Re: [PATCH 3/7] iommu/arm-smmu: Add tlb_sync implementation hook

[Robin Murphy]

On 30/08/2019 23:49, Krishna Reddy wrote:
> > > +	if (smmu->impl->tlb_sync) {
> > > +		smmu->impl->tlb_sync(smmu, page, sync, status);
>
> > What I'd hoped is that rather than needing a hook for this, you could just override smmu_domain->tlb_ops from .init_context to wire up the alternate .sync method directly. That would save this extra level of indirection.
>
> Hi Robin,  overriding tlb_ops->tlb_sync function is not enough here.
> There are direct references to arm_smmu_tlb_sync_context(),  arm_smmu_tlb_sync_global() functions.
> In arm-smmu.c.  we can replace these direct references with tlb_ops->tlb_sync() function except in one function arm_smmu_device_reset().
> When arm_smmu_device_reset() gets called, domains are not initialized and tlb_ops is not available.
> Should we add a new hook for arm_smmu_tlb_sync_global() or make this as a responsibility of impl->reset() hook as it gets
> called at the same place?

Ah, right, I'd forgotten about the TLB maintenance on reset. Looking 
again, though, I think you might need to implement impl->reset anyway 
for the sake of clearing GFSR correctly - since the value read from the 
base instance technically may not match whatever bits might happen to be 
set in the other instances - so I don't see any issue with just calling 
nsmmu_tlb_sync() from there as well.

Robin.