Re: [PATCH/RFC] Shared page tables

[Dave McCracken <dmccr@us.ibm.com>]

--On Tuesday, January 24, 2006 08:06:37 +0100 Arjan van de Ven
<arjan@infradead.org> wrote:

>> The randomization is not for cache coloring, but for security purposes
>> (except for the old very small stack randomization that was used
>> to avoid conflicts on HyperThreaded CPUs). I would be surprised if the
>> mmap made much difference because it's page aligned and at least
>> on x86 the L2 and larger caches are usually PI.
> 
> randomization to a large degree is more important between machines than
> within the same machine (except for setuid stuff but lets call that a
> special category for now). Imo prelink is one of the better bets to get
> "all code for a binary/lib on the same 2 mb page", all distros ship
> prelink nowadays anyway (it's too much of a win that nobody can afford
> to not ship it ;) and within prelink the balance between randomization
> for security and 2Mb sharing can be struck best. In fact it needs know
> about the 2Mb thing anyway to place it there properly and for all
> binaries... the kernel just can't do that.

Currently libc and most other system libraries have text segments smaller
than 2MB, so they won't share anyway.  We can't even coalesce adjacent
libraries since the linker puts unshareable data space after each library.

The main win for text sharing is applications with large text in the
program itself.  As long as that's loaded at the same address we'll share
page tables for it.

I thought the main security benefit for randomization of mapped regions was
for writeable data space anyway.  Isn't text space protected by not being
writeable?

Dave McCracken

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>