Am 29.10.2025 um 18:23 hat Kevin Wolf geschrieben:
> Am 28.10.2025 um 17:33 hat Hanna Czenczek geschrieben:
> > nvme_process_completion() must run in the main BDS context, so schedule
> > a BH for requests that aren’t there.
> >
> > The context in which we kick does not matter, but let’s just keep kick
> > and process_completion together for simplicity’s sake.
>
> Ok, fair, move the main BDS context for calling these functions. But
> doesn't that mean that we need to move back to the request context for
> calling the callback?
>
> In particular, I see this:
>
> static void nvme_rw_cb_bh(void *opaque)
> {
> NVMeCoData *data = opaque;
> qemu_coroutine_enter(data->co);
> }
>
> The next patch changes some things about coroutine wakeup, but it
> doesn't touch this qemu_coroutine_enter(). So I think the coroutine is
> now running in the wrong thread.
It actually isn't because the patch changes in which AioContext the BH
is called. Quite confusing with all the indirections. Let's get rid of
the BH with qemu_coroutine_enter() and just call aio_co_wake() directly.
Kevin
> I also feel that it gets a bit confusing what is running in which
> context, so maybe we can add comments to each of the callbacks telling
> that they are running in main BDS context or request coroutine context.
>
> > (For what it’s worth, a quick fio bandwidth test indicates that on my
> > test hardware, if anything, this may be a bit better than kicking
> > immediately before scheduling a pure nvme_process_completion() BH. But
> > I wouldn’t take more from those results than that it doesn’t really seem
> > to matter either way.)
> >
> > Cc: [email protected]
> > Signed-off-by: Hanna Czenczek <[email protected]>
>
> Kevin
