Reading /dev/mem where the GPU device is located reboots the system in Xavier

Autonomous Machines Jetson & Embedded Systems Jetson AGX Xavier
1

I am trying to read /dev/mem using the busybox toolset. My goal is to read some registers related to the integrated GPU. By looking at /proc/iomem, it is easy to understand that the gpu device is mapped in physical memory at the address 0x17000000. This is a chunk of my /proc/iomem:

...
15f00000-15ffffff : /host1x/vi@15c10000
16000000-167fffff : /host1x/pva0
16800000-16ffffff : /host1x/pva1
17000000-17ffffff : /gv11b                 <--- HERE
18000000-18ffffff : /gv11b
30000000-3003ffff : config
30040000-3007ffff : atu_dma
34000000-3403ffff : config
34040000-3407ffff : atu_dma
38000000-3803ffff : config
...

If you try to read with busybox at this address withe the command:

busybox devmem 0x17000000

Then the system hangs for a while and then it reboots.

Repeating this experiment with the Jetson TX2 does not produce any issue.

Do you have any clue about this?

2

Its mostly because, architecture and design is different.
hang and reboot happens due to no clock or power to the controller or no privilege to ccplex EL level.

3

Thank you,

The reason was, like you said, is no power to the controller because of GPU railgating. If you try to read the memory while a GPU kernel is running, then it is fine.

Another option is to disable the GPU railgate:

echo "0" > /sys/bus/platform/devices/17000000.gv11b/driver/17000000.gv11b/railgate_enable

and it works fine.