xen/x86: unable to build with gcc5

xen/x86: unable to build with gcc5

[Ian Campbell]

This was originally reported against Xen 4.4 in
http://bugs.debian.org/778181 but I have reproduced with current staging
using debug=n. Compiler I used is:
        $ gcc --version
        gcc (Debian 5-20150205-1) 5.0.0 20150205 (experimental) [trunk revision 220455]

Error: size of boot.o:.text is 0x026c
/build/gcc5/xen.git/xen/Rules.mk:180: recipe for target 'boot.init.o' failed

$ objdump -t ./xen/arch/x86/efi/boot.o | grep -P '[^t]\.text[^\.]'
0000000000000000 l    d  .text	0000000000000000 .text
0000000000000000 l     F .text	00000000000000e9 pre_parse.constprop.9
00000000000000f0 l     F .text	0000000000000123 get_value.constprop.10
0000000000000220 l     F .text	000000000000004c set_color.part.4.constprop.11

pre_parse, get_value and set_color are all annotated with __init, but I
guess this is something else. Any idea what to do? Possibly a gcc bug?

Building with -k -s I also see:

Error: size of unxz.o:.text is 0x06b
->
0000000000000000 l    d  .text	0000000000000000 .text
0000000000000000 l     F .text	000000000000006b dec_vli.isra.1

Error: size of dmi_scan.o:.text is 0x018f
->
 grep -P '[^t]\.text\b'
0000000000000000 l    d  .text	0000000000000000 .text
0000000000000000 l     F .text	000000000000018f _dmi_iterate.constprop.6

And:

symbols.c: In function 'symbols_lookup':
symbols.c:128:359: error: array subscript is above array bounds [-Werror=array-bounds]
symbols.c:136:176: error: array subscript is above array bounds [-Werror=array-bounds]

from common/symbols.c. I suppose because of
        extern const unsigned int symbols_offsets[1]
        #define symbols_address(n) (SYMBOLS_ORIGIN + symbols_offsets[n])

I'm not sure why that needs to be a singleton array instead of []?

Ian.

Re: xen/x86: unable to build with gcc5

[Jan Beulich]

>>> On 13.03.15 at 13:00, <ian.campbell@citrix.com> wrote:
> This was originally reported against Xen 4.4 in
> http://bugs.debian.org/778181 but I have reproduced with current staging
> using debug=n. Compiler I used is:
>         $ gcc --version
>         gcc (Debian 5-20150205-1) 5.0.0 20150205 (experimental) [trunk 
> revision 220455]
> 
> Error: size of boot.o:.text is 0x026c
> /build/gcc5/xen.git/xen/Rules.mk:180: recipe for target 'boot.init.o' failed
> 
> $ objdump -t ./xen/arch/x86/efi/boot.o | grep -P '[^t]\.text[^\.]'
> 0000000000000000 l    d  .text	0000000000000000 .text
> 0000000000000000 l     F .text	00000000000000e9 pre_parse.constprop.9
> 00000000000000f0 l     F .text	0000000000000123 get_value.constprop.10
> 0000000000000220 l     F .text	000000000000004c set_color.part.4.constprop.11
> 
> pre_parse, get_value and set_color are all annotated with __init, but I
> guess this is something else. Any idea what to do? Possibly a gcc bug?
> 
> Building with -k -s I also see:
> 
> Error: size of unxz.o:.text is 0x06b
> ->
> 0000000000000000 l    d  .text	0000000000000000 .text
> 0000000000000000 l     F .text	000000000000006b dec_vli.isra.1
> 
> Error: size of dmi_scan.o:.text is 0x018f
> ->
>  grep -P '[^t]\.text\b'
> 0000000000000000 l    d  .text	0000000000000000 .text
> 0000000000000000 l     F .text	000000000000018f _dmi_iterate.constprop.6

For all of the above, at a first glance I'd say compiler bug. But to
be sure, could you send the three .o-s?

> And:
> 
> symbols.c: In function 'symbols_lookup':
> symbols.c:128:359: error: array subscript is above array bounds 
> [-Werror=array-bounds]
> symbols.c:136:176: error: array subscript is above array bounds 
> [-Werror=array-bounds]
> 
> from common/symbols.c. I suppose because of
>         extern const unsigned int symbols_offsets[1]
>         #define symbols_address(n) (SYMBOLS_ORIGIN + symbols_offsets[n])
> 
> I'm not sure why that needs to be a singleton array instead of []?

I don't see a reason other than someone having wanted to be
standard C compliant (which we don't need to be here).

Jan

Re: xen/x86: unable to build with gcc5

[Ian Campbell]

[-- Attachment #1: Type: text/plain, Size: 1990 bytes --]

On Fri, 2015-03-13 at 12:10 +0000, Jan Beulich wrote:
[...]
> For all of the above, at a first glance I'd say compiler bug. But to
> be sure, could you send the three .o-s?

