.4V+ARM_1.2GHz - 60+ minutes stable DDR3_1.37V+ARM_1.2GHz - 100+ minutes stable
Haven't noticed the "hot RAM" problem Tony reported - can poweroff and then start fine (no reboot because of unrelated and known ABE issue "clock: dpll_abe_ck failed transition to 'locked'").
While running these tests, charger LED starts blinking. Reading charger's REG09 returns 0x01 - "Hot temperature". Indeed after taking out the battery it's too hot to hold in hand.
[1]: dmesg: thermal thermal_zone0: critical temperature reached(125 C),shutting down Seems like throttling code fails to keep the temperature under control.
So not much new here, just another data point. Looks like my board is very close to working as intended.
Grazvydas
On 10/20/2017 04:54 PM, H. Nikolaus Schaller wrote:
A suggested by Michael I did some more tests with the 4GB Samsung DDR3 board (C48):
- 1.35V did boot boot sometimes up to login: but then console is stuck without other indications
sometimes it was stuck right after Starting Kernel... Once I got this error (never seen before): [ 14.156466] unwind: Unknown symbol address c0200000 [ 14.161585] unwind: Index not found c0200000 [ 14.172344] unwind: Unknown symbol address c0200000 [ 14.177458] unwind: Index not found c0200000
1.50V boots fine and passed memtester 3GB without problems
1.40V wasn't able to boot in 4 attempts - did hang after 3 - 20 seconds kernel log
1.45V boots fine and passed memtester 3GB without problems
So all this shows how tight the margin between flaky and working is...
Now the next step was to try to improve the wiring on a board with Samsung DDR3L chips.
But when trying to boot his morning, it worked well and without flaws at 1.40V and passed memtester 3GB as well. And it simply was stuck at 1.35V before I could get to a login: I have no real idea what the difference is - maybe battery voltage. If battery gets drained the Palmas has to do a different VBatt / VDD-DDR3L translation ratio and that might change noise - even noise from the OMAP VDD-CORE.
So I simply have no board where I can compare "before adding wires" and "after adding wires" and clearly see a difference.
To summarise the 4GB boards (*):
- one IM 4GB works without problems at 1.35V (or more - not completely tested)
- one IM 4GB makes problems, less at 1.5, more at 1.35V - so it might be an issue of this individual board
- one Samsung 4GB works without problems at 1.45/1.50V but flaky at 1.35V/1.40V
- one Samsung 4GB (@Tony) works only with 1.5V and modified EMIF settings like DLL off and is also temperature dependent
From all this I would conclude that there is NO significant difference between Samsung and IMG chips... Maybe, iMG are a little more robust.
So it must be the power supply of the DDR3L chips and potentially the supply of the EMIF.
Therefore I think we can now close the analysis phase and have to rework the PCB layout. And then ask Juliano to simulate it again. And hope that it solves all problems. Even with Samsung chips.
BR, Nikolaus
(*) note that OMAP5 was limited to 1.0GHz and I wasn't able to boot it well at 1.7GHz. _______________________________________________ Kernel mailing list Kernel@pyra-handheld.com http://pyra-handheld.com/cgi-bin/mailman/listinfo/kernel