Re: [PATCH 1/2] io_uring/cmd: split io_uring_cmd_set_res() from io_uring_cmd_done()

[Ming Lei]

On Thu, Aug 27, 2026 at 4:44 PM Caleb Sander Mateos
<csander@xxxxxxxxxxxxxxx> wrote:
>
> On Thu, Aug 27, 2026 at 1:05 PM Joanne Koong <joannelkoong@xxxxxxxxx> wrote:
> >
> > Hi Caleb,
> >
> > On Thu, Aug 27, 2026 at 11:57 AM Caleb Sander Mateos
> > <csander@xxxxxxxxxxxxxxx> wrote:
> > >
> > > In preparation for setting the io_uring NVMe passthru CQE results from
> > > the blk-mq request completion rather than the task work callback, split
> > > out functions io_uring_cmd_set_res{,32}() from io_uring_cmd_done{,32}().
> > > This allows io_uring_cmd_done{,32}() and __io_uring_cmd_done() to be
> > > consolidated into a single CQE-size-agnostic function with 3 fewer
> > > arguments than __io_uring_cmd_done().
> > >
> >
> > The conversion looks logically correct to me but the interface feels a
> > bit annoying / fragile on the caller side.
> >
> > Is there a way nvme can get what it needs without changing the other
> > callers? Would it work to just keep the existing
> > io_uring_cmd_done{,32}() as is and add a separate API for special
> > callers like nvme, eg:
> >
> >   void io_uring_cmd_set_res32(struct io_uring_cmd *cmd, s32 ret, u64 res2);
> >   void __io_uring_cmd_done(struct io_uring_cmd *cmd, unsigned issue_flags);
> >
> > where io_uring_cmd_done{,32} would then just be inline wrappers over
> > those? afaict, then only the nvme callsite would have to change and
> > fuse, btrfs, block, ublk, and scsi_bsg could just stay as is. Do you
> > think something like that makes sense to do?

Yeah, this way is definitely much better from a driver viewpoint.



Thanks,
Ming Lei