Re: [PATCH 3/7] iommu/vt-d: Use device_block_translation() in dev_attach error path

[Baolu Lu]

On 2022/11/4 10:18, Tian, Kevin wrote:
> > From: Lu Baolu <baolu.lu@xxxxxxxxxxxxxxx>
> > Sent: Thursday, November 3, 2022 1:53 PM
> >
> > If domain attaching to device fails, the IOMMU driver should bring the
> > device to blocking DMA state. The upper layer is expected to recover it
> > by attaching a new domain. Use device_block_translation() in the error
> > path of dev_attach to make the behavior specific.
> >
> > The difference between device_block_translation() and the previous
> > dmar_remove_one_dev_info() is that the latter disables PCIe ATS and the
> > related PCIe features. This is unnecessary as these features are not per
> > domain capabilities, disabling them during domain switching is
> > unnecessary.
>
> well, the opposite argument is that when the DMA is blocked what is
> the point of enabling ATS/PRI on the device.

It's not "DMA is blocked", but "DMA without PASID is blocked". :-)

As above term implies, it's conceptually incorrect to disable ATS/PRI
only because DMA without PASID is about to blocked.

> btw this change is partial. @attach_dev still calls iommu_enable_pci_caps()
> which always tries to enable PCI capabilities w/o checking whether they
> have been enabled or not. Then user will hit -EBUSY when related PCI
> helpers are called.

Good catch!

How about moving iommu_enable/disable_pci_caps() into
iommu_probe/release_device() path? I may look into details if there's no
significant arch gaps.

> another difference worthy of pointing out is that in scalable mode it is
> the RID2PASID entry instead of context entry being cleared.

Yes. Will update the commit message.

Best regards,
baolu