Re: [Qemu-devel] [PATCH 1/2] vhost-user-blk: introduce a new vhost-user-blk host device

[Liu, Changpeng]

> -----Original Message-----
> From: Stefan Hajnoczi [mailto:address@hidden
> Sent: Thursday, July 27, 2017 5:49 PM
> To: Liu, Changpeng <address@hidden>
> Cc: address@hidden; address@hidden; address@hidden;
> address@hidden
> Subject: Re: [PATCH 1/2] vhost-user-blk: introduce a new vhost-user-blk host
> device
> 
> On Thu, Jul 27, 2017 at 12:29:45AM +0000, Liu, Changpeng wrote:
> > > > +    .fields = (VMStateField[]) {
> > > > +        VMSTATE_VIRTIO_DEVICE,
> > > > +        VMSTATE_END_OF_LIST()
> > > > +    },
> > > > +};
> > > > +
> > > > +static Property vhost_user_blk_properties[] = {
> > > > +    DEFINE_PROP_CHR("chardev", VHostUserBlk, chardev),
> > > > +    DEFINE_BLOCK_PROPERTIES(VHostUserBlk, blkcfg),
> > >
> > > DEFINE_BLOCK_PROPERTIES cannot be used directly because it includes the
> > > 'drive' (blkcfg.blk) parameter.  The command-line should not allow a
> > > drive= parameter.
> > >
> > > Also, parameters like the discard granularity, optimal I/O size, logical
> > > block size, etc can be initialized to sensible defaults by QEMU's block
> > > layer when drive= is used.  Since you are bypassing QEMU's block layer
> > > there is no way for QEMU to set good defaults.
> > >
> > > Are you relying on the managment tools populating these parameters with
> > > the correct values from the vhost-user-blk process?  Or did you have
> > > some other mechanism in mind?
> > Actually for this part and your comments about block resize, I think maybe 
> > add
> several
> > additional vhost user messages such like:
> VHOST_USER_GET_BLK_CFG/VHOST_USER_SET_BLK_CFG
> > makes the code more clear, I'm okay to add such messages.
> 
> New messages for virtio config space read/write might be useful for
> other devices besides virtio-blk.
I mean all the block device information like: block_size/capacity are stored in 
another process, 
so add a new vhost user message to Qemu vhost-user-blk can get those 
information when Qemu get
started, once those messages stored to virtio_pci config space, we will not 
change it.
> 
> It's worth checking how virtio config space is live migrated and how to
> do that consistently if the vhost-user process is involved.
Virtio will config spaces for virtio_blk, and even the config space migrated to 
another machine, it should be
same with the another I/O process, did I get your comments ? Thanks. 
> 
> Stefan