On Mon, Aug 5, 2024 at 5:13 PM Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx> wrote: > > On Mon, 5 Aug 2024 at 16:25, Nicholas Piggin <npiggin@xxxxxxxxx> wrote: > > > > Can userspace on other archs not unmap their vdsos? > > I think they can, and nobody cares. The "context.vdso" value stays at > some stale value, and anybody who tries to use it will just fail. > I want to seal the vdso :-), so I also care (not having it changeable from userspace) For the restore scenario, if vdso is sealed, I guess CRIU won't be able to relocate the vdso from userspace, I 'm interested in hearing vdso dev's input on this , e.g. is that possible to make CRIU compatible with memory sealing. > So what makes powerpc special is not "you can unmap the vdso", but > "powerpc cares". > > I just don't quite know _why_ powerpc cares. > > Judging by the comments and a quick 'grep', the reason may be > > arch/powerpc/perf/callchain_32.c > > which seems to have some vdso knowledge. > > But x86 does something kind of like that at signal frame generation > time, and doesn't care. > > I really think it's an issue of "if you screw with the vdso, you get > to keep both broken pieces". > > Linus
