On Tue, 07 Sep 2021 21:38:27 +0200, Mark Kettenis wrote: > I'm not convinced the original diff is right: > > * We have several places in the kernel where we store numbers of pages > in a (32-bit) int. Changing just one of these places is dangerous. > > * Changing the type of just vm_dsize makes no sense. We should change > them all (but see the point above). > > * Does ASAN really need to reserve that much VA space?
The oddity here is that p_vm_dsize in kinfo_proc actually corresponds to vm_dused, not vm_dsize. So it is not actually the size of the data segment alone. Since uvmspace_dused() returns vsize_t it does seem like vm_dused should be sized similarly. As things stand, vm_dused could wrap. - todd
