On 07/05/2025 3:40 pm, Jason Gunthorpe wrote:
On Wed, May 07, 2025 at 07:29:51AM -0700, Rob Clark wrote:
From: Rob Clark <[email protected]>
In situations where mapping/unmapping sequence can be controlled by
userspace, attempting to map over a region that has not yet been
unmapped is an error. But not something that should spam dmesg.
Now that there is a quirk, we can also drop the selftest_running
flag, and use the quirk instead for selftests.
Signed-off-by: Rob Clark <[email protected]>
---
Sending v2 stand-alone, since I'm not quite ready to send a new
iteration of the full VM_BIND series. And with selftest_running
removed, I think this patch stands on it's own. (And maybe there
is still time to sneak this in for v6.16, removing an iommu dep
for the VM_BIND series in v6.17?)
You should just remove the warn on completely, no other driver does this.
What do you mean "no other driver"?
From a quick look, these IOMMU drivers have some sort of visible
warning related to the state of an existing PTE during map/unmap:
apple-dart
arm-smmu
arm-smmu-v3
arm-smmu-qcom
exynos-iommu
ipmmu_vmsa
msm_iommu
mtk_iommu
rockchip-iommu
s390-iommu
sun50i-iommu
qcom_iommu
while these IOMMU drivers do not:
amd
intel
mtk_iommu_v1
omap-iommu
riscv
sprd-iommu
tegra-smmu
virtio-iommu
By driver count, warning is literally the majority behaviour :/
Thanks,
Robin.
