Hi Fengyuan,
On 2026/2/4 11:06, Fengyuan Yu wrote:
Hi,
This patch series adds a bare-metal qtest for the Intel IOMMU (VT-d) using
the iommu-testdev framework. The test exercises address translation paths
without requiring a full guest OS boot.
Motivation
----------
The Intel IOMMU implementation in QEMU supports various translation modes
including pass-through and translated (4-level paging) modes. Currently,
comprehensive testing of these translation paths requires booting a full
guest OS with appropriate drivers, which is time-consuming and makes
regression testing difficult.
This new test fills that gap by using iommu-testdev to trigger DMA
transactions and validate the IOMMU's translation logic directly.
Test Coverage
-------------
The new test provides:
- Legacy pass-through mode (identity mapping)
- Legacy translated mode with 4-level page table walks
- Root Entry Table and Context Entry Table configuration
- Complete 48-bit address space translation
- End-to-end DMA verification with memory validation
Testing
-------
QTEST_QEMU_BINARY=./build/qemu-system-x86_64 \
./build/tests/qtest/iommu-intel-test --tap -k
Thanks,
Fengyuan
Thanks for working on VT-d qtests. As a first-time patch contributor,
you’ve already done a great job with good cover letter and commit
messages to accurately summarize your work.
It seems that your code cannot be applied to the latest master branch.
And I also noticed multiple lines >80 columns (some > 90). QEMU style
says try to keep lines to 80 columns, only going a bit over when
wrapping would harm readability but never > 90 columns.
../tests/qtest/libqos/qos-intel-iommu.c: In function
‘qvtd_build_root_entry’:
../tests/qtest/libqos/qos-intel-iommu.c:168:31: error:
‘VTD_CONTEXT_ENTRY_SLPTPTR’ undeclared (first use in this function); did
you mean ‘VTD_CONTEXT_ENTRY_SSPTPTR’?
168 | lo = (context_table_ptr & VTD_CONTEXT_ENTRY_SLPTPTR) |
VTD_CONTEXT_ENTRY_P;
| ^~~~~~~~~~~~~~~~~~~~~~~~~
| VTD_CONTEXT_ENTRY_SSPTPTR
../tests/qtest/libqos/qos-intel-iommu.c:168:31: note: each undeclared
identifier is reported only once for each function it appears in
../tests/qtest/libqos/qos-intel-iommu.c: In function
‘qvtd_build_context_entry’:
../tests/qtest/libqos/qos-intel-iommu.c:190:25: error:
‘VTD_CONTEXT_ENTRY_SLPTPTR’ undeclared (first use in this function); did
you mean ‘VTD_CONTEXT_ENTRY_SSPTPTR’?
190 | (slptptr & VTD_CONTEXT_ENTRY_SLPTPTR);
| ^~~~~~~~~~~~~~~~~~~~~~~~~
| VTD_CONTEXT_ENTRY_SSPTPTR
../tests/qtest/libqos/qos-intel-iommu.c: In function
‘qvtd_setup_translation_tables’:
../tests/qtest/libqos/qos-intel-iommu.c:239:36: error: ‘VTD_SL_R’
undeclared (first use in this function); did you mean ‘VTD_SS_R’?
239 | pml4_entry = QVTD_PT_L3_BASE | VTD_SL_R | VTD_SL_W;
| ^~~~~~~~
| VTD_SS_R
../tests/qtest/libqos/qos-intel-iommu.c:239:47: error: ‘VTD_SL_W’
undeclared (first use in this function); did you mean ‘VTD_SS_W’?
239 | pml4_entry = QVTD_PT_L3_BASE | VTD_SL_R | VTD_SL_W;
| ^~~~~~~~
| VTD_SS_W
[92/2328] Compiling C object libblock.a.p/block.c.o
Besides when submitting patches it’s a good idea to read
`docs/devel/submitting-a-patch.rst`. It may be a bit long, but making
sure the code builds cleanly and running scripts/checkpatch.pl for style
checks beforehand can help avoid many basic issues.
Finally I think CC the experts on x86 and VT-d emulation might be a more
efficient way to review this sereis patch.
M: Paolo Bonzini <[email protected]>
R: Zhao Liu <[email protected]>
M: Michael S. Tsirkin <[email protected]>
R: Jason Wang <[email protected]>
R: Yi Liu <[email protected]>
R: Clément Mathieu--Drif <[email protected]>
Anyway I'm looking forward to your v2 patches.
Best regards,
Tao
