Hi Vincent,
On 2/25/19 3:22 PM, Vincent Stehlé wrote:
> Hi Eric,
>
> On Mon, Feb 18, 2019 at 02:55:00PM +0100, Eric Auger wrote:
>> This patch registers a fault handler which records faults in
>> a circular buffer and then signals an eventfd. This buffer is
>> exposed within the fault region.
>>
>> Signed-off-by: Eric Auger <eric.auger@xxxxxxxxxx>
>> ---
>> drivers/vfio/pci/vfio_pci.c | 49 +++++++++++++++++++++++++++++
>> drivers/vfio/pci/vfio_pci_private.h | 1 +
>> 2 files changed, 50 insertions(+)
>>
>> diff --git a/drivers/vfio/pci/vfio_pci.c b/drivers/vfio/pci/vfio_pci.c
>> index aaf63e5ca2b6..019c9fd380a5 100644
>> --- a/drivers/vfio/pci/vfio_pci.c
>> +++ b/drivers/vfio/pci/vfio_pci.c
> (..)
>> static int vfio_pci_init_fault_region(struct vfio_pci_device *vdev)
>> {
>> struct vfio_region_fault_prod *header;
>> @@ -276,6 +317,13 @@ static int vfio_pci_init_fault_region(struct vfio_pci_device *vdev)
>> header = (struct vfio_region_fault_prod *)vdev->fault_pages;
>> header->version = -1;
>> header->offset = PAGE_SIZE;
>> +
>> + ret = iommu_register_device_fault_handler(&vdev->pdev->dev,
>> + vfio_pci_iommu_dev_fault_handler,
>> + vdev);
>> + if (ret)
>> + goto out;
>> +
>> return 0;
>> out:
>> kfree(vdev->fault_pages);
>
> This patch calls iommu_register_device_fault_handler from
> vfio_pci_init_fault_region, leading to the following call stack:
>
> iommu_register_device_fault_handler
> vfio_pci_init_fault_region
> vfio_pci_enable
> vfio_pci_open
> vfio_group_get_device_fd
>
>> @@ -1420,6 +1468,7 @@ static void vfio_pci_remove(struct pci_dev *pdev)
>> vfio_iommu_group_put(pdev->dev.iommu_group, &pdev->dev);
>> kfree(vdev->region);
>> kfree(vdev->fault_pages);
>> + iommu_unregister_device_fault_handler(&pdev->dev);
>> mutex_destroy(&vdev->ioeventfds_lock);
>> kfree(vdev);
>
> And then this patch calls iommu_unregister_device_fault_handler from
> vfio_pci_remove, and not from vfio_pci_release.
Yes you're fully right. Thank you for the time spent debugging the
issue. this is a left-over from the previous version and indeed the
unregistration should be called from the release ops.
By the way, I will package a qemu version for testing this week. Sorry
for the delay.
Thanks
Eric
>
> I think this means a device cannot be used twice in a row without unloading the
> module.
>
> Here is an example sequence:
>
> 1. modprobe vfio-pci
> 2. Userspace uses VFIO, calls ioctl(VFIO_GROUP_GET_DEVICE_FD)
> 2.1. iommu_register_device_fault_handler is called
> 3. Userspace exits
> 3.1. vfio_pci_release is called,
> but iommu_unregister_device_fault_handler is not called
> 4. Userspace uses VFIO agin, calls ioctl(VFIO_GROUP_GET_DEVICE_FD) again
> 4.1. iommu_register_device_fault_handler is called again,
> notices a fault handler is already there,
> returns -EBUSY
>
> Unloading the vfio-pci module will call vfio_pci_remove.
>
> Maybe iommu_unregister_device_fault_handler should be called from
> vfio_pci_release instead of vfio_pci_remove?
>
> Best regards,
> Vincent.
>
