Date:   Wed, 12 Jan 2022 10:02:03 -0500
From:   "Michael S. Tsirkin" <mst@redhat.com>
To:     =?utf-8?B?5p2O5Lic5Y2H?= <lidongsheng@dayudpu.com>
Cc:     jasowang <jasowang@redhat.com>,
        virtualization <virtualization@lists.linux-foundation.org>,
        linux-kernel <linux-kernel@vger.kernel.org>
Subject: Re: [PATCH] virtio_vdpa: Support surprise removal of virtio vdpa
 device
Message-ID: <20220112100006-mutt-send-email-mst@kernel.org>
References: <tencent_7A45E1E967F13AE14B061269@qq.com>
 <20220111065033-mutt-send-email-mst@kernel.org>
 <CACGkMEuAoSwore14qnuMDgdEtWh-UOJf1=oR9vhPMff8hoEZEQ@mail.gmail.com>
 <20220112013018-mutt-send-email-mst@kernel.org>
 <tencent_18F8A88D22E327230EF313BE@qq.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
Content-Transfer-Encoding: 8bit
In-Reply-To: <tencent_18F8A88D22E327230EF313BE@qq.com>
Precedence: bulk


On Wed, Jan 12, 2022 at 04:44:08PM +0800, 李东升 wrote:
> Indeed, there are hidden dangers for storage devices.
> So maybe we can break the device before virtio_net unregister_netdev and before
> virtio_blk reset
> 
> Unlike pci device, there is a manufacturer-driven existence between virtio vdpa
> and hardware devices, 
> unless the manufacturer provides a state interface, virtio vdpa and
> higher-level drivers cannot obtain the real hardware state.

I think for surprise removal you need to propagate the presence status up.
No way around that imho.

> 
> ------------------ Original ------------------
> From:  "mst"<mst@redhat.com>;
> Date:  Wed, Jan 12, 2022 02:30 PM
> To:  "jasowang"<jasowang@redhat.com>;
> Cc:  "李东升"<lidongsheng@dayudpu.com>; "virtualization"
> <virtualization@lists.linux-foundation.org>; "linux-kernel"
> <linux-kernel@vger.kernel.org>;
> Subject:  Re: [PATCH] virtio_vdpa: Support surprise removal of virtio vdpa
> device
>  
> On Wed, Jan 12, 2022 at 10:23:07AM +0800, Jason Wang wrote:
> > On Tue, Jan 11, 2022 at 7:52 PM Michael S. Tsirkin <mst@redhat.com> wrote:
> > >
> > > On Tue, Jan 11, 2022 at 11:36:42AM +0800, 李东升 wrote:
> > > > When virtio vdpa device removed, the abnormal damage of the device cannot
> be
> > > > perceived normally, which will cause problems similar to:
> > > >
> > > > 43bb40c5b926
> > >
> > >
> > > Should include the subject of the patch too.
> > >
> > > > Hence, add the ability to abort the command on surprise removal
> > > >
> > > > Signed-off-by: dongsheng li <lidongsheng@dayudpu.com>
> > >
> > > When removing gracefully,
> > > I am not sure we should break device unconditionally like this
> > > before giving drivers a chance to clean up.
> > > Should we just do it for surprise removal?
> >
> > That requires a new method to query whether it's a surprise removal.
> >
> > Thanks
> 
> We can check pci_device_is_present like virtio does.
> 
> > >
> > > > ---
> > > >  drivers/virtio/virtio_vdpa.c | 1 +
> > > >  1 file changed, 1 insertion(+)
> > > >
> > > > diff --git a/drivers/virtio/virtio_vdpa.c b/drivers/virtio/virtio_vdpa.c
> > > > index 4a9ddb44b2a7..fd930409d190 100644
> > > > --- a/drivers/virtio/virtio_vdpa.c
> > > > +++ b/drivers/virtio/virtio_vdpa.c
> > > > @@ -374,6 +374,7 @@ static void virtio_vdpa_remove(struct vdpa_device
> *vdpa)
> > > >  {
> > > >   struct virtio_vdpa_device *vd_dev = vdpa_get_drvdata(vdpa);
> > > >
> > > > + virtio_break_device(vd_dev->vdev);
> > > >   unregister_virtio_device(&vd_dev->vdev);
> > > >  }
> > > >
> > > > --
> > > > 2.17.1
> > >
>