Hello Joao,
I think you should first convert all PHBs to PCIIOMMUOps to avoid all the
tests as below and adapt pci_setup_iommu_ops() with the new parameter.
OK, that's Yi's original patch:
https://lore.kernel.org/all/20210302203827.437645-5-yi.l....@intel.com/
I went with this one is that 1) it might take eons to get every single IOMMU
maintainer ack; and 2) it would allow each IOMMU to move at its own speed
specially as I can't test most of the other ones. essentially iterative, rather
than invasive change? Does that make sense?
I think it is ok to make global changes to replace a function by a struct
of ops. This is not major (unless the extra indirection has a major perf
impact on some platforms).
It should be a mechanical change. As the pci_setup_iommu_ops() should be
functionally equivalent to pci_setup_iommu() [...]
Thanks for going back to the previous proposal.
Getting acks from everyone will be difficultsince some PHBs are orphans.
[...] This is what gets me a bit hesitant
orphans shouldn't be an issue, nor the PPC emulated machines. We will see
what other maintainers have to say.
Thanks,
C.
