| Message ID | 0beddeaaa4a8a7a45ce93ff21c543ae58be64908.1746139811.git.nicolinc@nvidia.com |
|---|---|
| State | New |
| Headers | show |
| Series | iommufd: Add vIOMMU infrastructure (Part-4 vQUEUE) | expand |
On Thu, May 01, 2025 at 04:01:22PM -0700, Nicolin Chen wrote: > +- IOMMUFD_OBJ_VQUEUE, representing a hardware accelerated virtual queue, as a > + subset of IOMMU's virtualization features, for the IOMMU HW to directly read > + or write the virtual queue memory owned by a guest OS. This HW-acceleration > + allows VM to work with the IOMMU HW directly without a VM Exit, i.e. reducing > + overhead from the hypercalls. Along with this vQUEUE object, iommufd provides > + user space an mmap interface for VMM to mmap a physical MMIO region from the > + host physical address space to the guest physical address space, allowing the > + guest OS to control the allocated vQUEUE HW. Thus, when allocating a vQUEUE, > + the VMM must request a pair of VMA info (vm_pgoff/size) for an mmap syscall. > + The length argument of an mmap syscall can be smaller than the given size for > + a partial mmap, but the addr argument of the mmap syscall should never offset > + from the returned vm_pgoff, which implies that an mmap will always start from Did you mean never be offset from returned vm_pgoff? > + the beginning of the physical MMIO region. > + Confused...
On Fri, May 02, 2025 at 10:50:07AM +0700, Bagas Sanjaya wrote: > On Thu, May 01, 2025 at 04:01:22PM -0700, Nicolin Chen wrote: > > +- IOMMUFD_OBJ_VQUEUE, representing a hardware accelerated virtual queue, as a > > + subset of IOMMU's virtualization features, for the IOMMU HW to directly read > > + or write the virtual queue memory owned by a guest OS. This HW-acceleration > > + allows VM to work with the IOMMU HW directly without a VM Exit, i.e. reducing > > + overhead from the hypercalls. Along with this vQUEUE object, iommufd provides > > + user space an mmap interface for VMM to mmap a physical MMIO region from the > > + host physical address space to the guest physical address space, allowing the > > + guest OS to control the allocated vQUEUE HW. Thus, when allocating a vQUEUE, > > + the VMM must request a pair of VMA info (vm_pgoff/size) for an mmap syscall. > > + The length argument of an mmap syscall can be smaller than the given size for > > + a partial mmap, but the addr argument of the mmap syscall should never offset > > + from the returned vm_pgoff, which implies that an mmap will always start from > > Did you mean never be offset from returned vm_pgoff? Yes. Will fix this. > > + the beginning of the physical MMIO region. > > + > > Confused... Meaning that VMM should just use the given vm_pgoff as is, without adding any offset to the vm_pgoff. Thanks Nicolin
On Thu, May 01, 2025 at 10:29:58PM -0700, Nicolin Chen wrote: > On Fri, May 02, 2025 at 10:50:07AM +0700, Bagas Sanjaya wrote: > > On Thu, May 01, 2025 at 04:01:22PM -0700, Nicolin Chen wrote: > > > +- IOMMUFD_OBJ_VQUEUE, representing a hardware accelerated virtual queue, as a > > > + subset of IOMMU's virtualization features, for the IOMMU HW to directly read > > > + or write the virtual queue memory owned by a guest OS. This HW-acceleration > > > + allows VM to work with the IOMMU HW directly without a VM Exit, i.e. reducing > > > + overhead from the hypercalls. Along with this vQUEUE object, iommufd provides > > > + user space an mmap interface for VMM to mmap a physical MMIO region from the > > > + host physical address space to the guest physical address space, allowing the > > > + guest OS to control the allocated vQUEUE HW. Thus, when allocating a vQUEUE, > > > + the VMM must request a pair of VMA info (vm_pgoff/size) for an mmap syscall. > > > + The length argument of an mmap syscall can be smaller than the given size for > > > + a partial mmap, but the addr argument of the mmap syscall should never offset > > > + from the returned vm_pgoff, which implies that an mmap will always start from > > > > Did you mean never be offset from returned vm_pgoff? > > Yes. Will fix this. > > > > + the beginning of the physical MMIO region. > > > + > > > > Confused... > > Meaning that VMM should just use the given vm_pgoff as is, without > adding any offset to the vm_pgoff. Understood, thanks!
diff --git a/Documentation/userspace-api/iommufd.rst b/Documentation/userspace-api/iommufd.rst index b0df15865dec..ed32713a97a3 100644 --- a/Documentation/userspace-api/iommufd.rst +++ b/Documentation/userspace-api/iommufd.rst @@ -124,6 +124,20 @@ Following IOMMUFD objects are exposed to userspace: used to allocate a vEVENTQ. Each vIOMMU can support multiple types of vEVENTS, but is confined to one vEVENTQ per vEVENTQ type. +- IOMMUFD_OBJ_VQUEUE, representing a hardware accelerated virtual queue, as a + subset of IOMMU's virtualization features, for the IOMMU HW to directly read + or write the virtual queue memory owned by a guest OS. This HW-acceleration + allows VM to work with the IOMMU HW directly without a VM Exit, i.e. reducing + overhead from the hypercalls. Along with this vQUEUE object, iommufd provides + user space an mmap interface for VMM to mmap a physical MMIO region from the + host physical address space to the guest physical address space, allowing the + guest OS to control the allocated vQUEUE HW. Thus, when allocating a vQUEUE, + the VMM must request a pair of VMA info (vm_pgoff/size) for an mmap syscall. + The length argument of an mmap syscall can be smaller than the given size for + a partial mmap, but the addr argument of the mmap syscall should never offset + from the returned vm_pgoff, which implies that an mmap will always start from + the beginning of the physical MMIO region. + All user-visible objects are destroyed via the IOMMU_DESTROY uAPI. The diagrams below show relationships between user-visible objects and kernel @@ -270,6 +284,7 @@ User visible objects are backed by following datastructures: - iommufd_viommu for IOMMUFD_OBJ_VIOMMU. - iommufd_vdevice for IOMMUFD_OBJ_VDEVICE. - iommufd_veventq for IOMMUFD_OBJ_VEVENTQ. +- iommufd_vqueue for IOMMUFD_OBJ_VQUEUE. Several terminologies when looking at these datastructures:
With the introduction of the new object and its infrastructure, update the doc to reflect that. Signed-off-by: Nicolin Chen <nicolinc@nvidia.com> --- Documentation/userspace-api/iommufd.rst | 15 +++++++++++++++ 1 file changed, 15 insertions(+)