[PATCH] move __attribute__((__cold__)) functions back into final .text section

[PATCH] move __attribute__((__cold__)) functions back into final .text section

[Jan Beulich]

Due to the addition of __attribute__((__cold__)) to a few symbols
without adjusting the linker scripts, those symbols currently may end
up outside the [_stext,_etext) range, as they get placed in
.text.unlikely by (at least) gcc 4.3.0. This may confuse code not only
outside of the kernel, symbol_put_addr()'s BUG() could also trigger.
Hence we need to add .text.unlikely (and for future uses of
__attribute__((__hot__)) also .text.hot) to the TEXT_TEXT() macro.

Issue observed by Lukas Lipavsky.

Cc: Lukas Lipavsky <llipavsky@suse.cz>
Signed-off-by: Jan Beulich <jbeulich@novell.com>

---
 include/asm-generic/vmlinux.lds.h |    4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

--- linux-2.6.26-rc6/include/asm-generic/vmlinux.lds.h	2008-04-17 04:49:44.000000000 +0200
+++ 2.6.26-rc6-hot-cold-text/include/asm-generic/vmlinux.lds.h	2008-06-18 12:19:22.000000000 +0200
@@ -204,6 +204,7 @@
  * during second ld run in second ld pass when generating System.map */
 #define TEXT_TEXT							\
 		ALIGN_FUNCTION();					\
+		*(.text.hot)						\
 		*(.text)						\
 		*(.ref.text)						\
 		*(.text.init.refok)					\
@@ -213,7 +214,8 @@
 	CPU_KEEP(init.text)						\
 	CPU_KEEP(exit.text)						\
 	MEM_KEEP(init.text)						\
-	MEM_KEEP(exit.text)
+	MEM_KEEP(exit.text)						\
+		*(.text.unlikely)
 
 
 /* sched.text is aling to function alignment to secure we have same

Re: [PATCH] move __attribute__((__cold__)) functions back into final .text section

[Lukas Lipavsky]

Jan Beulich píše v St 18. 06. 2008 v 12:36 +0100:
> Due to the addition of __attribute__((__cold__)) to a few symbols
> without adjusting the linker scripts, those symbols currently may end
> up outside the [_stext,_etext) range, as they get placed in
> .text.unlikely by (at least) gcc 4.3.0. This may confuse code not only
> outside of the kernel, symbol_put_addr()'s BUG() could also trigger.
> Hence we need to add .text.unlikely (and for future uses of
> __attribute__((__hot__)) also .text.hot) to the TEXT_TEXT() macro.
> Cc: Lukas Lipavsky <llipavsky@suse.cz>
> Signed-off-by: Jan Beulich <jbeulich@novell.com>

Tested-by: Lukas Lipavsky <llipavsky@suse.cz>

Re: [PATCH] move __attribute__((__cold__)) functions back into final .text section

[Sam Ravnborg]

On Wed, Jun 18, 2008 at 12:36:01PM +0100, Jan Beulich wrote:
> Due to the addition of __attribute__((__cold__)) to a few symbols
> without adjusting the linker scripts, those symbols currently may end
> up outside the [_stext,_etext) range, as they get placed in
> .text.unlikely by (at least) gcc 4.3.0. This may confuse code not only
> outside of the kernel, symbol_put_addr()'s BUG() could also trigger.
> Hence we need to add .text.unlikely (and for future uses of
> __attribute__((__hot__)) also .text.hot) to the TEXT_TEXT() macro.
> 
> Issue observed by Lukas Lipavsky.
> 
> Cc: Lukas Lipavsky <llipavsky@suse.cz>
> Signed-off-by: Jan Beulich <jbeulich@novell.com>

Applied.

	Sam

Re: [PATCH] move __attribute__((__cold__)) functions back into final .text section

[Andrew Morton]

On Wed, 18 Jun 2008 12:36:01 +0100 "Jan Beulich" <jbeulich@novell.com> wrote:

> Due to the addition of __attribute__((__cold__)) to a few symbols
> without adjusting the linker scripts, those symbols currently may end
> up outside the [_stext,_etext) range, as they get placed in
> .text.unlikely by (at least) gcc 4.3.0. This may confuse code not only
> outside of the kernel, symbol_put_addr()'s BUG() could also trigger.
> Hence we need to add .text.unlikely (and for future uses of
> __attribute__((__hot__)) also .text.hot) to the TEXT_TEXT() macro.
> 
> Issue observed by Lukas Lipavsky.
> 
> Cc: Lukas Lipavsky <llipavsky@suse.cz>
> Signed-off-by: Jan Beulich <jbeulich@novell.com>
> 
> ---
>  include/asm-generic/vmlinux.lds.h |    4 +++-
>  1 file changed, 3 insertions(+), 1 deletion(-)
> 
> --- linux-2.6.26-rc6/include/asm-generic/vmlinux.lds.h	2008-04-17 04:49:44.000000000 +0200
> +++ 2.6.26-rc6-hot-cold-text/include/asm-generic/vmlinux.lds.h	2008-06-18 12:19:22.000000000 +0200
> @@ -204,6 +204,7 @@
>   * during second ld run in second ld pass when generating System.map */
>  #define TEXT_TEXT							\
>  		ALIGN_FUNCTION();					\
> +		*(.text.hot)						\
>  		*(.text)						\
>  		*(.ref.text)						\
>  		*(.text.init.refok)					\
> @@ -213,7 +214,8 @@
>  	CPU_KEEP(init.text)						\
>  	CPU_KEEP(exit.text)						\
>  	MEM_KEEP(init.text)						\
> -	MEM_KEEP(exit.text)
> +	MEM_KEEP(exit.text)						\
> +		*(.text.unlikely)
>  
>  
>  /* sched.text is aling to function alignment to secure we have same
> 

