Minimum toolchain requirements?

Minimum toolchain requirements?

[Rob Landley]

Although the README and Documentation/Changes both say the kernel  
builds with gcc 3.2, this is no loner the case. In reality the new 3.7  
kernel no longer builds under unpatched gcc 4.2.1 (the last GPLv2  
release).

Building for i686 breaks with "arch/x86/kernel/cpu/perf_event_p6.c:22:  
error: p6_hw_cache_event_ids causes a section type conflict" (trivial  
workaround: patch kernel so CONFIG_BROKEN_RODATA defaults to y).  
Building for mips breaks with "arch/mips/lib/delay.c:24:5: warning:  
"__SIZEOF_LONG__" is not defined". (Introduced January 2007, gcc git  
commit 6a60f216c210e. Easy enough to add an equivalent to my toolchain.)

Over in my Aboriginal Linux project I've been patching both the kernel  
and my GPLv2 toolchain for a while to work around these random  
breakages (such as when sh4 decided it would only build with binutils  
2.20, which had only been out for 3 months at the time). But since I'm  
supposed to be catching Documentation stuff that falls through the  
cracks, if the docs are clearly out of date I should probably update  
them.

I'm still regression testing each new kernel release against gcc 4.2.1,  
binutils 2.17, and make 3.81 (I.E. the last GPLv2 release of each  
tool). My personal preference would be to upgrade Documentation/Changes  
to say "gcc 4.2.1 and binutils 2.17 are the oldest supported versions",  
and then try to push patches upstream that prevent the kernel from  
building on those. Unfortunately, when I try to push patches to make  
older toolchains work, the reception isn't exactly warm.

So I ask the question: what are the current minimum requirements to  
build the kernel?

Rob

Re: Minimum toolchain requirements?

[Shaun Ruffell]

Hi Rob,

On Fri, Dec 14, 2012 at 04:25:10PM -0600, Rob Landley wrote:
> Although the README and Documentation/Changes both say the kernel
> builds with gcc 3.2, this is no loner the case. In reality the new
> 3.7 kernel no longer builds under unpatched gcc 4.2.1 (the last
> GPLv2 release).
> 
> Building for i686 breaks with
> "arch/x86/kernel/cpu/perf_event_p6.c:22: error:
> p6_hw_cache_event_ids causes a section type conflict" (trivial
> workaround: patch kernel so CONFIG_BROKEN_RODATA defaults to y).

I came across your email while searching for a solution to the above
build error.

In addition to setting CONFIG_BROKEN_RODATA=y, Jan Beulich sent a
patch [1] to LKML that also fixes this build problem for me.

[1] https://lkml.org/lkml/2012/11/23/308

It doesn't appear to have hit Linus' tree yet though, but with luck
someone will pick it up before 3.8 final.

Cheers,
Shaun

Re: Minimum toolchain requirements?

[Rob Landley]

On 12/27/2012 02:09:33 PM, Shaun Ruffell wrote:
> Hi Rob,
> 
> On Fri, Dec 14, 2012 at 04:25:10PM -0600, Rob Landley wrote:
> > Although the README and Documentation/Changes both say the kernel
> > builds with gcc 3.2, this is no loner the case. In reality the new
> > 3.7 kernel no longer builds under unpatched gcc 4.2.1 (the last
> > GPLv2 release).
> >
> > Building for i686 breaks with
> > "arch/x86/kernel/cpu/perf_event_p6.c:22: error:
> > p6_hw_cache_event_ids causes a section type conflict" (trivial
> > workaround: patch kernel so CONFIG_BROKEN_RODATA defaults to y).
> 
> I came across your email while searching for a solution to the above
> build error.
> 
> In addition to setting CONFIG_BROKEN_RODATA=y, Jan Beulich sent a
> patch [1] to LKML that also fixes this build problem for me.
> 
> [1] https://lkml.org/lkml/2012/11/23/308
> 
> It doesn't appear to have hit Linus' tree yet though, but with luck
> someone will pick it up before 3.8 final.
> 
> Cheers,
> Shaun

Having it in 3.7.x would be nice too. Copying stable.

Thanks,

Rob

Re: Minimum toolchain requirements?

[Greg KH]

On Thu, Dec 27, 2012 at 09:35:27PM -0600, Rob Landley wrote:
> On 12/27/2012 02:09:33 PM, Shaun Ruffell wrote:
> >Hi Rob,
> >
> >On Fri, Dec 14, 2012 at 04:25:10PM -0600, Rob Landley wrote:
> >> Although the README and Documentation/Changes both say the kernel
> >> builds with gcc 3.2, this is no loner the case. In reality the new
> >> 3.7 kernel no longer builds under unpatched gcc 4.2.1 (the last
> >> GPLv2 release).
> >>
> >> Building for i686 breaks with
> >> "arch/x86/kernel/cpu/perf_event_p6.c:22: error:
> >> p6_hw_cache_event_ids causes a section type conflict" (trivial
> >> workaround: patch kernel so CONFIG_BROKEN_RODATA defaults to y).
> >
> >I came across your email while searching for a solution to the above
> >build error.
> >
> >In addition to setting CONFIG_BROKEN_RODATA=y, Jan Beulich sent a
> >patch [1] to LKML that also fixes this build problem for me.
> >
> >[1] https://lkml.org/lkml/2012/11/23/308
> >
> >It doesn't appear to have hit Linus' tree yet though, but with luck
> >someone will pick it up before 3.8 final.
> >
> >Cheers,
> >Shaun
> 
> Having it in 3.7.x would be nice too. Copying stable.

Please let stable@vger.kernel.org know the git commit id in Linus's tree
when it makes it there, and we can take it.  Until then, nothing we can
do, sorry.

greg k-h