Skip to contentSkip to conversation

linux-next: build failure after merge of the tip tree

PATCH In review Stephen Rothwell 2 participants 12 messages Mar 12, 2010 – Mar 17, 2010

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

Hi all,

After merging the tip tree, today's linux-next build (powerpc ppc64_defconfig) failed like this:

ERROR: ".perf_arch_fetch_caller_regs" [fs/xfs/xfs.ko] undefined!
ERROR: ".perf_arch_fetch_caller_regs" [arch/powerpc/platforms/cell/spufs/spufs.ko] undefined!

Caused by commit 5331d7b84613b8325362dde53dc2bff2fb87d351 ("perf:
Introduce new perf_fetch_caller_regs() for hot regs snapshot") from the tip tree. Presumably commit 639fe4b12f92b54c9c3b38c82cdafaa38cfd3e63 ("perf: export perf_trace_regs and perf_arch_fetch_caller_regs") should have exported the weak version as well as the x86 one.

I have used the version of the tip tree from next-20100311 for today.

Signature · 3 lines
Cheers,
Stephen Rothwell                    [email protected]
http://www.canb.auug.org.au/~sfr/
#1 lore raw

PATCH 1/1 perf: Also export the weak version of perf_arch_fetch_caller_regs

Quoted 8 lines · Stephen Rothwell

Hi all,

After merging the tip tree, today's linux-next build (powerpc ppc64_defconfig) failed like this:

ERROR: ".perf_arch_fetch_caller_regs" [fs/xfs/xfs.ko] undefined!
ERROR: ".perf_arch_fetch_caller_regs" [arch/powerpc/platforms/cell/spufs/spufs.ko] undefined!

Quoted 7 lines

Caused by commit 5331d7b84613b8325362dde53dc2bff2fb87d351 ("perf:
Introduce new perf_fetch_caller_regs() for hot regs snapshot") from the tip tree. Presumably commit 639fe4b12f92b54c9c3b38c82cdafaa38cfd3e63 ("perf: export perf_trace_regs and perf_arch_fetch_caller_regs") should have exported the weak version as well as the x86 one.

I have used the version of the tip tree from next-20100311 for today.

Does that fixes the issue? Sorry I can not test for now...

Thanks!

---

Quoted 1 line

From 0b21e01edb23b4725d20a9e5fddb6df7fedd16e1 Mon Sep 17 00:00:00 2001

From: Frederic Weisbecker <[email protected]>
Date: Fri, 12 Mar 2010 06:23:22 +0100
Subject: [PATCH] perf: Also export the weak version of perf_arch_fetch_caller_regs

perf_arch_fetch_caller_regs() is exported in x86 but not for other archs that use the weak version.

Export it only once from the generic code, it will apply to either weak/overriden version.

---
 arch/x86/kernel/cpu/perf_event.c |    1 -
 kernel/perf_event.c              |    1 +
 2 files changed, 1 insertions(+), 1 deletions(-)
arch/x86/kernel/cpu/perf_event.c +0 -1
index a6d92c3..a3aff76 100644
@@ -1691,4 +1691,3 @@ void perf_arch_fetch_caller_regs(struct pt_regs *regs, unsigned long ip, int ski
 	regs->cs = __KERNEL_CS;
 	local_save_flags(regs->flags);
 }
-EXPORT_SYMBOL_GPL(perf_arch_fetch_caller_regs);
kernel/perf_event.c +1 -0
index 8bf6127..7aea58d 100644
@@ -2794,6 +2794,7 @@ __weak
 void perf_arch_fetch_caller_regs(struct pt_regs *regs, unsigned long ip, int skip)
 {
 }
+EXPORT_SYMBOL_GPL(perf_arch_fetch_caller_regs);
 
 /*
  * Output
Signature · 1 line
1.6.2.3
#2 lore raw

Re: [PATCH] perf: Also export the weak version of perf_arch_fetch_caller_regs

Hi Frederic,

Quoted 2 lines · On Fri, 12 Mar 2010 06:50:44 +0100 Frederic Weisbecker <[email protected]>…

Does that fixes the issue? Sorry I can not test for now...

I can't test it right now, sorry, but I have heard that EXPORT_SYMBOL's for weak symbols must be in a different file from the actual symbol definition.

