Re: [PATCH v2 06/12] iommu/vt-d: Cleanup intel_context_flush_present()

[Baolu Lu]

On 3/5/25 11:34, Yi Liu wrote:
> On 2025/3/5 10:21, Baolu Lu wrote:
> > On 3/4/25 16:43, Yi Liu wrote:
> > > On 2025/2/24 13:16, Lu Baolu wrote:
> > > > The intel_context_flush_present() is called in places where either the
> > > > scalable mode is disabled, or scalable mode is enabled but all PASID
> > > > entries are known to be non-present. In these cases, the flush_domains
> > > > path within intel_context_flush_present() will never execute. This dead
> > > > code is therefore removed.
> > >
> > > The reason for this path is the remaining caller of
> > > intel_context_flush_present() is only the domain_context_clear_one() 
> > > which
> > > is called in legacy mode path. Is it?
> > > If so, it seems unnecessary to keep __context_flush_dev_iotlb(info); 
> > > in the
> > > end of the new intel_context_flush_present(). Also, since this helper 
> > > is more for legacy mode, might be good to move it out of pasid.c.:)
> >
> > This helper is for invalidating various caches when a context entry is
> > present and certain fields are changed. It is used in both legacy and
> > scalable modes.
>
> hmmm. the kdoc says all the pasid entries are non-present, is it necessary
> to flush dev_tlb in such scenario? I suppose no present pasid entry means
> no pagetable as well.

The spec has defined the software behavior for cache invalidation in
"Table 28: Guidance to Software for Invalidations". This helper was
written according to it.

> > In the past, this helper would work even if some PASIDs were still in
> > use. After the changes introduced in this series, this PASID-in-use case
> > is removed. So remove the dead code.
>
> yeah, I got this part. As I mentioned, the only caller now is the
> domain_context_clear_one() which is used in the legacy path. That's why I
> feel like it is now more for legacy mode path now after this series.

It's also called by device_pasid_table_teardown().

Thanks,
baolu