Excerpts from Nicholas Piggin's message of November 11, 2020 2:46 pm: > Excerpts from Christophe Leroy's message of November 10, 2020 9:19 pm: >> >> >> Le 10/11/2020 à 09:34, Nicholas Piggin a écrit : >>> Excerpts from Christophe Leroy's message of November 6, 2020 6:14 pm: >>>> >>>> >>>> Le 05/11/2020 à 15:34, Nicholas Piggin a écrit : >>>>> This also moves the 32s DABR match to C. >>>> >>>> Is there a real benefit doing this ? >>> >>> Oh I missed doing it, but yes I think bad_page_fault and do_break should >>> probably be implemented with the DEFINE_INTERRUT_HANDLER wrappers. >>> >> >> Yes, anyway, do we need to do that change ? Can't the dispatch between >> do_break() and page fault >> handling remain in handle_page_fault() ? What's the benefit of going into >> do_page_fault() and coming >> back ? > > You might be right, I'll take another look at it.
For 32-bit, we need to come back to save NV GPRs. Certainly the 64s code stays in do_page_fault because it always saves them. Now I don't think that's the nicest thing to go in and out of the interrupt wrappers twice in these cases, but for a first pass I think it's okay. Either we could add another type of error-case wrapper that does some adjustment if it becomes necessary, or we find a nice way to save NVGPRs from C code. If we could somehow parse unwind data to find where the NVGPRs are saved by the compiler and generate a little code stub to load them out, would be the ultimate :) Maybe one day... Thanks, Nick
