hanging at alloc_pd is weird for that though, it's a fair way into attach

lots of commands have to have succeeded to get there

if you're after my changes to make the cmdq interrupt-driven, it _is_ the first thing we do after disabling polling though

so my guess would be that the interrupt isn't happening

I would go look at the cmdq (in mlx->mlx_cmd->mcmd_dma.mxdb_va, follow that and print it as an array of mlxcx_cmd_ent_t, look and see if the cmd has completed and we haven't woken up)