On Sun, Oct 20, 2024 at 03:16:11PM -0400, Kent Overstreet wrote: > On Sun, Oct 20, 2024 at 12:09:58PM -0700, Linus Torvalds wrote: > > On Sun, 20 Oct 2024 at 12:09, Linus Torvalds > > <torvalds@xxxxxxxxxxxxxxxxxxxx> wrote: > > > > > > How about you limit the amount of memory you use in the first place instead? > > > > .. and to clarify: we're not making other parts of the kernel less > > robust because *you* are doing something stupid and odd. > > Except, vmalloc() already behaves this way - so it seems to me you > already have. > > I've already added a stupid workaround to the darray code to switch to > calling vmalloc() directly, when necessary; this patch was a courtesy > because if bcachefs is hitting this limit no doubt other things will be > soon as well. > I was thinking to prevent "big" allocations to limit the vmalloc() by the INT_MAX sizes, i.e. to apply same limitation as kvmalloc() has. vmalloc() is stick to totalram_pages() which is way a lot. But it would break bcachefs, as i see it. -- Uladzislau Rezki
