Re: [PATCH v2 0/6] IOMMUFD: Deliver IO page faults to user space

From: Baolu Lu
Date: Fri Dec 08 2023 - 01:02:12 EST


On 12/1/23 10:24 PM, Jason Gunthorpe wrote:
On Thu, Oct 26, 2023 at 10:49:24AM +0800, Lu Baolu wrote:
Hi folks,

This series implements the functionality of delivering IO page faults to
user space through the IOMMUFD framework for nested translation. Nested
translation is a hardware feature that supports two-stage translation
tables for IOMMU. The second-stage translation table is managed by the
host VMM, while the first-stage translation table is owned by user
space. This allows user space to control the IOMMU mappings for its
devices.

When an IO page fault occurs on the first-stage translation table, the
IOMMU hardware can deliver the page fault to user space through the
IOMMUFD framework. User space can then handle the page fault and respond
to the device top-down through the IOMMUFD. This allows user space to
implement its own IO page fault handling policies.

User space indicates its capability of handling IO page faults by
setting the IOMMU_HWPT_ALLOC_IOPF_CAPABLE flag when allocating a
hardware page table (HWPT). IOMMUFD will then set up its infrastructure
for page fault delivery. On a successful return of HWPT allocation, the
user can retrieve and respond to page faults by reading and writing to
the file descriptor (FD) returned in out_fault_fd.

This is probably backwards, userspace should allocate the FD with a
dedicated ioctl and provide it during domain allocation.

Introducing a dedicated fault FD for fault handling seems promising. It
decouples the fault handling from any specific domain. I suppose we need
different fault fd for recoverable faults (a.k.a. IO page fault) and
unrecoverable faults. Do I understand you correctly?


If the userspace wants a fd per domain then it should do that. If it
wants to share fds between domains that should work too.

Yes, it's more flexible. The fault message contains the hwpt obj id, so
user space can recognize the hwpt on which the fault happened.

Best regards,
baolu