Am 2021-12-03 um 2:18 a.m. schrieb Christian König:
> Well NAK.
>
> We already discussed this and decided to not use any hardware
> acceleration for the debug access.
Conclusions from our offline discussion for the record:
We need amdgpu_ttm_access_memory (e.g. gdb accessing VRAM) for
post-mortem
Well NAK.
We already discussed this and decided to not use any hardware
acceleration for the debug access.
Apart from that you implementation is absolutely horrible and won't work
in all cases.
Regards,
Christian.
Am 02.12.21 um 22:43 schrieb Jonathan Kim:
To support better memory access
To support better memory access performance on non-Large BAR devices, use
SDMA copies instead of MM access.
SDMA access is restricted to PAGE_SIZE'd access to account for the PTRACED
process memory r/w operation use case. Any other access size will use
MMIO.
Failure to do an SDMA copy will resul
