Hello Alejandro,
On 9/19/25 23:34, Alejandro Jimenez wrote:
This series adds support for guests using the AMD vIOMMU to enable DMA remapping
for VFIO devices. Please see v1[0] cover letter for additional details such as
example QEMU command line parameters used in testing.
I have sanity tested on an AMD EPYC Genoa host, booting a Linux guest with
'iommu.passthrough=0' and several CX6 VFs, and there are no issues during
typical guest operation.
When using the non-default parameter 'iommu.forcedac=1' in the guest kernel
cmdline, this initially fails due to a VFIO integer overflow bug which requires
the following fix in the host kernel:
https://github.com/aljimenezb/linux/commit/014be8cafe7464d278729583a2dd5d94514e2e2a
This is a work in progress as there are other locations in the driver that are
susceptible to overflows, but the above is sufficient to fix the initial
problem.
Even after that fix is applied, I see an issue on guest reboot when 'forcedac=1'
is in use. Although the guest boots, the VF is not properly initialized, failing
with a timeout. Once the guest reaches userspace the VF driver can be reloaded
and it then works as expected. I am still investigating the root cause for this
issue, and will need to discuss all the steps I have tried to eliminate
potential sources of errors in a separate thread.
I am sending v3 despite this known issue since forcedac=1 is not a default or
commonly known/used setting. Having the large portions of the infrastructure for
DMA remapping already in place (and working) will make it easier to debug this
corner case and get feedback/testing from the community. I hope this is a viable
approach, otherwise I am happy to discuss all the steps I have taken to debug
this issue in this thread and test any suggestions to address it.
Changes since v2[2]:
- P5: Fixed missed check for AMDVI_FR_DTE_RTR_ERR in amdvi_do_translate()
(Sairaj)
- P6: Reword commit message to clarify the need to discern between empty PTEs
and errors (Vasant)
- P9: Use correct enum type for notifier flags and remove whitespace changes
(Sairaj)
- P11: Fixed integer overflow bug when guest uses iommu.forcedac=1. Fixed in
P8. (Sairaj)
- P15: Fixed typo in commit message (Sairaj)
- P16: On reset, use passthrough mode by default on all address spaces (Sairaj)
- P18: Enforce isolation by using DMA mode on errors retrieving DTE (Ethan &
Sairaj)
- P20: Removed unused pte_override_page_mask() and pte_get_page_mask() to avoid
-Wunused-function error.
- Add HATDis support patches from Joao Martins (HATDis available in Linux since
[1])
Thank you,
Alejandro
[0]
https://lore.kernel.org/all/[email protected]/
[1] https://lore.kernel.org/all/[email protected]/
[2]
https://lore.kernel.org/qemu-devel/[email protected]/
Alejandro Jimenez (20):
memory: Adjust event ranges to fit within notifier boundaries
amd_iommu: Document '-device amd-iommu' common options
amd_iommu: Reorder device and page table helpers
amd_iommu: Helper to decode size of page invalidation command
amd_iommu: Add helper function to extract the DTE
amd_iommu: Return an error when unable to read PTE from guest memory
amd_iommu: Add helpers to walk AMD v1 Page Table format
amd_iommu: Add a page walker to sync shadow page tables on
invalidation
amd_iommu: Add basic structure to support IOMMU notifier updates
amd_iommu: Sync shadow page tables on page invalidation
amd_iommu: Use iova_tree records to determine large page size on UNMAP
amd_iommu: Unmap all address spaces under the AMD IOMMU on reset
amd_iommu: Add replay callback
amd_iommu: Invalidate address translations on INVALIDATE_IOMMU_ALL
amd_iommu: Toggle memory regions based on address translation mode
amd_iommu: Set all address spaces to use passthrough mode on reset
amd_iommu: Add dma-remap property to AMD vIOMMU device
amd_iommu: Toggle address translation mode on devtab entry
invalidation
amd_iommu: Do not assume passthrough translation when DTE[TV]=0
amd_iommu: Refactor amdvi_page_walk() to use common code for page walk
Joao Martins (2):
i386/intel-iommu: Move dma_translation to x86-iommu
amd_iommu: HATDis/HATS=11 support
hw/i386/acpi-build.c | 6 +-
hw/i386/amd_iommu.c | 1056 ++++++++++++++++++++++++++++++-----
hw/i386/amd_iommu.h | 51 ++
The current status of AMD-Vi Emulation in MAINTAINERS is Orphan.
Since this series is about to be merged, should AMD-Vi be considered
maintained now ? and if so by whom ?
Thanks,
C.
hw/i386/intel_iommu.c | 5 +-
hw/i386/x86-iommu.c | 1 +
include/hw/i386/x86-iommu.h | 1 +
qemu-options.hx | 23 +
system/memory.c | 10 +-
8 files changed, 999 insertions(+), 154 deletions(-)>
base-commit: ab8008b231e758e03c87c1c483c03afdd9c02e19
