Re: msync() bug

[Linus Torvalds]

On Mon, 9 Jul 2001, Hugh Dickins wrote:
> >
> > that cannot happen, remap_pte_range only maps invalid pages or reserved
> > pages.
>
> Anyone know why mmap() of /dev/mem behaves in this way - solves the
> problem Andrew raises, but surely that could be solved in a better way?
> Seems strange that mmap() cannot give you what read() and write() can.

Simple: we MUST NOT muck around with the page counts of random pages.

And if the pages aren't marked Reserved, the page counts _would_ get
corrupted by the VM page handling.

Ergo: you can only mmap Reserved pages (or you have to create a nice
mapping and allocate the pages properly, which in the case of /dev/mem is
obviously not a possibility)

		Linus