Linus Torvalds <torva...@linux-foundation.org> writes: > On Fri, Jun 5, 2020 at 9:38 AM Michael Ellerman <m...@ellerman.id.au> wrote: >> >> I've pushed the result of my resolution of the conflicts to the powerpc/merge >> branch, if you want to look at that, though I've also tried to describe it in >> full below. > > I ended up doing the machine_check_exception() differently, because I > felt the code itself was done wrong and I wanted to add a note about > that. > > Having the same function have completely different semantics depending > on a platform issue is just fundamentally wrong, and makes not just > for fragile code, but also means that you can't do single image > kernels.
Yeah I agree it's not that nicely structured. In this case CONFIG_PPC_BOOK3S_64 is one of our top-level compile time switches anyway, so the single image thing at least isn't a concern. > It should be two different functions, possibly just > > non_nmi_fn() { ... } > > nmi_fn() { nmi_enter(); non_nmi_fn(); nmi_exit(); } > > and now you don't have odd rules for the same function that depends on > how the platform happens to call it. Agreed. > I didn't do the above. I did something that looked like the old code, > but had a comment. Oh well. Thanks, we can restructure it later. > But thanks for describing the merge, I'd have missed the place where > there was a new use of pgd_oiffset(). Yeah I missed it to begin with :} > ..and then when I actually compared whether I otherwise got the same > result as you, I realized that this all depends on the module tree. > > I'll go merge that first, and then re-do this all. Oh well. Darn it. I figured you'd have merged that by the time you saw my pull, will mention anything similar in future. cheers