Re: [PATCH v3] iommu/s390: Add support for iommu_device handling

[Joerg Roedel]

Hi Sebastian,

On Fri, Aug 11, 2017 at 07:02:36PM +0200, Sebastian Ott wrote:
> ..but I found the bug, actually 2 bugs:
> 
> * That patch embedded a struct iommu_device within struct zpci_dev but
> the iommu_device has a release function (via its class) - so when
> the release function gets called it frees memory that was never allocated.
> The fix is to not embedd struct iommu_device in zpci_dev (see below)
> 
> * iommu_release_device must not release the struct device but the
> structure it is embedded in: struct iommu_device (I'll send a patch
> for that)

Thanks a lot for your in-depth analysis, this bug has been around since
4.11 time :-/ Unfortunatly I can't test iommu-unplug here, so I didn't
notice it until your report.

Meanwhile I worked on a different fix that make the 'struct device' in
iommu_device a pointer again. As Gerald and you noticed already,
struct iommu_device is embedded in other structs as well, and I'd like
to keep it that way. The reason is that in the future the code in
iommu.c should call into iommu-drivers and supply a struct iommu_device,
and the driver can then just do a container_of to get its own data about
that hardware iommu.

So just making the dev member a pointer is a simpler fix for now. The
other option would have been to call back into the iommu-drivers from
the release-function.

I attach the patch I wrote to fix this, can you please test it together
with the initial patch in this thread?

Thanks,

	Joerg