On Wed, Mar 30, 2022 at 02:33:23PM -0300, Jason Gunthorpe wrote:
> On Wed, Mar 30, 2022 at 08:19:37PM +0300, Tony Lindgren wrote:
>
> > > > __iommu_probe_device from probe_iommu_group+0x2c/0x38
> > > > probe_iommu_group from bus_for_each_dev+0x74/0xbc
> > > > bus_for_each_dev from bus_iommu_probe+0x34/0x2e8
> > > > bus_iommu_probe from bus_set_iommu+0x80/0xc8
> > > > bus_set_iommu from omap_iommu_init+0x88/0xcc
> > > > omap_iommu_init from do_one_initcall+0x44/0x24c
> > > >
> > > > Any ideas for a fix?
> > > >
> > > > It would be good to fix this quickly so we don't end up with a broken
> > > > v5.18-rc1..
> > > >
> > > > For reference, this is mainline commit 41bb23e70b50 ("iommu: Remove unused
> > > > argument in is_attach_deferred").
> > >
> > > Are you confident in the bisection? I don't see how that commit could
> > > NULL deref..
> >
> > Oops sorry you're right, the breaking commit is a different patch, it's
> > 3f6634d997db ("iommu: Use right way to retrieve iommu_ops") instead. I must
> > have picked the wrong commit while testing.
>
> That makes alot more sense
>
> > > Can you find the code that is the NULL deref?
> >
> > I can debug a bit more tomorrow.
>
> Looks like omap's omap_iommu_probe_device() is buggy, it returns 0 on
> error paths:
>
> num_iommus = of_property_count_elems_of_size(dev->of_node, "iommus",
> sizeof(phandle));
> if (num_iommus < 0)
> return 0;
>
> This function needs to return an errno -ENODEV
>
> Otherwise it causes a NULL dev->iommu->iommu_dev and dev_iommu_ops() will
> crash.
>
> Jason
I tried to boot current mainline (787af64d05cd) on am57xx-evm and it
failed to boot. I made the change you suggested and it boots okay now:
https://gist.github.com/pdp7/918eefe03b5c5db3b5d28d819f6a27f6
thanks,
drew
diff --git a/drivers/iommu/omap-iommu.c b/drivers/iommu/omap-iommu.c
index 4aab631ef517..d9cf2820c02e 100644
--- a/drivers/iommu/omap-iommu.c
+++ b/drivers/iommu/omap-iommu.c
@@ -1661,7 +1661,7 @@ static struct iommu_device *omap_iommu_probe_device(struct device *dev)
num_iommus = of_property_count_elems_of_size(dev->of_node, "iommus",
sizeof(phandle));
if (num_iommus < 0)
- return 0;
+ return ERR_PTR(-ENODEV);
arch_data = kcalloc(num_iommus + 1, sizeof(*arch_data), GFP_KERNEL);
if (!arch_data)
