I want to reserve a tiny bit of the address space just below 1<<47 on x86_64 for kernel purposes but without stealing away management of the page tables. It seems like the way to do that is to set USER_PGTABLES_CEILING to 0 and then make some adjustment to exit_mmap() to free the tables on exit. The problem is that free_pgd_range(), free_pgtables, etc are quite opaque to me, and I'm having a hard time understanding the pagetable freeing code. Some questions I haven't figured out: - What is the intended purpose of addr, end, floor, and ceiling? What are the pagetable freeing functions actually *supposed* to do? - Are there any invariants that, for example, there is never a pagetable that doesn't have any vmas at all under it? I can understand how all the code would be correct if this invariant were to exist, but I don't see what would preserve it. But maybe free_pgd_range(), etc really do preserve it. - What keeps mm->mmap pointing to the lowest-addressed vma? I see lots of code that seems to assume that you can start at mm->mmap, follow the vm_next links, and find all vmas, but I can't figure out why this would work. - What happens if a process exits while mm->mmap is NULL? - Is there any piece of code that makes it obvious that all the pagetables are gone by the time the exit_mmap() finishes? Because I'm staring to wonder whether some weird combination of maps and unmaps will just leak pagetables, and the code is rather complicated, subtle, and completely lacking in documentation, and I've learned to be quite suspicious of such things. --Andy -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxx. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>
