lkml.org 
[lkml]   [2023]   [Jan]   [29]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
    /
    From
    SubjectRE: [PATCH v2 3/3] iommufd/device: Change iommufd_hw_pagetable_has_group to device centric
    Date
    > From: Nicolin Chen <nicolinc@nvidia.com>
    > Sent: Sunday, January 29, 2023 6:39 PM
    >
    > On Sun, Jan 29, 2023 at 09:37:00AM +0000, Tian, Kevin wrote:
    > > External email: Use caution opening links or attachments
    > >
    > >
    > > > From: Nicolin Chen <nicolinc@nvidia.com>
    > > > Sent: Sunday, January 29, 2023 5:18 AM
    > > >
    > > > -static bool iommufd_hw_pagetable_has_group(struct
    > > > iommufd_hw_pagetable *hwpt,
    > > > - struct iommu_group *group)
    > > > +static bool iommufd_hw_pagetable_has_device(struct
    > > > iommufd_hw_pagetable *hwpt,
    > > > + struct device *dev)
    > > > {
    > > > - struct iommufd_device *cur_dev;
    > > > -
    > > > - list_for_each_entry(cur_dev, &hwpt->devices, devices_item)
    > > > - if (cur_dev->group == group)
    > > > - return true;
    > > > - return false;
    > > > + /*
    > > > + * iommu_get_domain_for_dev() returns an iommu_group->domain
    > > > ptr, if it
    > > > + * is the same domain as the hwpt->domain, it means that this hwpt
    > > > has
    > > > + * the iommu_group/device.
    > > > + */
    > > > + return hwpt->domain == iommu_get_domain_for_dev(dev);
    > > > }
    > >
    > > Here we could have three scenarios:
    > >
    > > 1) the device is attached to blocked domain;
    > > 2) the device is attached to hwpt->domain;
    > > 3) the device is attached to another hwpt->domain;
    > >
    > > if this function returns false then iommufd_device_do_attach() will attach
    > > the device to the specified hwpt. But then it's wrong for 3).
    > >
    > > Has 3) been denied in earlier path? If yes at least a WARN_ON for
    > > case 3) makes sense here.
    >
    > The case #3 means the device is already attached to some other
    > domain? Then vfio_iommufd_physical_attach_ioas returns -EBUSY
    > at the sanity of vdev->iommufd_attached. And the case #3 feels
    > like a domain replacement use case to me. So probably not that
    > necessary to add a wARN_ON?
    >

    You are right. I thought about the cdev case where the device is
    not attached in vfio but has a valid domain due to attach status
    of other devices in the group. But even in this case it's user's
    responsibility to not break group boundary. So yes it's just a
    domain replacement and WARN_ON is not required.

    \
     
     \ /
      Last update: 2023-03-27 00:01    [W:3.674 / U:0.004 seconds]
    ©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site