[Bug]: Total Ram detection in XF40H less than og stock fw reported.

[Luizardoh]

Pre-submission checklist

• I am using the latest version of the project
• I have searched existing issues to make sure this is not a duplicate
• I have provided a screenshot from the DTB detection tool at https://lcdyk0517.github.io/dtbTools.html
• I have described which DTB I selected in the issue

Firmware Type

arkos4clone

Version

ArkOS4Clone-08152026

DTB File

rk3326-xf40h-linux.dtb

DTB Detection Screenshot

Console Output / Error

Bug Description

I bought an XF40H knowing I could put ArkOS4Clone on it, because I use PortMaster a lot and I don't like EmuELEC. However, I was surprised to see that the console reports quite a bit less RAM than when it runs with EmuELEC (whether the one on the eMMC it shipped with, or a second SD I flashed with its same stock image). In ArkOS it reports 611MB, with about 400 available to use, while the stock CFW reports a maximum of 975MB of RAM, with 321 free while in RetroArch (I know that firmware isn't the best at resource optimization anyway). But I was wondering what this difference in detection is due to.

For reference, the stock CFW runs kernel 5.11, while ArkOS uses kernel 4.4 — in case the kernel version matters at all for this.

I've done some tests but none have panned out:

I've seen that no DTB (arkos xf40h, emuelec rf3640g4ka, odroidgo2) declares a /memory node → the size is set by u-boot, not the dtb.
No mem= in bootargs; ARM64 uses the tree's /memory node (it doesn't add RAM via mem=).
I compared the boot regions byte by byte: the DDR firmware differs — ArkOS DDR fwver v2.12 vs stock DDR V2.08 20220817, same 333 MHz training clock; u-boot builds differ (arkos4clone 2026 vs pangjianwen 2025).
Test A — swapped the full stock boot region: it didn't boot ArkOS (my mistake: I didn't account for the partition table being rewritten + the stock u-boot uses extlinux, not boot.ini). Reverted.
Test B — spliced the DDR bin v2.08 into ArkOS's idbloader: it booted, but RAM stayed at 661, so the DDR blob at 0x8800 isn't the deciding factor.
Test C — full stock u-boot + an extlinux.conf pointing at the ArkOS kernel: booted ArkOS, but RAM still showed 661/758 MB physical → the SD bootloader isn't the deciding factor.
I analyzed reserved-memory, chosen, and dmc in the DTBs: only ~1 MB of reservations, none affecting the total.
Test max_ddrfreq=333: no change (rules out frequency).
As a loose idea I'm not sure about: maybe the total RAM is decided by u-boot's DDR training during boot, and over SD it only ever trains ~758MB no matter which SD bootloader is used — while the full 1GB (975MB) only appears through the sealed eMMC boot path. If that were true, it would be something baked into the boot/branding rather than something fixable through the SD. I can't access the eMMC's bootloader to confirm, though.
P.D. — I've included the photos from EmuELEC instead ssh ss reports because EmuELEC doesn't allow SSH access in my case, neither on its stock eMMC version nor on the copy I flashed to an SD card. As a cross-check, RetroArch inside ArkOS reports the same RAM figure I obtained over SSH, so the measurement method doesn't skew the result.

Steps to Reproduce

Expected Behavior (Optional)

No response

Screenshots / Video / Logs (Optional)

RAM reported by ArkOS4Clone over SSH

==========

Retroarch in Emuelec reporting 975mb Ram

=======

Emuelec version in eMMC (same version in the another sd)

Additional Context (Optional)

No response

[lcdyk0517]

That's quite interesting – it looks like the xf40h also suffers from the fake 1GB RAM issue. Could you try this with the stock firmware: launch a game, open the RetroArch menu, navigate to Settings > User Interface > On‑Screen Notifications > Notifications Visibility, and enable Display Memory Usage. Then resume the game and check whether it reports 1GB. Also, if you're able to back up the original minimal firmware, I'd be happy to take a look at it.

[Luizardoh]

Yes, of course, here it is.

Fifa 2006 in gba, the retroarch overlay shows the same 975mb

What exactly do you mean by "original minimal firmware"? It came on my eMMC with version 9b0486_20251231and I don't know how to extract it from there. The repository only has up to ba8947_20250702, but it behaves the same way. Both versions report 975MB of RAM. Or are you referring to the binaries for the initial 16MB raw data that define the MBR and the uboot+ idbloader?

Here's the link to the emuelec image repository:

https://github.com/handheld520/XiFan-XF40H/releases

[lcdyk0517]

Do you mean that the image from the HandHeld520 repository shows 975M for you? Could you write a shell script that uses dd to dump the entire eMMC?

[Luizardoh]

Both images the stock one on the eMMC and the one from the HandHeld520 repo report those 975mb.
Regarding the script to dump from the eMMC, I'm not sure I can do it; maybe with AI help I can. But what you want is the whole image?Because for that, I feel that the HandHeld520 might be useful.

[lcdyk0517]

Try and see if aurknix is normal.https://github.com/[AveyondFly/distribution_rocknix](https://github.com/AveyondFly/distribution_rocknix)

[Luizardoh]

amazing, aurknix shows another value , 730mb of ram lol

[lcdyk0517]

Actually, no matter what is displayed, the available memory is always around 300 MB. More research should be done to see how it makes 750 MB of RAM show up as 975 MB.