On 06/19/2018 06:57 PM, Dan Williams wrote: > On Tue, Jun 19, 2018 at 6:34 PM, John Hubbard <jhubbard@xxxxxxxxxx> wrote: >> On 06/19/2018 06:24 PM, Dan Williams wrote: >>> On Tue, Jun 19, 2018 at 11:11 AM, John Hubbard <jhubbard@xxxxxxxxxx> wrote: >>>> On 06/19/2018 03:41 AM, Jan Kara wrote: >>>>> On Tue 19-06-18 02:02:55, Matthew Wilcox wrote: >>>>>> On Tue, Jun 19, 2018 at 10:29:49AM +0200, Jan Kara wrote: >>> [..] >>>>> And then there's the aspect that both these approaches are a bit too >>>>> heavyweight for some get_user_pages_fast() users (e.g. direct IO) - Al Viro >>>>> had an idea to use page lock for that path but e.g. fs/direct-io.c would have >>>>> problems due to lock ordering constraints (filesystem ->get_block would >>>>> suddently get called with the page lock held). But we can probably leave >>>>> performance optimizations for phase two. >>>> >>>> >>>> So I assume that phase one would be to apply this approach only to >>>> get_user_pages_longterm. (Please let me know if that's wrong.) >>> >>> I think that's wrong, because get_user_pages_longterm() is only a >>> filesystem-dax avoidance mechanism, it's not trying to address all the >>> problems that Jan is talking about. I don't see any viable half-step >>> solutions. >>> >> >> OK, but in that case, I'm slightly confused by Jan's comment above, about leaving >> performance optimizations until phase two. Because that *is* a half-step approach: >> phase one, phase two. > > No, sorry, I might be confusing things. The half step is leaving > truncate broken, or my strawman that only addressed unmap. > >> Are you disagreeing with Jan, or are you suggesting "fix get_user_pages first, and >> leave get_user_pages_fast alone for now?" > > I'm agreeing with Jan, we need to fix page_mkclean() and > try_to_unmap() without regressing truncate behavior. > OK, perfect, thanks for clarifying. It all sounds consistent now. :)
