AIE-ML Programming & optimization give me a 2GB (over 12GB) RAM linux on vek280

Hello,

I am contacting you because I try to build & run the following tutorial on vek280: https://github.com/Xilinx/Vitis-Tutorials/tree/2025.1/AI_Engine_Development/AIE-ML/Design_Tutorials/01-AIE-ML-programming-and-optimization

The project compiles well but when I run it on the board, I only have 2GB of memory available on Linux.

Thus, I compared my boot image with the AMD-EDF prebuilt WIC image for vek280 provided here: https://www.xilinx.com/member/forms/download/xef.html?filename=edf-linux-disk-image-amd-cortexa72-common%20versal-vek280-sdt-seg.rootfs-07292043.wic.xz

And this image has 12GB of RAM so it seems that the AIE/PL/HOST process provided in the example is failing at something.

We can see in boot logs that in amd-edf image, there is only 300MB reserved, but on our image there is 10GB reserved.

AMD-EDF BOOT:
[ 0.137708] Memory: 11214824K/12582912K available (16960K kernel code, 1088K rwdata, 4976K rodata, 3136K init, 484K bss, 339448K reserved, 1024000K cma-reserved)

OUR BOOT:
[ 5.157516] Memory: 1271296K/12582912K available (16960K kernel code, 1088K rwdata, 4976K rodata, 3136K init, 484K bss, 10783192K reserved, 524288K cma-reserved)

The device tree seems to be involved in this

AMD-EDF BOOT:
[ 0.000000] Zone ranges:
[ 0.000000] DMA [mem 0x0000000000000000-0x00000000ffffffff]
[ 0.000000] DMA32 empty
[ 0.000000] Normal [mem 0x0000000100000000-0x00000600ffffffff]
[ 0.000000] Movable zone start for each node
[ 0.000000] Early memory node ranges
[ 0.000000] node 0: [mem 0x0000000000000000-0x000000007fffffff]
[ 0.000000] node 0: [mem 0x0000000800000000-0x000000087fffffff]
[ 0.000000] node 0: [mem 0x0000050000000000-0x00000500ffffffff]
[ 0.000000] node 0: [mem 0x0000060000000000-0x00000600ffffffff]
[ 0.000000] Initmem setup node 0 [mem 0x0000000000000000-0x00000600ffffffff]
[ 0.000000] cma: Reserved 1000 MiB at 0x0000000039600000 on node -1

OUR BOOT:
[ 0.000000] OF: reserved mem: 0x0000000800000000..0x000000087fffffff (2097152 KiB) nomap non-reusable buffer@0
[ 0.000000] OF: reserved mem: 0x0000050000000000..0x00000501ffffffff (8388608 KiB) nomap non-reusable buffer@1
[ 0.000000] Zone ranges:
[ 0.000000] DMA [mem 0x0000000000000000-0x00000000ffffffff]
[ 0.000000] DMA32 empty
[ 0.000000] Normal [mem 0x0000000100000000-0x00000501ffffffff]
[ 0.000000] Movable zone start for each node
[ 0.000000] Early memory node ranges
[ 0.000000] node 0: [mem 0x0000000000000000-0x000000007fffffff]
[ 0.000000] node 0: [mem 0x0000000800000000-0x000000087fffffff]
[ 0.000000] node 0: [mem 0x0000050000000000-0x00000501ffffffff]
[ 0.000000] Initmem setup node 0 [mem 0x0000000000000000-0x00000501ffffffff]
[ 0.000000] cma: Reserved 512 MiB at 0x000000004be00000 on node -1

Would you have any idea of what should be modified in the tutorial so that tutorial correctly boots with 12GB of RAM ?

(This is the default behavior of the tutorial on our side)

Start with the AI_Engine_Development/AIE-ML/Design_Tutorials/01-AIE-ML-programming-and-optimization tutorial and compare its boot image and device-tree configuration with the AMD-EDF VEK280 image. Use the provided boot logs to trace the reserved memory regions. Done means the tutorial boots on the VEK280 with approximately 12GB of RAM available to Linux.