Re: [PATCH] vfio/mdev: Check globally for duplicate devices

[Alex Williamson]

On Wed, 16 May 2018 11:05:06 +0200
Cornelia Huck <cohuck@xxxxxxxxxx> wrote:

> On Tue, 15 May 2018 14:17:04 -0600
> Alex Williamson <alex.williamson@xxxxxxxxxx> wrote:
> 
> > When we create an mdev device, we check for duplicates against the
> > parent device and return -EEXIST if found, but the mdev device
> > namespace is global since we'll link all devices from the bus.  We do
> > catch this later in sysfs_do_create_link_sd() to return -EEXIST, but
> > with it comes a kernel warning and stack trace for trying to create
> > duplicate sysfs links, which makes it an undesirable response.
> > 
> > Therefore we should really be looking for duplicates across all mdev
> > parent devices, or as implemented here, against our mdev device list.
> > 
> > Notably, mdev_parent.lock really only seems to be serializing device
> > creation and removal per parent.  I'm not sure if this is necessary,
> > mdev vendor drivers could easily provide this serialization if it
> > is required, but a side-effect of holding the mdev_list_lock to
> > protect the namespace is actually greater serialization across the
> > create and remove paths, so mdev_parent.lock is removed.  If we can
> > show that vendor drivers handle the create/remove paths themselves,
> > perhaps we can refine the locking granularity.  
> 
> I'm not sure whether more locking granularity on the create/remove
> paths is really worth the effort.

Perhaps not, but I thought I should at least mention it as a
consideration.

> > Signed-off-by: Alex Williamson <alex.williamson@xxxxxxxxxx>
> > ---
> >  drivers/vfio/mdev/mdev_core.c    |   79 ++++++++++----------------------------
> >  drivers/vfio/mdev/mdev_private.h |    1 
> >  2 files changed, 20 insertions(+), 60 deletions(-)  
> 
> In general, I think this patch makes sense; some nits below.
> 
> > 
> > diff --git a/drivers/vfio/mdev/mdev_core.c b/drivers/vfio/mdev/mdev_core.c
> > index 126991046eb7..3d8898a2baaf 100644
> > --- a/drivers/vfio/mdev/mdev_core.c
> > +++ b/drivers/vfio/mdev/mdev_core.c  
> 
> > @@ -376,12 +346,13 @@ int mdev_device_remove(struct device *dev, bool force_remove)
> >  	struct mdev_device *mdev, *tmp;
> >  	struct mdev_parent *parent;
> >  	struct mdev_type *type;
> > -	int ret;
> > +	int ret = 0;  
> 
> I don't think you need to init this, as ret should either be set to
> -ENODEV or the return code of mdev_device_remove_ops(), shouldn't it?

Yep, I think this is a leftover from before I decided I should goto a
common out, it's unnecessary now.  Removed.
 
> >  	bool found = false;
> >  
> >  	mdev = to_mdev_device(dev);
> >  
> >  	mutex_lock(&mdev_list_lock);
> > +  
> 
> unrelated whitespace change

Intentional, previously mdev_list_lock was only protecting this
sub-section of the code, so the lock was tucked up to it.  Now we're
holding the lock across the whole function so I wanted to make it
separate.
 
> >  	list_for_each_entry(tmp, &mdev_list, next) {
> >  		if (tmp == mdev) {
> >  			found = true;
> > @@ -389,35 +360,25 @@ int mdev_device_remove(struct device *dev, bool force_remove)
> >  		}
> >  	}
> >  
> > -	if (found)
> > -		list_del(&mdev->next);
> > -
> > -	mutex_unlock(&mdev_list_lock);
> > -
> > -	if (!found)
> > -		return -ENODEV;
> > +	if (!found) {
> > +		ret = -ENODEV;
> > +		goto out;
> > +	}
> >  
> >  	type = to_mdev_type(mdev->type_kobj);
> >  	parent = mdev->parent;
> > -	mutex_lock(&parent->lock);
> >  
> >  	ret = mdev_device_remove_ops(mdev, force_remove);
> > -	if (ret) {
> > -		mutex_unlock(&parent->lock);
> > -
> > -		mutex_lock(&mdev_list_lock);
> > -		list_add(&mdev->next, &mdev_list);
> > -		mutex_unlock(&mdev_list_lock);
> > -
> > -		return ret;
> > -	}
> > +	if (ret)
> > +		goto out;  
> 
> This change really simplyfies the code, nice.

Agreed, thanks for the review!

Alex