So this is now in Sam's
git+ssh://master.kernel.org/pub/scm/linux/kernel/git/sam/kbuild-next.git.
But it is not in the present linux-next.  Either because Sam added it
after Stephen pulled Sam's tree, or because something screwed up.

So I'll now be dropping it.  Sam, please ensure that it gets into
2.6.26 and 2.6.25.x (IMO).

Thanks.

Re: [PATCH] move __attribute__((__cold__)) functions back into final .text section

[Stephen Rothwell]

[-- Attachment #1: Type: text/plain, Size: 579 bytes --]

Hi Andrew,

On Tue, 1 Jul 2008 23:17:17 -0700 Andrew Morton <akpm@linux-foundation.org> wrote:
>
> So this is now in Sam's
> git+ssh://master.kernel.org/pub/scm/linux/kernel/git/sam/kbuild-next.git.
> But it is not in the present linux-next.  Either because Sam added it
> after Stephen pulled Sam's tree, or because something screwed up.

Its in the linux-next tree I have just (in the last 30 minutes) published
(but not yet announced) (tag next-20080702).

-- 
Cheers,
Stephen Rothwell                    sfr@canb.auug.org.au
http://www.canb.auug.org.au/~sfr/

[-- Attachment #2: Type: application/pgp-signature, Size: 197 bytes --]

Re: [PATCH] move __attribute__((__cold__)) functions back into final .text section

[Sam Ravnborg]

On Tue, Jul 01, 2008 at 11:17:17PM -0700, Andrew Morton wrote:
> On Wed, 18 Jun 2008 12:36:01 +0100 "Jan Beulich" <jbeulich@novell.com> wrote:
> 
> > Due to the addition of __attribute__((__cold__)) to a few symbols
> > without adjusting the linker scripts, those symbols currently may end
> > up outside the [_stext,_etext) range, as they get placed in
> > .text.unlikely by (at least) gcc 4.3.0. This may confuse code not only
> > outside of the kernel, symbol_put_addr()'s BUG() could also trigger.
> > Hence we need to add .text.unlikely (and for future uses of
> > __attribute__((__hot__)) also .text.hot) to the TEXT_TEXT() macro.
> > 
> > Issue observed by Lukas Lipavsky.
> > 
> > Cc: Lukas Lipavsky <llipavsky@suse.cz>
> > Signed-off-by: Jan Beulich <jbeulich@novell.com>
> > 
> > ---
> >  include/asm-generic/vmlinux.lds.h |    4 +++-
> >  1 file changed, 3 insertions(+), 1 deletion(-)
> > 
> > --- linux-2.6.26-rc6/include/asm-generic/vmlinux.lds.h	2008-04-17 04:49:44.000000000 +0200
> > +++ 2.6.26-rc6-hot-cold-text/include/asm-generic/vmlinux.lds.h	2008-06-18 12:19:22.000000000 +0200
> > @@ -204,6 +204,7 @@
> >   * during second ld run in second ld pass when generating System.map */
> >  #define TEXT_TEXT							\
> >  		ALIGN_FUNCTION();					\
> > +		*(.text.hot)						\
> >  		*(.text)						\
> >  		*(.ref.text)						\
> >  		*(.text.init.refok)					\
> > @@ -213,7 +214,8 @@
> >  	CPU_KEEP(init.text)						\
> >  	CPU_KEEP(exit.text)						\
> >  	MEM_KEEP(init.text)						\
> > -	MEM_KEEP(exit.text)
> > +	MEM_KEEP(exit.text)						\
> > +		*(.text.unlikely)
> >  
> >  
> >  /* sched.text is aling to function alignment to secure we have same
> > 
> 
> So this is now in Sam's
> git+ssh://master.kernel.org/pub/scm/linux/kernel/git/sam/kbuild-next.git.
> But it is not in the present linux-next.  Either because Sam added it
> after Stephen pulled Sam's tree, or because something screwed up.
> 
> So I'll now be dropping it.  Sam, please ensure that it gets into
> 2.6.26 and 2.6.25.x (IMO).

I have added a Cc: <stable@kernel.org> so the stable team should pick
it when applied to -linus.

	Sam