Re: Non-Exec stack patches

[Jakub Jelinek <jakub@redhat.com>]

On Wed, Mar 24, 2004 at 09:12:30AM -0800, David Mosberger wrote:
>   David> That's why there is mprotect().
> 
>   John> But mprotect() costs enough (hundreds of cycles) to be a
>   John> significant burden in some cases.  Generating code to a stack
>   John> region that is inherently executable is inexpensive (even
>   John> allowing for restrictive alignment and avoiding I/D cache
>   John> conflicts), is thread safe, is async-signal safe, and takes
>   John> less work than other alternatives.  Yes, the "black hats" do
>   John> this; so do the "white hats."  Please do not increase the
>   John> minimum cost for applications that want generate-and-execute
>   John> on the stack at upredictable high frequency.
> 
> Huh?  Only one mprotect() call is needed to make the entire stack
> executable.

Nope.  Think about multithreaded apps.  Furthermore, getting the exact
extents of the particular stack is difficult to find for applications,
but e.g. the threading library has to know such things.

	Jakub