On 10/2/25 15:38, Daniel Gimpelevich wrote: > On Thu, 2025-10-02 at 14:55 -0700, Dave Hansen wrote: >> That's not a bad idea. Or, even if you can pick two amenable >> architectures to start with it will make it really obvious that this is >> useful. Two architectures means a *lot*, IMNHO. Two is a billion times >> better than one. > I think it's a bad idea, if I understand it correctly. The patchset > conceptually patches a mechanism of the kernel as a whole, but one which > just so happens to need to be implemented separately for each arch. > Breaking it down like you suggest creates an embarrassingly high > likelihood of different architectures' implementations of it going out > of sync, a previous situation that this patchset was partly intended to > address. I say keep it atomic. If it breaks on an arch or two but not > others and nobody notices right away, that would be better addressed > with a new patch when someone eventually does notice. Just my 2¢…
How is the approach to "keep it atomic" working out so far? ;) The kernel isn't exactly developed in secret. It's also not hard at all to, say, once a week to peek at linux-next and do a lore search (or use lei) if anyone is desperately worried about the ~50 lines per architecture going out of sync.
