Re: [Qemu-devel] [RFC v2 1/2] virtio: add pmem driver

[Stefan Hajnoczi <stefanha@redhat.com>]

[-- Attachment #1: Type: text/plain, Size: 1374 bytes --]

On Thu, Apr 26, 2018 at 11:44:59AM -0400, Pankaj Gupta wrote:
> > > +	int err;
> > > +
> > > +	sg_init_one(&sg, buf, sizeof(buf));
> > > +
> > > +	err = virtqueue_add_outbuf(vpmem->req_vq, &sg, 1, buf, GFP_KERNEL);
> > > +
> > > +	if (err) {
> > > +		dev_err(&vdev->dev, "failed to send command to virtio pmem device\n");
> > > +		return;
> > > +	}
> > > +
> > > +	virtqueue_kick(vpmem->req_vq);
> > 
> > Is any locking necessary?  Two CPUs must not invoke virtio_pmem_flush()
> > at the same time.  Not sure if anything guarantees this, maybe you're
> > relying on libnvdimm but I haven't checked.
> 
> I thought about it to some extent, and wanted to go ahead with simple version first:
> 
> - I think file 'inode -> locking' sill is there for request on single file.
> - For multiple files, our aim is to just flush the backend block image.
> - Even there is collision for virt queue read/write entry it should just trigger a Qemu fsync. 
>   We just want most recent flush to assure guest writes are synced properly.
> 
> Important point here: We are doing entire block fsync for guest virtual disk.

I don't understand your answer.  Is locking necessary or not?

From the virtqueue_add_outbuf() documentation:

 * Caller must ensure we don't call this with other virtqueue operations
 * at the same time (except where noted).

Stefan

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 455 bytes --]