[PATCH] x86/perf_events: Fix "section type conflict" build error.

[PATCH] x86/perf_events: Fix "section type conflict" build error.

[Shaun Ruffell]

From: Jan Beulich <JBeulich@suse.com>

This patch fixes a build regression first introduced in 3.7 with
(e09df47 "perf/x86: Update/fix generic events on P6 PMU").

At least some older versions of gcc, like (GCC) 4.1.2 20080704 (Red Hat
4.1.2-51), dislike mixing constant and non-const data in the same
section. Without this patch a build will fail with the following error:

    CC      arch/x86/kernel/cpu/perf_event_p6.o
  arch/x86/kernel/cpu/perf_event_p6.c:22: error: p6_hw_cache_event_ids causes a section type conflict
  make[3]: *** [arch/x86/kernel/cpu/perf_event_p6.o] Error 1

Newer versions of gcc simply emits the section as writable (which isn't what
we want, but also is not a big problem as it gets discarded post-init anyway).

Also get the Knight's Corner definitions in sync.

Signed-off-by: Jan Beulich <jbeulich@suse.com>
Cc: stable@vger.kernel.org # 3.7.x only
[sruffell@digium.com: Added details to the commit message.]
Signed-off-by: Shaun Ruffell <sruffell@digium.com>

---

[I had the wrong stable email address when I previously sent this.
Sorry about the extra noise.]

Hans, Thomas,

Any chance of picking this up for 3.8? This doesn't seem too
controversial since it only makes sure that data in the __initconst
section is const.

Rob Landley also reported this to the list here:
https://lkml.org/lkml/2012/12/14/511. 

Thanks,
Shaun

 arch/x86/kernel/cpu/perf_event_knc.c | 4 ++--
 arch/x86/kernel/cpu/perf_event_p6.c  | 2 +-
 2 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/arch/x86/kernel/cpu/perf_event_knc.c b/arch/x86/kernel/cpu/perf_event_knc.c
index 4b7731b..838fa87 100644
--- a/arch/x86/kernel/cpu/perf_event_knc.c
+++ b/arch/x86/kernel/cpu/perf_event_knc.c
@@ -17,7 +17,7 @@ static const u64 knc_perfmon_event_map[] =
   [PERF_COUNT_HW_BRANCH_MISSES]		= 0x002b,
 };
 
-static __initconst u64 knc_hw_cache_event_ids
+static const u64 __initconst knc_hw_cache_event_ids
 				[PERF_COUNT_HW_CACHE_MAX]
 				[PERF_COUNT_HW_CACHE_OP_MAX]
 				[PERF_COUNT_HW_CACHE_RESULT_MAX] =
@@ -284,7 +284,7 @@ static struct attribute *intel_knc_formats_attr[] = {
 	NULL,
 };
 
-static __initconst struct x86_pmu knc_pmu = {
+static const struct x86_pmu knc_pmu __initconst = {
 	.name			= "knc",
 	.handle_irq		= knc_pmu_handle_irq,
 	.disable_all		= knc_pmu_disable_all,
diff --git a/arch/x86/kernel/cpu/perf_event_p6.c b/arch/x86/kernel/cpu/perf_event_p6.c
index f2af39f..b1e2fe1 100644
--- a/arch/x86/kernel/cpu/perf_event_p6.c
+++ b/arch/x86/kernel/cpu/perf_event_p6.c
@@ -19,7 +19,7 @@ static const u64 p6_perfmon_event_map[] =
 
 };
 
-static __initconst u64 p6_hw_cache_event_ids
+static const u64 __initconst p6_hw_cache_event_ids
 				[PERF_COUNT_HW_CACHE_MAX]
 				[PERF_COUNT_HW_CACHE_OP_MAX]
 				[PERF_COUNT_HW_CACHE_RESULT_MAX] =
-- 
1.8.1

Re: [PATCH] x86/perf_events: Fix "section type conflict" build error.

[Shaun Ruffell]

Bump. Still an issue with v3.8-rc4. Am I mistaken in thinking this
patch is trivial and obviously correct?

