[Buildroot] [PATCH] package/binutils: Disable -Os for Thumb-2 only platforms

[Buildroot] [PATCH] package/binutils: Disable -Os for Thumb-2 only platforms

[Romain Naour]

ERROR: reloc type R_ARM_THM_MOVW_ABS_NC unsupported in this context
ERROR: reloc type R_ARM_THM_MOVT_ABS unsupported in this context
2 bad relocscollect2: error: ld returned 1 exit status

Binutils build fine with -O1, -O2 or -O3.

Fixes:
http://autobuild.buildroot.net/results/e72/e727e6e9cf361acc786acec192a88f5f5444a2f6

Signed-off-by: Romain Naour <romain.naour@gmail.com>
Cc: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
---
 package/binutils/binutils.mk | 6 ++++++
 1 file changed, 6 insertions(+)

diff --git a/package/binutils/binutils.mk b/package/binutils/binutils.mk
index ff90a16..6946e7e 100644
--- a/package/binutils/binutils.mk
+++ b/package/binutils/binutils.mk
@@ -77,6 +77,12 @@ ifeq ($(BR2_bfin),y)
 BINUTILS_CONF_ENV += CFLAGS="$(TARGET_CFLAGS) -O1"
 endif
 
+# Workaround a build issue with -Os on Thumb-2 only platforms
+# (such as Cortex-M). (Binutils 2.25.1 and 2.26.1)
+ifeq ($(BR2_ARM_CPU_HAS_ARM)$(BR2_OPTIMIZE_S),yy)
+BINUTILS_CONF_ENV += CFLAGS="$(TARGET_CFLAGS) -O1"
+endif
+
 # Install binutils after busybox to prefer full-blown utilities
 ifeq ($(BR2_PACKAGE_BUSYBOX),y)
 BINUTILS_DEPENDENCIES += busybox
-- 
2.5.5

[Buildroot] [PATCH] package/binutils: Disable -Os for Thumb-2 only platforms

[Baruch Siach]

Hi Romain,

On Sun, Sep 04, 2016 at 10:49:50AM +0200, Romain Naour wrote:
> ERROR: reloc type R_ARM_THM_MOVW_ABS_NC unsupported in this context
> ERROR: reloc type R_ARM_THM_MOVT_ABS unsupported in this context
> 2 bad relocscollect2: error: ld returned 1 exit status
> 
> Binutils build fine with -O1, -O2 or -O3.

Is there an upstream bug report we can refer to?

> Fixes:
> http://autobuild.buildroot.net/results/e72/e727e6e9cf361acc786acec192a88f5f5444a2f6
> 
> Signed-off-by: Romain Naour <romain.naour@gmail.com>
> Cc: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
> ---
>  package/binutils/binutils.mk | 6 ++++++
>  1 file changed, 6 insertions(+)
> 
> diff --git a/package/binutils/binutils.mk b/package/binutils/binutils.mk
> index ff90a16..6946e7e 100644
> --- a/package/binutils/binutils.mk
> +++ b/package/binutils/binutils.mk
> @@ -77,6 +77,12 @@ ifeq ($(BR2_bfin),y)
>  BINUTILS_CONF_ENV += CFLAGS="$(TARGET_CFLAGS) -O1"
>  endif
>  
> +# Workaround a build issue with -Os on Thumb-2 only platforms
> +# (such as Cortex-M). (Binutils 2.25.1 and 2.26.1)
> +ifeq ($(BR2_ARM_CPU_HAS_ARM)$(BR2_OPTIMIZE_S),yy)
> +BINUTILS_CONF_ENV += CFLAGS="$(TARGET_CFLAGS) -O1"

Isn't -O2 the more "standard" optimization option?

> +endif

baruch

-- 
     http://baruch.siach.name/blog/                  ~. .~   Tk Open Systems
=}------------------------------------------------ooO--U--Ooo------------{=
   - baruch at tkos.co.il - tel: +972.52.368.4656, http://www.tkos.co.il -

[Buildroot] [PATCH] package/binutils: Disable -Os for Thumb-2 only platforms

[Romain Naour]

Hi Baruch,

Le 04/09/2016 ? 10:55, Baruch Siach a ?crit :
> Hi Romain,
> 
> On Sun, Sep 04, 2016 at 10:49:50AM +0200, Romain Naour wrote:
>> ERROR: reloc type R_ARM_THM_MOVW_ABS_NC unsupported in this context
>> ERROR: reloc type R_ARM_THM_MOVT_ABS unsupported in this context
>> 2 bad relocscollect2: error: ld returned 1 exit status
>>
>> Binutils build fine with -O1, -O2 or -O3.
> 
> Is there an upstream bug report we can refer to?

No, I can't find any bug report.
Also I don't understand the meaning of this error, I just try some other build
options and figured out that the issue was caused by -Os.

> 
>> Fixes:
>> http://autobuild.buildroot.net/results/e72/e727e6e9cf361acc786acec192a88f5f5444a2f6
>>
>> Signed-off-by: Romain Naour <romain.naour@gmail.com>
>> Cc: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
>> ---
>>  package/binutils/binutils.mk | 6 ++++++
>>  1 file changed, 6 insertions(+)
>>
>> diff --git a/package/binutils/binutils.mk b/package/binutils/binutils.mk
>> index ff90a16..6946e7e 100644
>> --- a/package/binutils/binutils.mk
>> +++ b/package/binutils/binutils.mk
>> @@ -77,6 +77,12 @@ ifeq ($(BR2_bfin),y)
>>  BINUTILS_CONF_ENV += CFLAGS="$(TARGET_CFLAGS) -O1"
>>  endif
>>  
>> +# Workaround a build issue with -Os on Thumb-2 only platforms
>> +# (such as Cortex-M). (Binutils 2.25.1 and 2.26.1)
>> +ifeq ($(BR2_ARM_CPU_HAS_ARM)$(BR2_OPTIMIZE_S),yy)
>> +BINUTILS_CONF_ENV += CFLAGS="$(TARGET_CFLAGS) -O1"
> 
> Isn't -O2 the more "standard" optimization option?

