ok so the story goes like this:
back in august 2016 i got the A20 card up and running with mainline 4.7rc1 (or thereabouts), including patching in NAND support for a proper mtd device. it worked... up to a point... except when u-boot did a complete scan it CORRUPTED the NAND flash... just from reads.
i figured this was "just software" as it's pretty experimental, and the 3.4.104 sunxi-nand works perfectly.
fast-foward to a few months ago, when i was testing the external micro-sd slot, weirdly it resulted in a kernel segfault *from the sunxi-nand driver*. i thought, "that's weird, might be to do with current-fighting from the micro-desktop PCB not having level-shifting"
sooo... that's now fixed: the micro-sd slot works perfectly through the new revision 1.7 microdesktop's level-shifting IC. however i got exactly the same kernel segfault in the sunxi-nand driver, so i temporarily took it out of the kernel config, tested again, and yes, the micro-sd worked fine.
... except that when i booted it up again, the nand flash had been corrupted. now, bear in mind there's *NO DRIVER INSTALLED*.
this is just too weird for me to deal with. not to mention, because of the age of the A20's Boot ROM there is now a limited set of "legacy" NAND ICs available i've had it with them.
now, i _could_ convert to eMMC but it's too much of a major redesign: it involves disrupting the RGB/TTL tracks, and may require at least two more rounds of pre-production prototyping...
... it's too much: it's too risky... and i'm getting fed up. so. what i'm going to do instead is: cut the NAND IC entirely, then wire SDC2 (which is the same pins as the NAND) to the *ON-BOARD* Micro-SD card slot, instead.
what that will give is *two* bootable Micro-SD card options, priority being first the external one (MMC0) and second the on-board one (MMC2).
the reason why this will be possible when it was not possible before was described in another post to the list a couple weeks back: there's simply no way to route the tracks on such a small 6-layer PCB with the NAND tracks going from right-to-left, MMC0 has no room to go left-to-right and at the same time MMC3 going left-to-right. that's over 30 tracks trying to cross through space that will only take about 15.
BUT...
... if the NAND tracks are *removed*.... *NOW* i can route MMC2 (right-hand-side of the A20 pins) to the on-board MicroSD slot (right of the Card).
i can always do a variant which has eMMC later.... but right now i just want to get something out the door as soon as possible with as little risk as possible. this approach does have the advantage that the BoM is about $5 less.... but it does mean i will need to get something like... 1,000 MicroSD cards, instead!
l.
On Thu Apr 6 06:12:16 BST 2017, Luke Kenneth Casson Leighton wrote:
... it's too much: it's too risky... and i'm getting fed up. so. what i'm going to do instead is: cut the NAND IC entirely, then wire SDC2 (which is the same pins as the NAND) to the *ON-BOARD* Micro-SD card slot, instead.
what that will give is *two* bootable Micro-SD card options, priority being first the external one (MMC0) and second the on-board one (MMC2).
I can't comment on the technical issues but I think this is an improvement in terms of usability, at least from my experience using other systems that relied on booting from Micro-SD instead of from NAND flash.
David
On 04/06/2017 12:12 AM, Luke Kenneth Casson Leighton wrote:
now, i _could_ convert to eMMC but it's too much of a major redesign: it involves disrupting the RGB/TTL tracks, and may require at least two more rounds of pre-production prototyping...
... it's too much: it's too risky... and i'm getting fed up. so. what i'm going to do instead is: cut the NAND IC entirely, then wire SDC2 (which is the same pins as the NAND) to the *ON-BOARD* Micro-SD card slot, instead.
Not that I'm against dual SD slots, but from what I see, NAND and eMMC share the same pins so it should be easy enough to add the bga footprint and the required passives.
