Re: serious performance regression due to NX patch

[Ingo Molnar]

On Sun, 11 Jul 2004, Jakub Jelinek wrote:

> > --- linux/fs/binfmt_elf.c.orig3	
> > +++ linux/fs/binfmt_elf.c	
> > @@ -627,8 +627,14 @@ static int load_elf_binary(struct linux_
> >  				executable_stack = EXSTACK_DISABLE_X;
> >  			break;
> >  		}
> > +#ifdef __i386_
> > +	/*
> > +	 * Legacy x86 binaries have an expectation of executability for
> > +	 * virtually all their address-space - turn executability on:
> > +	 */
> >  	if (i == elf_ex.e_phnum)
> >  		def_flags |= VM_EXEC | VM_MAYEXEC;
> > +#endif
> 
> This looks incorrect. There are many arches where legacy binaries expect
> the executability for virtually all their address-space (my guess is all
> but x86-64 and ia64), and even on those two legacy binaries expected at
> least stack executable.

so ... this should be #ifndef ia64?

to all the purists: this cannot be done via VM_DATA_DEFAULT_FLAGS, because
some architectures enforce the X bit, some dont. In fact only ia64 seems
to enforce it reliably for all binaries.

the #ifdef could be made an arch inline or define. But it's really
academic as only ia64 seems to have this problem. So i'd suggest the patch
below.

	Ingo

--- linux/fs/binfmt_elf.c.orig
+++ linux/fs/binfmt_elf.c
@@ -627,8 +627,14 @@ static int load_elf_binary(struct linux_
 				executable_stack = EXSTACK_DISABLE_X;
 			break;
 		}
+#ifndef __ia64__
+	/*
+	 * Legacy binaries (except ia64) have an expectation of executability
+	 * for virtually all their address-space - turn executability on:
+	 */
 	if (i == elf_ex.e_phnum)
 		def_flags |= VM_EXEC | VM_MAYEXEC;
+#endif
 
 	/* Some simple consistency checks for the interpreter */
 	if (elf_interpreter) {
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/