On 4/28/23 00:33, Christophe JAILLET wrote: > Le 28/04/2023 à 01:12, Chaitanya Kulkarni a écrit : >> (+Keith, for host side) >>> --- >>> More aggressive grouping could be done to be more future proof, but the >>> way the struct nvmet_ns is written suggest that some fields should be >>> kept together. So keep grouping as-is. >>> >>> >> >> you can send RFC and I'll be happy to take a look if it's going >> have any benefit, it will take some time though.. >> >> for host side :- >> >> while you are at it, it might be useful to take a look at the structures >> that are accessed in the fast path on the host side ? > > Ok, why not, but can you help identifying these structures or places > considered as fast path? > > CJ you can start with nvme_ns/nvme_ctrl as I remember nvme_ns is used in nvme_setup_rw_cmd() on host and nvme_ctrl is used in nvmet_passthru_execute_cmd(). In general nvme_queue_rq()/nvme_rdma_queue_rq()/ nvmet_bdev_execute_rw()/nvmet_file_execute_rw()/ nvmet_passthru_execute_cmd() are functions to start with, then you can see structs starting with nvme_ prefix (mainly related to ns and ctrl) which are worth taking a look and their benefits, but be careful what you are moving ... Ohh and nvmet_req that is also ... -ck
