On Wed, Jun 5, 2019 at 6:18 AM Marek Szyprowski <m.szyprowski@xxxxxxxxxxx> wrote: > > Hi Rob, > > On 2019-06-05 14:57, Rob Clark wrote: > > On Tue, Jun 4, 2019 at 11:58 PM Tomasz Figa <tfiga@xxxxxxxxxxxx> wrote: > >> But first of all, I remember Marek already submitted some patches long > >> ago that extended struct driver with some flag that means that the > >> driver doesn't want the IOMMU to be attached before probe. Why > >> wouldn't that work? Sounds like a perfect opt-out solution. > > Actually, yeah.. we should do that. That is the simplest solution. > > Tomasz has very good memory. It took me a while to find that old patches: > > https://patchwork.kernel.org/patch/4677251/ > https://patchwork.kernel.org/patch/4677941/ > https://patchwork.kernel.org/patch/4677401/ > > It looks that my idea was a bit ahead of its time ;) > if I re-spin this, was their a reason not to just use bitfields, ie: - bool suppress_bind_attrs; /* disables bind/unbind via sysfs */ + bool suppress_bind_attrs : 1; /* disables bind/unbind via sysfs */ + bool has_own_iommu_manager : 1; /* driver explictly manages IOMMU */ That seems like it would have been a bit less churn and a bit nicer looking (IMO at least) BR, -R
