Linux Virtualization - Jul 2014 - [PATCH v2 3/4] virtio: rng: delay hwrng_register() till driver is ready

On Mon, Jul 21, 2014 at 05:15:51PM +0530, Amit Shah
wrote:
> Instead of calling hwrng_register() in the probe routing, call it in the
> scan routine.  This ensures that when hwrng_register() is successful,
> and it requests a few random bytes to seed the kernel's pool at init,
> we're ready to service that request.
> 
> This will also enable us to remove the workaround added previously to
> check whether probe was completed, and only then ask for data from the
> host.  The revert follows in the next commit.
> 
> There's a slight behaviour change here on unsuccessful
hwrng_register().
> Previously, when hwrng_unregister() failed, the probe() routine would
> fail, and the vqs would be torn down, and driver would be marked not
> initialized.  Now, the vqs will remain initialized, driver would be
> marked initialized as well, but won't be available in the list of RNGs
> available to hwrng core.  To fix the failures, the procedure remains the
> same, i.e. unload and re-load the module, and hope things succeed the
> next time around.
I'm not too comfortable with this.  I'll try to take a closer look
tonight, but in the meantime...
> Signed-off-by: Amit Shah <amit.shah at redhat.com>
> ---
>  drivers/char/hw_random/virtio-rng.c | 25 +++++++++++++++----------
>  1 file changed, 15 insertions(+), 10 deletions(-)
> 
> diff --git a/drivers/char/hw_random/virtio-rng.c
b/drivers/char/hw_random/virtio-rng.c
> index a156284..d9927eb 100644
> --- a/drivers/char/hw_random/virtio-rng.c
> +++ b/drivers/char/hw_random/virtio-rng.c
> @@ -35,6 +35,7 @@ struct virtrng_info {
>  	unsigned int data_avail;
>  	int index;
>  	bool busy;
> +	bool hwrng_register_done;
>  };
>  
>  static bool probe_done;
> @@ -136,15 +137,6 @@ static int probe_common(struct virtio_device *vdev)
>  		return err;
>  	}
>  
> -	err = hwrng_register(&vi->hwrng);
> -	if (err) {
> -		vdev->config->del_vqs(vdev);
> -		vi->vq = NULL;
> -		kfree(vi);
> -		ida_simple_remove(&rng_index_ida, index);
> -		return err;
> -	}
> -
This needs to stay.  register, and failure to do so, should occur in the
probe routine.
>  	probe_done = true;
>  	return 0;
>  }
> @@ -152,9 +144,11 @@ static int probe_common(struct virtio_device *vdev)
>  static void remove_common(struct virtio_device *vdev)
>  {
>  	struct virtrng_info *vi = vdev->priv;
> +
>  	vdev->config->reset(vdev);
>  	vi->busy = false;
> -	hwrng_unregister(&vi->hwrng);
> +	if (vi->hwrng_register_done)
> +		hwrng_unregister(&vi->hwrng);
>  	vdev->config->del_vqs(vdev);
>  	ida_simple_remove(&rng_index_ida, vi->index);
>  	kfree(vi);
> @@ -170,6 +164,16 @@ static void virtrng_remove(struct virtio_device *vdev)
>  	remove_common(vdev);
>  }
>  
> +static void virtrng_scan(struct virtio_device *vdev)
> +{
> +	struct virtrng_info *vi = vdev->priv;
> +	int err;
> +
> +	err = hwrng_register(&vi->hwrng);
> +	if (!err)
> +		vi->hwrng_register_done = true;
Instead, perhaps we should just feed the entropy pool from here?  We
would still need to prevent the core from doing so.  Perhaps back to the
flag idea?

thx,

Jason.
> +}
> +
>  #ifdef CONFIG_PM_SLEEP
>  static int virtrng_freeze(struct virtio_device *vdev)
>  {
> @@ -194,6 +198,7 @@ static struct virtio_driver virtio_rng_driver = {
>  	.id_table =	id_table,
>  	.probe =	virtrng_probe,
>  	.remove =	virtrng_remove,
> +	.scan =		virtrng_scan,
>  #ifdef CONFIG_PM_SLEEP
>  	.freeze =	virtrng_freeze,
>  	.restore =	virtrng_restore,
> -- 
> 1.9.3
>

