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

From: Sebastian Ott
Date: Fri Aug 11 2017 - 13:02:48 EST


Hello Joerg,

On Fri, 11 Aug 2017, Joerg Roedel wrote:
> Hey Sebastian,
>
> On Thu, Aug 10, 2017 at 09:07:06PM +0200, Sebastian Ott wrote:
> > With this patch pci hot-unplug fails with a use after free or refcounting
> > issue - I'm currently trying to understand what's going on...
>
> Let me know if I can help with debugging the issue, do you have a
> backtrace for me to look at?

I would have send backtraces but everyone looked different: random mem
corruptions, panic during unrelated allocations and stuff like that.

..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)

With these fixed it works fine.

Sebastian

---
arch/s390/include/asm/pci.h | 2 +-
drivers/iommu/s390-iommu.c | 20 ++++++++++++--------
2 files changed, 13 insertions(+), 9 deletions(-)

diff --git a/arch/s390/include/asm/pci.h b/arch/s390/include/asm/pci.h
index 386df9a..de3129e 100644
--- a/arch/s390/include/asm/pci.h
+++ b/arch/s390/include/asm/pci.h
@@ -123,7 +123,7 @@ struct zpci_dev {
unsigned long iommu_pages;
unsigned int next_bit;

- struct iommu_device iommu_dev; /* IOMMU core handle */
+ struct iommu_device *iommu_dev; /* IOMMU core handle */

char res_name[16];
struct zpci_bar_struct bars[PCI_BAR_COUNT];
diff --git a/drivers/iommu/s390-iommu.c b/drivers/iommu/s390-iommu.c
index 85f3bc5..58a7414 100644
--- a/drivers/iommu/s390-iommu.c
+++ b/drivers/iommu/s390-iommu.c
@@ -174,7 +174,7 @@ static int s390_iommu_add_device(struct device *dev)
return PTR_ERR(group);

iommu_group_put(group);
- iommu_device_link(&zdev->iommu_dev, dev);
+ iommu_device_link(zdev->iommu_dev, dev);

return 0;
}
@@ -201,7 +201,7 @@ static void s390_iommu_remove_device(struct device *dev)
s390_iommu_detach_device(domain, dev);
}

- iommu_device_unlink(&zdev->iommu_dev, dev);
+ iommu_device_unlink(zdev->iommu_dev, dev);
iommu_group_remove_device(dev);
}

@@ -336,21 +336,25 @@ int zpci_init_iommu(struct zpci_dev *zdev)
{
int rc = 0;

- rc = iommu_device_sysfs_add(&zdev->iommu_dev, NULL, NULL,
+ zdev->iommu_dev = kzalloc(sizeof(*zdev->iommu_dev), GFP_KERNEL);
+ if (!zdev->iommu_dev)
+ return -ENOMEM;
+
+ rc = iommu_device_sysfs_add(zdev->iommu_dev, NULL, NULL,
"s390-iommu.%08x", zdev->fid);
if (rc)
goto out_err;

- iommu_device_set_ops(&zdev->iommu_dev, &s390_iommu_ops);
+ iommu_device_set_ops(zdev->iommu_dev, &s390_iommu_ops);

- rc = iommu_device_register(&zdev->iommu_dev);
+ rc = iommu_device_register(zdev->iommu_dev);
if (rc)
goto out_sysfs;

return 0;

out_sysfs:
- iommu_device_sysfs_remove(&zdev->iommu_dev);
+ iommu_device_sysfs_remove(zdev->iommu_dev);

out_err:
return rc;
@@ -358,8 +362,8 @@ int zpci_init_iommu(struct zpci_dev *zdev)

void zpci_destroy_iommu(struct zpci_dev *zdev)
{
- iommu_device_unregister(&zdev->iommu_dev);
- iommu_device_sysfs_remove(&zdev->iommu_dev);
+ iommu_device_unregister(zdev->iommu_dev);
+ iommu_device_sysfs_remove(zdev->iommu_dev);
}

static struct iommu_ops s390_iommu_ops = {
--
2.5.5