Re: [PULL 25/93] tests/qtest: add RISC-V IOMMU MRIF notice tests

On Thu, 2026-08-20 at 07:17 -0700, Richard Henderson wrote:
> On 8/19/26 20:43, [email protected] wrote:
> > From: Flavien Solt <[email protected]>
> > 
> > Exercise MRIF-mode MSI writes through iommu-testdev with the
> > destination
> > interrupt-enable bit both clear and set. Each case requires the
> > pending
> > bit to be stored plus the configured notice ID to reach guest
> > memory.
> > 
> > The enable-clear case regresses the notice suppression fixed by the
> > previous patch. The enable-set case provides a control for the
> > existing
> > path.
> > 
> > Signed-off-by: Flavien Solt <[email protected]>
> > Reviewed-by: Chao Liu <[email protected]>
> > Message-ID: <[email protected]>
> > Signed-off-by: Alistair Francis <[email protected]>
> > ---
> >   tests/qtest/iommu-riscv-test.c | 102
> > +++++++++++++++++++++++++++++++++
> >   1 file changed, 102 insertions(+)
> 
> Something in here has an endian bug, as this new test fails on s390x:
> 
> https://gitlab.com/qemu-project/qemu/-/jobs/16009267995#L5106
> 
> ▶  694/1142 /riscv64/iommu-testdev/mrif-notice/enable-clear - 
> ERROR:../tests/qtest/iommu-riscv-
> test.c:349:run_riscv_iommu_mrif_notice: assertion failed 
> (pending & (1ull << RISCV_MRIF_INTID) != 0): (0 != 0) FAIL
> 
> Full output when running manually
> 
> # Found PCI device: vendor=0x1b36 device=0x0014 devfn=0x08
> # Found riscv-iommu-pci at devfn=0x08
> # RISC-V IOMMU MMIO base address: 0x40000000
> # Searching for iommu-testdev (vendor=0x1b36, device=0x0005)
> # Found PCI device: vendor=0x1b36 device=0x0005 devfn=0x10
> # MRIF notice: enable=0 pending=0x2000000000000000 notice=0x2d010000
> **
> ERROR:../src/tests/qtest/iommu-riscv-
> test.c:349:run_riscv_iommu_mrif_notice: assertion 
> failed (pending & (1ull << RISCV_MRIF_INTID) != 0): (0 != 0)
> not ok /riscv64/iommu-testdev/mrif-notice/enable-clear - 
> ERROR:../src/tests/qtest/iommu-riscv-
> test.c:349:run_riscv_iommu_mrif_notice: assertion 
> failed (pending & (1ull << RISCV_MRIF_INTID) != 0): (0 != 0)
> Bail out!
> Aborted (core dumped)
> 
> I assume pending was expected to equal 0x20, not the byte-reverse.

I have dropped the entire series from my PR and will send a v2.

Alistair

> 
> 
> r~