Attached.

> > symbols.c: In function 'symbols_lookup':
> > symbols.c:128:359: error: array subscript is above array bounds 
> > [-Werror=array-bounds]
> > symbols.c:136:176: error: array subscript is above array bounds 
> > [-Werror=array-bounds]
> > 
> > from common/symbols.c. I suppose because of
> >         extern const unsigned int symbols_offsets[1]
> >         #define symbols_address(n) (SYMBOLS_ORIGIN + symbols_offsets[n])
> > 
> > I'm not sure why that needs to be a singleton array instead of []?
> 
> I don't see a reason other than someone having wanted to be
> standard C compliant (which we don't need to be here).

I've switched to Keir's current address in case he has a comment, but
the following fixes it for me:

8<-------------

>From fbe732e5be6352afe2d9302f5d56439b9b9b6dee Mon Sep 17 00:00:00 2001
From: Ian Campbell <ian.campbell@citrix.com>
Date: Fri, 13 Mar 2015 12:22:25 +0000
Subject: [PATCH] xen: common: Use unbounded array for symbols_offset.

Using a singleton array causes gcc5 to report:
symbols.c: In function 'symbols_lookup':
symbols.c:128:359: error: array subscript is above array bounds [-Werror=array-bounds]
symbols.c:136:176: error: array subscript is above array bounds [-Werror=array-bounds]

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
---
 xen/common/symbols.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/xen/common/symbols.c b/xen/common/symbols.c
index bc2fde6..fc7c9e7 100644
--- a/xen/common/symbols.c
+++ b/xen/common/symbols.c
@@ -19,7 +19,7 @@
 #include <xen/spinlock.h>
 
 #ifdef SYMBOLS_ORIGIN
-extern const unsigned int symbols_offsets[1];
+extern const unsigned int symbols_offsets[];
 #define symbols_address(n) (SYMBOLS_ORIGIN + symbols_offsets[n])
 #else
 extern const unsigned long symbols_addresses[];
-- 
2.1.4




[-- Attachment #2: boot.o --]
[-- Type: application/x-object, Size: 231648 bytes --]

[-- Attachment #3: dmi_scan.o --]
[-- Type: application/x-object, Size: 75560 bytes --]

[-- Attachment #4: unxz.o --]
[-- Type: application/x-object, Size: 152144 bytes --]

[-- Attachment #5: Type: text/plain, Size: 126 bytes --]

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel

Re: xen/x86: unable to build with gcc5

[Jan Beulich]

>>> On 13.03.15 at 13:25, <ian.campbell@citrix.com> wrote:
> From: Ian Campbell <ian.campbell@citrix.com>
> Date: Fri, 13 Mar 2015 12:22:25 +0000
> Subject: [PATCH] xen: common: Use unbounded array for symbols_offset.
> 
> Using a singleton array causes gcc5 to report:
> symbols.c: In function 'symbols_lookup':
> symbols.c:128:359: error: array subscript is above array bounds 
> [-Werror=array-bounds]
> symbols.c:136:176: error: array subscript is above array bounds 
> [-Werror=array-bounds]
> 
> Signed-off-by: Ian Campbell <ian.campbell@citrix.com>

Acked-by: Jan Beulich <jbeulich@suse.com>

> ---
>  xen/common/symbols.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/xen/common/symbols.c b/xen/common/symbols.c
> index bc2fde6..fc7c9e7 100644
> --- a/xen/common/symbols.c
> +++ b/xen/common/symbols.c
> @@ -19,7 +19,7 @@
>  #include <xen/spinlock.h>
>  
>  #ifdef SYMBOLS_ORIGIN
> -extern const unsigned int symbols_offsets[1];
> +extern const unsigned int symbols_offsets[];
>  #define symbols_address(n) (SYMBOLS_ORIGIN + symbols_offsets[n])
>  #else
>  extern const unsigned long symbols_addresses[];
> -- 
> 2.1.4

Re: xen/x86: unable to build with gcc5

[Jan Beulich]

>>> On 13.03.15 at 13:00, <ian.campbell@citrix.com> wrote:
> Error: size of boot.o:.text is 0x026c
> /build/gcc5/xen.git/xen/Rules.mk:180: recipe for target 'boot.init.o' failed
> 
> $ objdump -t ./xen/arch/x86/efi/boot.o | grep -P '[^t]\.text[^\.]'
> 0000000000000000 l    d  .text	0000000000000000 .text
> 0000000000000000 l     F .text	00000000000000e9 pre_parse.constprop.9
> 00000000000000f0 l     F .text	0000000000000123 get_value.constprop.10
> 0000000000000220 l     F .text	000000000000004c set_color.part.4.constprop.11
> 
> pre_parse, get_value and set_color are all annotated with __init, but I
> guess this is something else. Any idea what to do? Possibly a gcc bug?
> 
> Building with -k -s I also see:
> 
> Error: size of unxz.o:.text is 0x06b
> ->
> 0000000000000000 l    d  .text	0000000000000000 .text
> 0000000000000000 l     F .text	000000000000006b dec_vli.isra.1
> 
> Error: size of dmi_scan.o:.text is 0x018f
> ->
>  grep -P '[^t]\.text\b'
> 0000000000000000 l    d  .text	0000000000000000 .text
> 0000000000000000 l     F .text	000000000000018f _dmi_iterate.constprop.6

