01:44:38 <arekinath> hanging at alloc_pd is weird for that though, it's a fair way into attach
01:44:45 <arekinath> lots of commands have to have succeeded to get there
01:46:36 <arekinath> if you're after my changes to make the cmdq interrupt-driven, it _is_ the first thing we do after disabling polling though
01:46:59 <arekinath> so my guess would be that the interrupt isn't happening
01:49:36 <arekinath> 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)