DMA appears to be stuck after GPU warm reboot issue in Guest on i.MX8QM board.

[oleksiimoisieiev]

Hello.

I'm currently building Xen based system on IMX8QM board.

In my configuration I have gpu core 1 in Dom0 which includes most of the hardware, and passed-through gpu core 0 to the Guest Domain (DomU). GPU cores can be split between the guests, according to the Section 8.5 of i.MX_Graphics_User's_Guide_Linux 

Everything works fine on the first boot. But after rebooting DomU I see the following error on Galcore initialization:

[ 1.647193] **************************
[ 1.647210] *** GPU STATE DUMP ***
[ 1.647225] **************************
[ 1.647238] axi = 0x0000025F
[ 1.647252] idle = 0x7FFFFFFE
[ 1.647265] FE not idle
[ 1.647276] DMA appears to be stuck at this address:
[ 1.647294] 0x003E47E0

Some investigation gave me that reset_gpu function does nothing because there is no registered reset control. Not sure if there is the cause.

I'm using commit a68e31b63f864ff71cd4adb40fbc9e1edc75c250 from linux-imx 

Bootlog is attached. It includes first successful start and unsuccessful restart.

Also attaching kernel config.

Maybe somebody have met this issue before.

[oleksiimoisieiev]

We managed to solve an issue. It appears to be connected with memory management. Memory wasn't cleared in the right way after destroying guest, so Galcore firmware was left in undefined state.

[mayqueen]

Any patches? I have same issue now, thanks