Signature · 3 lines
Cheers,
Stephen Rothwell                    [email protected]
http://www.canb.auug.org.au/~sfr/
#3 lore raw

Re: [PATCH] perf: Also export the weak version of perf_arch_fetch_caller_regs

Hi Frederic,

Quoted 8 lines · On Fri, 12 Mar 2010 17:20:43 +1100 Stephen Rothwell <[email protected]>…
Quoted 2 lines · On Fri, 12 Mar 2010 06:50:44 +0100 Frederic Weisbecker <[email protected]>…

Does that fixes the issue? Sorry I can not test for now...

I can't test it right now, sorry, but I have heard that EXPORT_SYMBOL's for weak symbols must be in a different file from the actual symbol definition.

See http://lkml.org/lkml/2010/2/17/467

Signature · 3 lines
Cheers,
Stephen Rothwell                    [email protected]
http://www.canb.auug.org.au/~sfr/
#4 lore raw

PATCH 1/1 perf: Fix unexported generic perf_arch_fetch_caller_regs

perf_arch_fetch_caller_regs() is exported for the overriden x86 version, but not for the generic weak version.
As a general rule, weak functions should have their symbol exported in the same file they are defined.

So let's export it on trace_event_perf.c as it is used by trace events only.

This fixes:
ERROR: ".perf_arch_fetch_caller_regs" [fs/xfs/xfs.ko] undefined!
ERROR: ".perf_arch_fetch_caller_regs" [arch/powerpc/platforms/cell/spufs/spufs.ko] undefined!

And also only build it if trace events are enabled.

---
 arch/x86/kernel/cpu/perf_event.c |    1 -
 kernel/trace/trace_event_perf.c  |    2 ++
 2 files changed, 2 insertions(+), 1 deletions(-)
arch/x86/kernel/cpu/perf_event.c +0 -1
index 7645fae..5fb490c 100644
@@ -1713,4 +1713,3 @@ void perf_arch_fetch_caller_regs(struct pt_regs *regs, unsigned long ip, int ski
 	regs->cs = __KERNEL_CS;
 	local_save_flags(regs->flags);
 }
-EXPORT_SYMBOL_GPL(perf_arch_fetch_caller_regs);
kernel/trace/trace_event_perf.c +2 -0
index 0709e4f..7d79a10 100644
@@ -12,6 +12,8 @@
 DEFINE_PER_CPU(struct pt_regs, perf_trace_regs);
 EXPORT_PER_CPU_SYMBOL_GPL(perf_trace_regs);
 
+EXPORT_SYMBOL_GPL(perf_arch_fetch_caller_regs);
+
 static char *perf_trace_buf;
 static char *perf_trace_buf_nmi;
 
Signature · 1 line
1.6.2.3
#5 lore raw

PATCH v2 1/1 perf: Fix unexported generic perf_arch_fetch_caller_regs

perf_arch_fetch_caller_regs() is exported for the overriden x86 version, but not for the generic weak version.
As a general rule, weak functions should have their symbol exported in the same file they are defined.

So let's export it on trace_event_perf.c as it is used by trace events only.

This fixes:
ERROR: ".perf_arch_fetch_caller_regs" [fs/xfs/xfs.ko] undefined!
ERROR: ".perf_arch_fetch_caller_regs" [arch/powerpc/platforms/cell/spufs/spufs.ko] undefined!

v2: And also only build it if trace events are enabled.

---
 arch/x86/kernel/cpu/perf_event.c |    3 ++-
 kernel/perf_event.c              |    2 ++
 kernel/trace/trace_event_perf.c  |    2 ++
 3 files changed, 6 insertions(+), 1 deletions(-)
arch/x86/kernel/cpu/perf_event.c +2 -1
index 7645fae..60398a0 100644
@@ -1702,6 +1702,7 @@ struct perf_callchain_entry *perf_callchain(struct pt_regs *regs)
 	return entry;
 }
 