On (Mon) 21 Jul 2014 [08:11:16], Jason Cooper wrote:
> On Mon, Jul 21, 2014 at 05:15:51PM +0530, Amit Shah wrote:
> > Instead of calling hwrng_register() in the probe routing, call it in
the
> > scan routine.  This ensures that when hwrng_register() is successful,
> > and it requests a few random bytes to seed the kernel's pool at
init,
> > we're ready to service that request.
> > 
> > This will also enable us to remove the workaround added previously to
> > check whether probe was completed, and only then ask for data from the
> > host.  The revert follows in the next commit.
> > 
> > There's a slight behaviour change here on unsuccessful
hwrng_register().
> > Previously, when hwrng_unregister() failed, the probe() routine would
> > fail, and the vqs would be torn down, and driver would be marked not
> > initialized.  Now, the vqs will remain initialized, driver would be
> > marked initialized as well, but won't be available in the list of
RNGs
> > available to hwrng core.  To fix the failures, the procedure remains
the
> > same, i.e. unload and re-load the module, and hope things succeed the
> > next time around.
> 
> I'm not too comfortable with this.  I'll try to take a closer look
> tonight, but in the meantime...
> 
> > Signed-off-by: Amit Shah <amit.shah at redhat.com>
> > ---
> >  drivers/char/hw_random/virtio-rng.c | 25 +++++++++++++++----------
> >  1 file changed, 15 insertions(+), 10 deletions(-)
> > 
> > diff --git a/drivers/char/hw_random/virtio-rng.c
b/drivers/char/hw_random/virtio-rng.c
> > index a156284..d9927eb 100644
> > --- a/drivers/char/hw_random/virtio-rng.c
> > +++ b/drivers/char/hw_random/virtio-rng.c
> > @@ -35,6 +35,7 @@ struct virtrng_info {
> >  	unsigned int data_avail;
> >  	int index;
> >  	bool busy;
> > +	bool hwrng_register_done;
> >  };
> >  
> >  static bool probe_done;
> > @@ -136,15 +137,6 @@ static int probe_common(struct virtio_device
*vdev)
> >  		return err;
> >  	}
> >  
> > -	err = hwrng_register(&vi->hwrng);
> > -	if (err) {
> > -		vdev->config->del_vqs(vdev);
> > -		vi->vq = NULL;
> > -		kfree(vi);
> > -		ida_simple_remove(&rng_index_ida, index);
> > -		return err;
> > -	}
> > -
> 
> This needs to stay.  register, and failure to do so, should occur in the
> probe routine.
Can you elaborate why?
> >  	probe_done = true;
> >  	return 0;
> >  }
> > @@ -152,9 +144,11 @@ static int probe_common(struct virtio_device
*vdev)
> >  static void remove_common(struct virtio_device *vdev)
> >  {
> >  	struct virtrng_info *vi = vdev->priv;
> > +
> >  	vdev->config->reset(vdev);
> >  	vi->busy = false;
> > -	hwrng_unregister(&vi->hwrng);
> > +	if (vi->hwrng_register_done)
> > +		hwrng_unregister(&vi->hwrng);
> >  	vdev->config->del_vqs(vdev);
> >  	ida_simple_remove(&rng_index_ida, vi->index);
> >  	kfree(vi);
> > @@ -170,6 +164,16 @@ static void virtrng_remove(struct virtio_device
*vdev)
> >  	remove_common(vdev);
> >  }
> >  
> > +static void virtrng_scan(struct virtio_device *vdev)
> > +{
> > +	struct virtrng_info *vi = vdev->priv;
> > +	int err;
> > +
> > +	err = hwrng_register(&vi->hwrng);
> > +	if (!err)
> > +		vi->hwrng_register_done = true;
> 
> Instead, perhaps we should just feed the entropy pool from here?  We
> would still need to prevent the core from doing so.  Perhaps back to the
> flag idea?
No way hwrng knows the difference between probe and scan for
virtio-rng, so it's back to the delayed workqueue idea, if this isn't
usable..

But I need to understand why this isn't workable.

Thanks,

		Amit

On Mon, Jul 21, 2014 at 08:11:16AM -0400, Jason Cooper
wrote:
>
> > @@ -136,15 +137,6 @@ static int probe_common(struct virtio_device
*vdev)
> >  		return err;
> >  	}
> >  
> > -	err = hwrng_register(&vi->hwrng);
> > -	if (err) {
> > -		vdev->config->del_vqs(vdev);
> > -		vi->vq = NULL;
> > -		kfree(vi);
> > -		ida_simple_remove(&rng_index_ida, index);
> > -		return err;
> > -	}
> > -
> 
> This needs to stay.  register, and failure to do so, should occur in the
> probe routine.
Why? Probing involves finding out whether the hardware is present.
While hwrng_register is about adding an entry in the hwrng system
which may only be one aspect of the given hardware.

For example, the same hardware may support other features that are
used outside of the hwrng system, e.g., crypto.

So there's nothing special about probe that requires us to have the
hwrng_register call there.  In practice, the only difference between
having it in probe vs. scan is that you can return errors.  The only
error that can be returned is ENOMEM which isn't of much interest to
the caller of probe anyway.

On the other hand, if you are calling hwrng_register you better be
damn sure that your hardware is ready to answer requests from the
hwrng system.  Please don't add silly flags to work around this.

Cheers,
-- 
Email: Herbert Xu <herbert at gondor.apana.org.au>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt

On Mon, Jul 21, 2014 at 08:41:24PM +0800, Herbert Xu
wrote:
> On Mon, Jul 21, 2014 at 08:11:16AM -0400, Jason Cooper wrote:
> >
> > > @@ -136,15 +137,6 @@ static int probe_common(struct virtio_device
*vdev)
> > >  		return err;
> > >  	}
> > >  
> > > -	err = hwrng_register(&vi->hwrng);
> > > -	if (err) {
> > > -		vdev->config->del_vqs(vdev);
> > > -		vi->vq = NULL;
> > > -		kfree(vi);
> > > -		ida_simple_remove(&rng_index_ida, index);
> > > -		return err;
> > > -	}
> > > -
> > 
> > This needs to stay.  register, and failure to do so, should occur in
the
> > probe routine.
> 
> Why? 
hmm, yes. On reviewing the code, my objection was silly.  It mostly
stemmed from the commit message's mentioning of reduced error handling.

thx,

Jason.