Re: [PATCH] PT_GNU_STACK

[Ian Lance Taylor <ian at airs dot com>]

Ulrich Drepper <drepper@redhat.com> writes:

> Ian Lance Taylor wrote:
> 
> > Because gcc would generate the call to it when it generated the
> > trampoline code, just as it already does on several architectures.
> > (For example, grep for TRANSFER_FROM_TRAMPOLINE in gcc/config/i386.)
> > [...]
> 
> > The cost is irritatingly high.  I object to this approach without more
> > consideration.
> 
> How can you write these two things in the same mail?  If anything is
> high, it's the cost of dynamically changing the permissions in the
> generated code.  Any runtime cost in unacceptable while a few more bytes
> in object files is absolutely no problem whatsoever.  There are no
> "irritatingly high" costs whatsoever.  The object file size varies more
> for each non-trivial file if you use a different compiler version.  We
> are talking about 8 bytes in the not section plus the section header
> table entry.

Every so often I need a reminder of why I avoid getting into
discussions with you.  Thanks for providing it.

If my choice is between silently accepting what I think is a mistake,
or fighting with you, I'll silently accept.  I will not comment
further on this issue.

Ian