The MarsBoard A20 has pads for both NAND and eMMC in the same footprint so you can use one or the other without losing board space.
http://www.haoyuelectronics.com/service/A10-A20/Schematics/CM-A10&A20/
--- crowd-funded eco-conscious hardware: https://www.crowdsupply.com/eoma68
On Thu, Apr 6, 2017 at 1:38 PM, Joseph Honold mozzwald@gmail.com wrote:
On 04/06/2017 12:12 AM, Luke Kenneth Casson Leighton wrote:
now, i _could_ convert to eMMC but it's too much of a major redesign: it involves disrupting the RGB/TTL tracks, and may require at least two more rounds of pre-production prototyping...
... it's too much: it's too risky... and i'm getting fed up. so. what i'm going to do instead is: cut the NAND IC entirely, then wire SDC2 (which is the same pins as the NAND) to the *ON-BOARD* Micro-SD card slot, instead.
Not that I'm against dual SD slots, but from what I see, NAND and eMMC share the same pins so it should be easy enough to add the bga footprint and the required passives.
should... but (a) if i get it wrong it's yet another $1500-$2000 and yet another 8 week delay, which means that it could be $3000-$4000 and 16 weeks because it will be necessary to do *two* more revisions, one to find out that the PCB design was wrong and one to add corrections
and (b) yes the same wires can be used but no the boot order cannot be changed because as it's the exact same wires there's still no room to cross over MMC3 and MMC0 to make MMC0 the "on-board" boot and MMC3 the EOMA68 off-board microsd.
the other alternative is to turn the A20 round by 90 degrees and to use two DDR3x16 RAM ICs instead of four DDR3x8 RAM ICs.
howevver...
(a) the cost of 2x DDR3x16 RAM ICs to make up 2GB is a COMPLETELY INSANE $20 just for the memory alone.
(b) replacing the memory layout and adjusting the power management is basically a total redesign of the board, i might as well chuck the ENTIRE design away and start completely from scratch.
i estimate that would take about three revisions, thus would be somewhere between $4,500-$6,000 and take an estimated 10-12 weeks.
and it'll be a $40 PCB. which there isn't enough cash for.
basic logical reasoning says: remove the NAND IC.
l.
--- crowd-funded eco-conscious hardware: https://www.crowdsupply.com/eoma68
On Thu, Apr 6, 2017 at 5:38 PM, Vincent Legoll vincent.legoll@gmail.com wrote:
basic logical reasoning says: remove the NAND IC.
Yes, stop the madness, make it work as-is/as-you-can, and keep the fancy for V2 :-)
Half-joke aside, I'm with you, get it out the door and take a break.
If I undestand, the V2 may not even be A20-based...
slight misunderstanding: i'm working on a second card that happens to have an RK3288 processor, named EOMA68-RK3288, current revision 0.1. that cannot be called "V2" i.e. is totally separate and distinct from the card named EOMA68-A20, current revision 2.7.4
l.
On 04/06/2017 07:05 PM, Luke Kenneth Casson Leighton wrote:
crowd-funded eco-conscious hardware: https://www.crowdsupply.com/eoma68
On Thu, Apr 6, 2017 at 5:38 PM, Vincent Legoll vincent.legoll@gmail.com wrote:
basic logical reasoning says: remove the NAND IC.
Yes, stop the madness, make it work as-is/as-you-can, and keep the fancy for V2 :-)
Half-joke aside, I'm with you, get it out the door and take a break.
If I undestand, the V2 may not even be A20-based...
slight misunderstanding: i'm working on a second card that happens to have an RK3288 processor, named EOMA68-RK3288, current revision 0.1. that cannot be called "V2" i.e. is totally separate and distinct from the card named EOMA68-A20, current revision 2.7.4
I am sure you will succeed in reverse engineering all the blobs out. and once you do, I will most definitely buy one and laptop casing. and if you want I will donate an extra 30$ because I didn't buy the original a20.
He does indeed have a point though, once you get it out the door take a break. You deserve it. :)
l.
arm-netbook mailing list arm-netbook@lists.phcomp.co.uk http://lists.phcomp.co.uk/mailman/listinfo/arm-netbook Send large attachments to arm-netbook@files.phcomp.co.uk
arm-netbook@lists.phcomp.co.uk