On Fri, May 22, 2026 at 09:48:35PM +0800, Barry Song wrote: > On Fri, May 22, 2026 at 9:36 PM Barry Song <[email protected]> wrote: > > > > On Fri, May 22, 2026 at 9:09 PM Matthew Wilcox <[email protected]> wrote: > > > > > > On Fri, May 22, 2026 at 10:33:05AM +0800, Barry Song (Xiaomi) wrote: > > > > need to touch `filemap.c` at all (probably because you are already > > > > maintaining `filemap.c` perfectly): > > > > > > I'm going to give you one chance to apologise for that. > > > > Apologies if my wording caused any misunderstanding. > > That was not my intention at all. > > > > What I meant is that filemap.c already has a very > > solid design. > > > > For memory.c, I had to touch several places for the > > blacklist; otherwise, the kernel would hang. > > > > But for filemap.c, I basically didn't need to touch > > anything, and preliminary testing shows no issues after > > moving it from the whitelist to the blacklist. This is > > Sorry, I feel I may be causing some misunderstanding > again. > > By "whitelist", I mean I used to allow certain cases > to use per-vma retry. > > By "blacklist", I mean I am now moving to disallow > certain cases from using per-vma retry. > > Right now, I have to add several cases in memory.c > to the blacklist; otherwise, the kernel would hang. > > But it seems that everything in filemap.c is fine so > far based on testing. > > I'm not sure if I've explained things clearly. Please > let me know if anything is still unclear or insufficient.
Barry - this thread is completely out of hand and getting _rapidly_ unproductive. It's certainly about as clear as mud where we stand right now, so here's my suggestion - let's just stop adding to the noise here :) and instead, you take the approach suggested by Suren at LSF and send that as an _RFC_ series. That way we can look at that and hopefully actually circle in on a solution rather than have endless sub threads and sub discussions :) It's far too sunny out in the UK right now for that ;) Thanks, Lorenzo
