On Wed, May 20, 2020 at 03:02:45PM +0530, Sai Prakash Ranjan wrote: > On 2020-05-19 20:41, Rob Clark wrote: > > On Tue, May 19, 2020 at 2:26 AM Sai Prakash Ranjan > > <saiprakash.ranjan@xxxxxxxxxxxxxx> wrote: > > > On 2020-05-18 21:15, Will Deacon wrote: > > > > So this sounds like an erratum to me, and I'm happy to set HUPCF if we > > > > detect the broken implementation. However, it will need an entry in > > > > Documentation/arm64/silicon-errata.rst and a decent comment in the > > > > driver > > > > to explain what we're doing and why. > > > > > > > > > > AFAIK there is no erratum documented internally for this behaviour and > > > this > > > exists from MSM8996 SoC time and errata usually don't survive this > > > long > > > across generation of SoCs and there is no point for us in disguising > > > it. > > > > possibly longer, qcom_iommu sets CFCFG.. > > > > Oh right, I was still in college when those SoCs were released ;) > > > > Is it OK if we clearly mention it as the "design limitation" or some > > > other > > > term which we can agree upon along with the description which Rob and > > > Jordan > > > provided for setting HUPCF in the driver when we add the set_hupcf > > > callback? > > > > I'm not too picky on what we call it, but afaict it has been this way > > since the beginning of time, rather than specific to a certain SoC or > > generation of SoCs. So it doesn't seem like the hw designers consider > > it a bug. > > > > (I'm not sure what the expected behavior is.. nor if any other SMMU > > implementation encounters this sort of situation..) > > Yes, that was my point as well that its probably not counted as a bug > by the hw designers. So I'm going to post setting HUPCF on QCOM > implementation with clear comments based on yours and Jordan's description > of this problem, but I wanted to have a way to set this only for GPU context > bank and not GMU as Jordan mentioned earlier that GMU doesnt need HUPCF set. > I was checking as to how do we map cb to device, if it was possible then we > can have > a compatibility thing like we did for identity mapping. Any ideas Robin? Right, see my reply over at: https://lore.kernel.org/r/20200520125700.GD25815@willie-the-truck Hopefully something like that can be made to work. Will