Yes maybe, it's just because I borrowed the same line above for Blackfin :)
As I said on the commit log, Binutils build fine with O1,O2 and O3 so use O2 if
it's more "standard" optimization option.

Thanks!

Best regards,
Romain

> 
>> +endif
> 
> baruch
> 

[Buildroot] [PATCH] package/binutils: Disable -Os for Thumb-2 only platforms

[Thomas Petazzoni]

Hello,

On Sun,  4 Sep 2016 10:49:50 +0200, Romain Naour wrote:

> +# Workaround a build issue with -Os on Thumb-2 only platforms
> +# (such as Cortex-M). (Binutils 2.25.1 and 2.26.1)
> +ifeq ($(BR2_ARM_CPU_HAS_ARM)$(BR2_OPTIMIZE_S),yy)

Unless I'm missing something, I don't see how the code is matching the
comment above. Your comment is "with Thumb-2 only platforms, there is
an issue with the -Os optimization level" while your condition is "with
ARM-capable platforms, there is an issue with the -Os optimization
level".

Are you sure what want is not:

ifeq ($(BR2_ARM_CPU_HAS_ARM):$(BR2_OPTIMIZE_S),:y)
....
endif

Also, I'm a bit surprised this happens only on Thumb-2 only platforms,
and not on all ARM platforms, when Thumb-2 is chosen. But I'm far from
understanding all the fine details of relocations.

Thanks,

Thomas
-- 
Thomas Petazzoni, CTO, Free Electrons
Embedded Linux and Kernel engineering
http://free-electrons.com

[Buildroot] [PATCH] package/binutils: Disable -Os for Thumb-2 only platforms

[Romain Naour]

Hi Thomas,

Le 04/09/2016 ? 14:43, Thomas Petazzoni a ?crit :
> Hello,
> 
> On Sun,  4 Sep 2016 10:49:50 +0200, Romain Naour wrote:
> 
>> +# Workaround a build issue with -Os on Thumb-2 only platforms
>> +# (such as Cortex-M). (Binutils 2.25.1 and 2.26.1)
>> +ifeq ($(BR2_ARM_CPU_HAS_ARM)$(BR2_OPTIMIZE_S),yy)
> 
> Unless I'm missing something, I don't see how the code is matching the
> comment above. Your comment is "with Thumb-2 only platforms, there is
> an issue with the -Os optimization level" while your condition is "with
> ARM-capable platforms, there is an issue with the -Os optimization
> level".
> 
> Are you sure what want is not:
> 
> ifeq ($(BR2_ARM_CPU_HAS_ARM):$(BR2_OPTIMIZE_S),:y)
> ....
> endif

You're right, this is what I wanted to do. :-/

> 
> Also, I'm a bit surprised this happens only on Thumb-2 only platforms,
> and not on all ARM platforms, when Thumb-2 is chosen. But I'm far from
> understanding all the fine details of relocations.

Neither do I.
I tried with an ARM toolchain with BR2_ARM_INSTRUCTIONS_THUMB2 set and with the
same build options (static only and Os) but Binutils build fine.

I'll send a v2 shortly.

Best regards,
Romain

> 
> Thanks,
> 
> Thomas
> 

[Buildroot] [PATCH] package/binutils: Disable -Os for Thumb-2 only platforms

[Nicholas Walton]

On Sun, Sep 4, 2016 at 5:43 AM Thomas Petazzoni <
thomas.petazzoni@free-electrons.com> wrote:

> On Sun,  4 Sep 2016 10:49:50 +0200, Romain Naour wrote:
>
> > +# Workaround a build issue with -Os on Thumb-2 only platforms
> > +# (such as Cortex-M). (Binutils 2.25.1 and 2.26.1)
>
> I'm a bit surprised this happens only on Thumb-2 only platforms,
> and not on all ARM platforms, when Thumb-2 is chosen. But I'm far from
> understanding all the fine details of relocations.
>

I found a bug report about a similar error in a different package, and I
saw it on a non-Thumb-2 platform (ARM9). I'll send more details when I'm in
the office on Tuesday, as I don't have the error message, package name, or
bug report in front of me.

Regards,
Nicholas
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.busybox.net/pipermail/buildroot/attachments/20160904/b144a93e/attachment.html>

[Buildroot] [PATCH] package/binutils: Disable -Os for Thumb-2 only platforms

[Nicholas Walton]

Please disregard my previous message:

On Sun, Sep 4, 2016 at 10:52 AM Nicholas Walton <nicholas.walton@gmail.com>
wrote:

> I found a bug report about a similar error in a different package, and I
> saw it on a non-Thumb-2 platform (ARM9).
>

The error I was remembering was bad code generation causing a segfault
(which I saw in libfb.so). Restricting xserver_xorg-server to -O2
(5500211e201f84b6e2ce0dd5ff3013fb92628860)
fixed it. I suspect the root cause was related to *Bug 63346*
<https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63346> -
"xserver_xorg-server-1.15.1
crash on RaspberryPi . . . gcc-4.9", and/or* Bug 69614*
<https://gcc.gnu.org/bugzilla/show_bug.cgi?id=69614> - "[4.9/5 Regression]
wrong code with -Os -fno-expensive . . . @ armv7a". Bug 69614 is fixed in
6.0 and the patch was backported to 5, but not to 4.9.

Apologies,
Nicholas Walton
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.busybox.net/pipermail/buildroot/attachments/20160906/cbfe59a1/attachment.html>