En réponse à Cedric Degea <cdegea@xxxxxxxxxxxxxxxx>: > Rudy Gingles <rudyatek@xxxxxxxxxxxxxx> wrote: > > >Anyway. What I wanted to ask you all is this - does anyone know of > > >MAP_COPY or MAP_PRIVATE being used in real software? I can write > > > I always saw mmap as private, until it was specifically shared, in > > which case you are correct about keeping all changes in real-time. > > > > As far as MAP_COPY, the only big usage that comes to mind is POSIX > > and fork(). But even elsewhere, technically, there could be many > > cases where you pass a buffer to another app, and use MAP_COPY rather > > > than specifically making a copy yourself. Just an idea. > > Kernel newbie question: what will the kernelland counterpart > to userland mmap() be called? (the BeOS current map_memory() > call used by accelerant's drivers I mean). > > At any rate, that one will definitely need the "straight forward" > version of mapping, in the context of accelerants (ie in the context > of mapping "memory" to memory), no funky copy-on-write feature there. > I think you mean map_physical_memory() ? This should still be provided, and isn't really exactly the same (even if QNX does what map_physical_memory() with mmap() and a special flag, but this is for userland stuff). François.