[RFC 2/3] nvme: passthrough helper with callback

[Kanchan Joshi]

On Fri, Mar 5, 2021 at 9:44 AM Chaitanya Kulkarni
<Chaitanya.Kulkarni at wdc.com> wrote:
>
> On 3/4/21 03:14, Kanchan Joshi wrote:
> > On Wed, Mar 3, 2021 at 8:52 PM Chaitanya Kulkarni
> > <Chaitanya.Kulkarni at wdc.com> wrote:
> >> On 3/2/21 23:22, Kanchan Joshi wrote:
> >>> -void nvme_execute_passthru_rq(struct request *rq)
> >>> +void nvme_execute_passthru_rq_common(struct request *rq,
> >>> +                     rq_end_io_fn *done)
> >>>  {
> >>>       struct nvme_command *cmd = nvme_req(rq)->cmd;
> >>>       struct nvme_ctrl *ctrl = nvme_req(rq)->ctrl;
> >>> @@ -1135,9 +1136,17 @@ void nvme_execute_passthru_rq(struct request *rq)
> >>>       u32 effects;
> >>>
> >>>       effects = nvme_passthru_start(ctrl, ns, cmd->common.opcode);
> >>> -     blk_execute_rq(disk, rq, 0);
> >>> +     if (!done)
> >>> +             blk_execute_rq(disk, rq, 0);
> >>> +     else
> >>> +             blk_execute_rq_nowait(disk, rq, 0, done);
> >>>       nvme_passthru_end(ctrl, effects);
> >> This needs a detailed explanation in order to prove the correctness.
> > Do you see something wrong here?
> > blk_execute_rq() employs the same helper (i.e. nowait one) and uses
> > additional completion-variable to make it sync.
> >
>
> There is no gurantee that command will finish between call to
> blk_execute_rq_nowait()
> and nvme_passthru_end() is there ?

Got your concern now, thanks for elaborating.
I will move the passthru_end to when completion actually occurs, and
get it processed via workqueue.

And perhaps I should send a micro-optimization patch which will call
passthru_end only for non-zero command-effects.
Currently we do three checks.