Brendan Cunningham
Brendan Cunningham
> Thanks for the clarification, I discussed this with the internal team. Unfortunately, we do not provide support for this. From our point of view, `hfi1` needs to have full...
> Sorry for the delay, I've gotten some answers from the internal team. > > > Are all ROCm-VA:GPU-page mappings mapped into the process virtual address space? If so, can...
Our reproducer is `osu_bibw` built from the OSU Microbenchmarks 7.4 source code available [here](https://mvapich.cse.ohio-state.edu/benchmarks/). I am 99% certain that the only ROCm calls in the benchmark code are HIP runtime...
Yes, [hfi1/pin_amd.c `amd_node_mmu_register()`](https://github.com/cornelisnetworks/opa-distro-drivers/blob/5e315a00e9aad3a5dc5d008f1a66c1f98b838a0c/drivers/infiniband/hw/hfi1/pin_amd.c#L142) is where our driver subscribes with `mmu_interval_notifier` for notifications about the ROCm VA range it was given.
This is satisfactory for us then. Thanks for your help.
Two other things I didn't note in my initial report: 1. I observed these problems on a 6.5.0 development kernel built with `CONFIG_DEBUG_KMEMLEAK=y`; the driver code I pointed you to...
No, I haven't run a longer test yet.