The same with these other two cases too - they're properly
annotated. I.e. the compiler forgets to propagate the section
attribute when generating variants of the functions (e.g. after
having detected that it always gets called with some argument
set to some specific value). Do you want to report this as a bug
(assuming there isn't one already; I for one can't find any) from
the Debian side, or should I open one?

Jan

Re: xen/x86: unable to build with gcc5

[M A Young]

On Fri, 13 Mar 2015, Ian Campbell wrote:

> On Fri, 2015-03-13 at 12:10 +0000, Jan Beulich wrote:
> [...]
> > For all of the above, at a first glance I'd say compiler bug. But to
> > be sure, could you send the three .o-s?

See https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64963

	Michael Young

Re: xen/x86: unable to build with gcc5

[Ian Campbell]

On Fri, 2015-03-13 at 13:25 +0000, Jan Beulich wrote:
> >>> On 13.03.15 at 13:00, <ian.campbell@citrix.com> wrote:
> > Error: size of boot.o:.text is 0x026c
> > /build/gcc5/xen.git/xen/Rules.mk:180: recipe for target 'boot.init.o' failed
> > 
> > $ objdump -t ./xen/arch/x86/efi/boot.o | grep -P '[^t]\.text[^\.]'
> > 0000000000000000 l    d  .text	0000000000000000 .text
> > 0000000000000000 l     F .text	00000000000000e9 pre_parse.constprop.9
> > 00000000000000f0 l     F .text	0000000000000123 get_value.constprop.10
> > 0000000000000220 l     F .text	000000000000004c set_color.part.4.constprop.11
> > 
> > pre_parse, get_value and set_color are all annotated with __init, but I
> > guess this is something else. Any idea what to do? Possibly a gcc bug?
> > 
> > Building with -k -s I also see:
> > 
> > Error: size of unxz.o:.text is 0x06b
> > ->
> > 0000000000000000 l    d  .text	0000000000000000 .text
> > 0000000000000000 l     F .text	000000000000006b dec_vli.isra.1
> > 
> > Error: size of dmi_scan.o:.text is 0x018f
> > ->
> >  grep -P '[^t]\.text\b'
> > 0000000000000000 l    d  .text	0000000000000000 .text
> > 0000000000000000 l     F .text	000000000000018f _dmi_iterate.constprop.6
> 
> The same with these other two cases too - they're properly
> annotated. I.e. the compiler forgets to propagate the section
> attribute when generating variants of the functions (e.g. after
> having detected that it always gets called with some argument
> set to some specific value). Do you want to report this as a bug
> (assuming there isn't one already; I for one can't find any) from
> the Debian side, or should I open one?

I'll bounce the Debian Xen bug back to the Debian gcc package with your
comments and I suppose the maintainer will then take it upstream.

Ian.

Re: xen/x86: unable to build with gcc5

[Ian Campbell]

On Fri, 2015-03-13 at 13:32 +0000, M A Young wrote:
> On Fri, 13 Mar 2015, Ian Campbell wrote:
> 
> > On Fri, 2015-03-13 at 12:10 +0000, Jan Beulich wrote:
> > [...]
> > > For all of the above, at a first glance I'd say compiler bug. But to
> > > be sure, could you send the three .o-s?
> 
> See https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64963

Indeed, it turns out I was using an old gcc-5 snapshot (5-20150205-1).
I've just updated to 5-20150307-1 and the issue is already fixed as
indicated in that bug report. The symbols.c patch is still needed.

Sorry for the noise. 

Ian.

Re: xen/x86: unable to build with gcc5

[Ian Campbell]

On Fri, 2015-03-13 at 12:58 +0000, Jan Beulich wrote:
> >>> On 13.03.15 at 13:25, <ian.campbell@citrix.com> wrote:
> > From: Ian Campbell <ian.campbell@citrix.com>
> > Date: Fri, 13 Mar 2015 12:22:25 +0000
> > Subject: [PATCH] xen: common: Use unbounded array for symbols_offset.
> > 
> > Using a singleton array causes gcc5 to report:
> > symbols.c: In function 'symbols_lookup':
> > symbols.c:128:359: error: array subscript is above array bounds 
> > [-Werror=array-bounds]
> > symbols.c:136:176: error: array subscript is above array bounds 
> > [-Werror=array-bounds]
> > 
> > Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
> 
> Acked-by: Jan Beulich <jbeulich@suse.com>

Applied, thanks.