Re: [PATCH] virtio-blk: Add a VIRTIO_BLK_T_OUT_FUA request type

On Wed, May 7, 2025 at 8:40 PM Daniel Verkamp <[email protected]> wrote:
> On Wed, May 7, 2025 at 8:30 AM Alberto Faria <[email protected]> wrote:
> >
> > This is a variant of the VIRTIO_BLK_T_OUT request type that further
> > ensures the write becomes stable once the request completes, commonly
> > known as a Force Unit Access (FUA) request.
> >
> > Also add a VIRTIO_BLK_F_OUT_FUA feature bit signaling support for this
> > request type.
>
> How does this interact with the writeback field? Presumably the value
> of writeback is ignored by FUA commands, but it would be good to make
> that explicit.

That was the intention, yes. I'll make it explicit.

> Also, the device requirements "write is considered volatile when it is
> submitted"/"write becomes stable" section seems like it should be
> updated to include FUA requests.

I added a fourth item to the list that follows those words:

    +\item\label{item:flush4} the write was performed using a
VIRTIO_BLK_T_OUT_FUA
    +  request.

That felt the cleanest to me.

> [...]
> >
> > +Drivers SHOULD NOT negotiate VIRTIO_BLK_F_OUT_FUA if they are incapable of
> > +sending VIRTIO_BLK_T_OUT_FUA commands.
>
> What harm would be caused if a driver accepts the FUA feature bit but
> never sends a FUA command?

There's a similar sentence for VIRTIO_BLK_F_FLUSH, so I followed the
precedent, but I don't think there's any need for it for FUA. I'll
drop it.

Thanks,
Alberto

> Thanks,
> -- Daniel
>