[Intel-gfx] [PATCH v3] vfio: fix deadlock between group lock and kvm lock

[Alex Williamson]

On Fri, 3 Feb 2023 14:54:44 +0000
"Liu, Yi L" <yi.l.liu at intel.com> wrote:

> > From: Alex Williamson <alex.williamson at redhat.com>
> > Sent: Friday, February 3, 2023 9:50 PM
> > 
> > On Fri, 3 Feb 2023 13:32:09 +0000
> > "Liu, Yi L" <yi.l.liu at intel.com> wrote:
> >   
> > > > From: Tian, Kevin <kevin.tian at intel.com>
> > > > Sent: Friday, February 3, 2023 10:00 AM
> > > >  
> > > > > From: Alex Williamson <alex.williamson at redhat.com>
> > > > > Sent: Friday, February 3, 2023 7:13 AM
> > > > >
> > > > > On Thu, 2 Feb 2023 23:04:10 +0000
> > > > > "Tian, Kevin" <kevin.tian at intel.com> wrote:
> > > > >  
> > > > > > > From: Alex Williamson <alex.williamson at redhat.com>
> > > > > > > Sent: Friday, February 3, 2023 3:42 AM
> > > > > > >
> > > > > > >
> > > > > > > LGTM.  I'm not sure moving the functions to vfio_main really buys  
> > us  
> > > > > > > anything since we're making so much use of group fields.  The cdev
> > > > > > > approach will necessarily be different, so the bulk of the get code  
> > will  
> > > > > > > likely need to move back to group.c anyway.
> > > > > > >  
> > > > > >
> > > > > > well my last comment was based on Matthew's v2 where the get  
> > code  
> > > > > > gets a kvm passed in instead of implicitly retrieving group ref_lock
> > > > > > internally. In that case the get/put helpers only contain device logic
> > > > > > thus fit in vfio_main.c.
> > > > > >
> > > > > > with v3 then they have to be in group.c since we don't want to use
> > > > > > group fields in vfio_main.c.
> > > > > >
> > > > > > but I still think v2 of the helpers is slightly better. The only difference
> > > > > > between cdev and group when handling this race is using different
> > > > > > ref_lock. the symbol get/put part is exactly same. So even if we
> > > > > > merge v3 like this, very likely Yi has to change it back to v2 style
> > > > > > to share the get/put helpers while just leaving the ref_lock part
> > > > > > handled differently between the two path.  
> > > > >
> > > > > I'm not really a fan of the asymmetry of the v2 version where the get
> > > > > helper needs to be called under the new kvm_ref_lock, but the put
> > > > > helper does not.  Having the get helper handle that makes the caller
> > > > > much cleaner.  Thanks,
> > > > >  
> > > >
> > > > What about passing the lock pointer into the helper? it's still slightly
> > > > asymmetry as the put helper doesn't carry the lock pointer but it
> > > > could also be interpreted as if the pointer has been saved in the get
> > > > then if it needs to be referenced by the put there is no need to pass
> > > > it in again.  
> > >
> > > For cdev, I may modify vfio_device_get_kvm_safe() to accept
> > > struct kvm and let its caller hold a kvm_ref_lock (field within
> > > struct vfio_device_file). Meanwhile, the group path holds
> > > the group->kvm_ref_lock before invoking vfio_device_get_kvm_safe().
> > > vfio_device_get_kvm_safe() just includes the symbol get/put and
> > > the device->kvm and put_kvm set.  
> > 
> > Sounds a lot like v2 :-\   
> 
> Yes, like v2. 😊
> 
> > I'd look more towards group and cdev specific
> > helpers that handle the locking so that the callers aren't exposed to
> > the asymmetry of get vs put, and reduce a new
> > _vfio_device_get_kvm_safe() in common code that only does the symbol
> > work.  Thanks,  
> 
> If so, looks like Matthew needs a v4. I'm waiting for the final version
> of this patch and sending a new cdev series based on it. wish to see
> it soon ^_^.

cdev support is a future feature, why does it become a requirement for
a fix to the current base?  The refactoring could also happen in the
cdev series.  Thanks,

Alex