Thanks,
Shaun

On Wed, Jan 09, 2013 at 03:59:42PM -0600, Shaun Ruffell wrote:
> From: Jan Beulich <JBeulich@suse.com>
> 
> This patch fixes a build regression first introduced in 3.7 with
> (e09df47 "perf/x86: Update/fix generic events on P6 PMU").
> 
> At least some older versions of gcc, like (GCC) 4.1.2 20080704 (Red Hat
> 4.1.2-51), dislike mixing constant and non-const data in the same
> section. Without this patch a build will fail with the following error:
> 
>     CC      arch/x86/kernel/cpu/perf_event_p6.o
>   arch/x86/kernel/cpu/perf_event_p6.c:22: error: p6_hw_cache_event_ids causes a section type conflict
>   make[3]: *** [arch/x86/kernel/cpu/perf_event_p6.o] Error 1
> 
> Newer versions of gcc simply emits the section as writable (which isn't what
> we want, but also is not a big problem as it gets discarded post-init anyway).
> 
> Also get the Knight's Corner definitions in sync.
> 
> Signed-off-by: Jan Beulich <jbeulich@suse.com>
> Cc: stable@vger.kernel.org # 3.7.x only
> [sruffell@digium.com: Added details to the commit message.]
> Signed-off-by: Shaun Ruffell <sruffell@digium.com>
> 
> ---
> 
> [I had the wrong stable email address when I previously sent this.
> Sorry about the extra noise.]
> 
> Hans, Thomas,
> 
> Any chance of picking this up for 3.8? This doesn't seem too
> controversial since it only makes sure that data in the __initconst
> section is const.
> 
> Rob Landley also reported this to the list here:
> https://lkml.org/lkml/2012/12/14/511. 
> 
> Thanks,
> Shaun
> 
>  arch/x86/kernel/cpu/perf_event_knc.c | 4 ++--
>  arch/x86/kernel/cpu/perf_event_p6.c  | 2 +-
>  2 files changed, 3 insertions(+), 3 deletions(-)
> 
> diff --git a/arch/x86/kernel/cpu/perf_event_knc.c b/arch/x86/kernel/cpu/perf_event_knc.c
> index 4b7731b..838fa87 100644
> --- a/arch/x86/kernel/cpu/perf_event_knc.c
> +++ b/arch/x86/kernel/cpu/perf_event_knc.c
> @@ -17,7 +17,7 @@ static const u64 knc_perfmon_event_map[] =
>    [PERF_COUNT_HW_BRANCH_MISSES]		= 0x002b,
>  };
>  
> -static __initconst u64 knc_hw_cache_event_ids
> +static const u64 __initconst knc_hw_cache_event_ids
>  				[PERF_COUNT_HW_CACHE_MAX]
>  				[PERF_COUNT_HW_CACHE_OP_MAX]
>  				[PERF_COUNT_HW_CACHE_RESULT_MAX] =
> @@ -284,7 +284,7 @@ static struct attribute *intel_knc_formats_attr[] = {
>  	NULL,
>  };
>  
> -static __initconst struct x86_pmu knc_pmu = {
> +static const struct x86_pmu knc_pmu __initconst = {
>  	.name			= "knc",
>  	.handle_irq		= knc_pmu_handle_irq,
>  	.disable_all		= knc_pmu_disable_all,
> diff --git a/arch/x86/kernel/cpu/perf_event_p6.c b/arch/x86/kernel/cpu/perf_event_p6.c
> index f2af39f..b1e2fe1 100644
> --- a/arch/x86/kernel/cpu/perf_event_p6.c
> +++ b/arch/x86/kernel/cpu/perf_event_p6.c
> @@ -19,7 +19,7 @@ static const u64 p6_perfmon_event_map[] =
>  
>  };
>  
> -static __initconst u64 p6_hw_cache_event_ids
> +static const u64 __initconst p6_hw_cache_event_ids
>  				[PERF_COUNT_HW_CACHE_MAX]
>  				[PERF_COUNT_HW_CACHE_OP_MAX]
>  				[PERF_COUNT_HW_CACHE_RESULT_MAX] =
> -- 
> 1.8.1