[PATCH rc v5 1/4] iommu/arm-smmu-v3: Add update_safe bits to fix STE update sequence

[Jason Gunthorpe]

On Thu, Jan 15, 2026 at 08:25:05AM -0800, Nicolin Chen wrote:
> On Thu, Jan 15, 2026 at 09:11:51AM -0400, Jason Gunthorpe wrote:
> > On Tue, Jan 13, 2026 at 04:51:12PM -0400, Jason Gunthorpe wrote:
> > > > -       safe_bits[1] |= cpu_to_le64(STRTAB_STE_1_EATS);
> > > > +       if (!((cur[2] | target[2]) & cpu_to_le64(STRTAB_STE_2_S2S)))
> > > > +               safe_bits[1] |= cpu_to_le64(
> > > > +                       FIELD_PREP(STRTAB_STE_1_EATS, STRTAB_STE_1_EATS_TRANS));
> > > > --------------------------------------------------------------------------
> > > > 
> > > > @will, does this look good to you? I can send a v7 with this.
> > > 
> > > That is an easy way to address Will's observation, makes sense to me.
> > 
> > Ah, but it looks like it can generate an errant view of a EATS that is
> > neither old or new. Ie value 3, reserved.
> > 
> > I think you should just check if old or new has EATS bit 1 set:
> > 
> > if (!((cur[2] | target[2]) & cpu_to_le64(STRTAB_STE_2_S2S)) &&
> >     !((cur[1] | target[1]) & cpu_to_le64(FIELD_PREP(STRTAB_STE_1_EATS, 2))))
> > 
> > Which the current driver never does..
> 
> The EATS field is completely controlled by the driver. So, we are
> safe for now, right?
> 
> Should we add this when the driver has the actual support for the
> split stage thing?

If we have figured it out now I would add it because it would be a big
leap to think the next person will remember about this detail..

But yes, this and the S2S thing don't effect the driver as it is now,
it is just doing work to help future people.

Jason