Make sure to drop the reference taken to the iommu platform device when
looking up its driver data during probe_device().

Note that commit 9826e393e4a8 ("iommu/tegra-smmu: Fix missing
put_device() call in tegra_smmu_find") fixed the leak in an error path,
but the reference is still leaking on success.

Fixes: 891846516317 ("memory: Add NVIDIA Tegra memory controller support")
Cc: stable@vger.kernel.org	# 3.19: 9826e393e4a8
Cc: Thierry Reding <treding@nvidia.com>
Cc: Miaoqian Lin <linmq006@gmail.com>
Acked-by: Robin Murphy <robin.murphy@arm.com>
Signed-off-by: Johan Hovold <johan@kernel.org>
---
 drivers/iommu/tegra-smmu.c | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)

diff --git a/drivers/iommu/tegra-smmu.c b/drivers/iommu/tegra-smmu.c
index 36cdd5fbab07..f6f26a072820 100644
--- a/drivers/iommu/tegra-smmu.c
+++ b/drivers/iommu/tegra-smmu.c
@@ -830,10 +830,9 @@ static struct tegra_smmu *tegra_smmu_find(struct device_node *np)
 		return NULL;
 
 	mc = platform_get_drvdata(pdev);
-	if (!mc) {
-		put_device(&pdev->dev);
+	put_device(&pdev->dev);
+	if (!mc)
 		return NULL;
-	}
 
 	return mc->smmu;
 }
-- 
2.49.1

On Tue, Oct 07, 2025 at 11:43:27AM +0200, Johan Hovold wrote:
> Make sure to drop the reference taken to the iommu platform device when
> looking up its driver data during probe_device().
> 
> Note that commit 9826e393e4a8 ("iommu/tegra-smmu: Fix missing
> put_device() call in tegra_smmu_find") fixed the leak in an error path,
> but the reference is still leaking on success.
> 
> Fixes: 891846516317 ("memory: Add NVIDIA Tegra memory controller support")
> Cc: stable@vger.kernel.org	# 3.19: 9826e393e4a8
> Cc: Thierry Reding <treding@nvidia.com>
> Cc: Miaoqian Lin <linmq006@gmail.com>
> Acked-by: Robin Murphy <robin.murphy@arm.com>
> Signed-off-by: Johan Hovold <johan@kernel.org>
> ---
>  drivers/iommu/tegra-smmu.c | 5 ++---
>  1 file changed, 2 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/iommu/tegra-smmu.c b/drivers/iommu/tegra-smmu.c
> index 36cdd5fbab07..f6f26a072820 100644
> --- a/drivers/iommu/tegra-smmu.c
> +++ b/drivers/iommu/tegra-smmu.c
> @@ -830,10 +830,9 @@ static struct tegra_smmu *tegra_smmu_find(struct device_node *np)
>  		return NULL;
>  
>  	mc = platform_get_drvdata(pdev);
> -	if (!mc) {
> -		put_device(&pdev->dev);
> +	put_device(&pdev->dev);
> +	if (!mc)
>  		return NULL;
> -	}
>  
>  	return mc->smmu;

pdev->dev is what's backing mc, so if we use put_device() here, then the
MC could go away at any time, right?

So the goal here was to make sure that the MC stays around during the
entire lifetime of the IOMMU attachment. We don't currently release that
reference, ever, so there is a leak, but wouldn't it be more appropriate
to release it in a .release_device implementation?

Thierry

On Thu, Oct 09, 2025 at 09:56:18AM +0200, Thierry Reding wrote:
> On Tue, Oct 07, 2025 at 11:43:27AM +0200, Johan Hovold wrote:

> > @@ -830,10 +830,9 @@ static struct tegra_smmu *tegra_smmu_find(struct device_node *np)
> >  		return NULL;
> >  
> >  	mc = platform_get_drvdata(pdev);
> > -	if (!mc) {
> > -		put_device(&pdev->dev);
> > +	put_device(&pdev->dev);
> > +	if (!mc)
> >  		return NULL;
> > -	}
> >  
> >  	return mc->smmu;
> 
> pdev->dev is what's backing mc, so if we use put_device() here, then the
> MC could go away at any time, right?

Holding a reference to a device does not prevent its driver data from
going away so there is no point in keeping the reference.

But from what I can tell, you don't need to worry about that anyway
since it's the memory controller driver that registers the iommu (and
the driver can't be unbound).

Johan

On Thu, Oct 09, 2025 at 10:27:55AM +0200, Johan Hovold wrote:
> On Thu, Oct 09, 2025 at 09:56:18AM +0200, Thierry Reding wrote:
> > On Tue, Oct 07, 2025 at 11:43:27AM +0200, Johan Hovold wrote:
> 
> > > @@ -830,10 +830,9 @@ static struct tegra_smmu *tegra_smmu_find(struct device_node *np)
> > >  		return NULL;
> > >  
> > >  	mc = platform_get_drvdata(pdev);
> > > -	if (!mc) {
> > > -		put_device(&pdev->dev);
> > > +	put_device(&pdev->dev);
> > > +	if (!mc)
> > >  		return NULL;
> > > -	}
> > >  
> > >  	return mc->smmu;
> > 
> > pdev->dev is what's backing mc, so if we use put_device() here, then the
> > MC could go away at any time, right?
> 
> Holding a reference to a device does not prevent its driver data from
> going away so there is no point in keeping the reference.
> 
> But from what I can tell, you don't need to worry about that anyway
> since it's the memory controller driver that registers the iommu (and
> the driver can't be unbound).

That's true. It'd be nice to at least conceptually do the right thing
here, but not sure it's worth it. As you said, driver data going away
would need special handling and it's not even clear what that would
mean in terms of the clients...

Acked-by: Thierry Reding <treding@nvidia.com>