+#ifdef CONFIG_EVENT_TRACING
 void perf_arch_fetch_caller_regs(struct pt_regs *regs, unsigned long ip, int skip)
 {
 	regs->ip = ip;
@@ -1713,4 +1714,4 @@ void perf_arch_fetch_caller_regs(struct pt_regs *regs, unsigned long ip, int ski
 	regs->cs = __KERNEL_CS;
 	local_save_flags(regs->flags);
 }
-EXPORT_SYMBOL_GPL(perf_arch_fetch_caller_regs);
+#endif
kernel/perf_event.c +2 -0
index fb3031c..574ee58 100644
@@ -2786,10 +2786,12 @@ __weak struct perf_callchain_entry *perf_callchain(struct pt_regs *regs)
 	return NULL;
 }
 
+#ifdef CONFIG_EVENT_TRACING
 __weak
 void perf_arch_fetch_caller_regs(struct pt_regs *regs, unsigned long ip, int skip)
 {
 }
+#endif
 
 /*
  * Output
kernel/trace/trace_event_perf.c +2 -0
index 0709e4f..7d79a10 100644
@@ -12,6 +12,8 @@
 DEFINE_PER_CPU(struct pt_regs, perf_trace_regs);
 EXPORT_PER_CPU_SYMBOL_GPL(perf_trace_regs);
 
+EXPORT_SYMBOL_GPL(perf_arch_fetch_caller_regs);
+
 static char *perf_trace_buf;
 static char *perf_trace_buf_nmi;
 
Signature · 1 line
1.6.2.3
#6 lore raw

Re: [PATCH v2] perf: Fix unexported generic perf_arch_fetch_caller_regs

Hi Frederic,

Quoted 4 lines · On Mon, 15 Mar 2010 23:17:22 +0100 Frederic Weisbecker <[email protected]>…

perf_arch_fetch_caller_regs() is exported for the overriden x86 version, but not for the generic weak version.
As a general rule, weak functions should have their symbol exported

                                          ^

"not"

Quoted 4 lines

in the same file they are defined.

So let's export it on trace_event_perf.c as it is used by trace events only.

Otherwise, looks good (though I haven't tested it).

Signature · 3 lines
Cheers,
Stephen Rothwell                    [email protected]
http://www.canb.auug.org.au/~sfr/
#7 lore raw

Re: [PATCH v2] perf: Fix unexported generic perf_arch_fetch_caller_regs

Quoted 9 lines · Stephen Rothwell

Hi Frederic,

Quoted 4 lines · On Mon, 15 Mar 2010 23:17:22 +0100 Frederic Weisbecker <[email protected]>…

perf_arch_fetch_caller_regs() is exported for the overriden x86 version, but not for the generic weak version.
As a general rule, weak functions should have their symbol exported

                                          ^

"not"

Ooh, right.

Quoted 6 lines
Quoted 4 lines

in the same file they are defined.

So let's export it on trace_event_perf.c as it is used by trace events only.

Otherwise, looks good (though I haven't tested it).

I've tested with some trace event built as module and it looks fine so far.

Thanks.

#8 lore raw

PATCH v3 1/1 perf: Fix unexported generic perf_arch_fetch_caller_regs

In reply to Stephen Rothwell

perf_arch_fetch_caller_regs() is exported for the overriden x86 version, but not for the generic weak version.
As a general rule, weak functions should not have their symbol exported in the same file they are defined.

So let's export it on trace_event_perf.c as it is used by trace events only.

This fixes:
ERROR: ".perf_arch_fetch_caller_regs" [fs/xfs/xfs.ko] undefined!
ERROR: ".perf_arch_fetch_caller_regs" [arch/powerpc/platforms/cell/spufs/spufs.ko] undefined!

v2: And also only build it if trace events are enabled.
v3: Fix changelog mistake

---
 arch/x86/kernel/cpu/perf_event.c |    3 ++-
 kernel/perf_event.c              |    2 ++
 kernel/trace/trace_event_perf.c  |    2 ++
 3 files changed, 6 insertions(+), 1 deletions(-)
arch/x86/kernel/cpu/perf_event.c +2 -1
index 7645fae..60398a0 100644
@@ -1702,6 +1702,7 @@ struct perf_callchain_entry *perf_callchain(struct pt_regs *regs)
 	return entry;
 }
 
+#ifdef CONFIG_EVENT_TRACING
 void perf_arch_fetch_caller_regs(struct pt_regs *regs, unsigned long ip, int skip)
 {
 	regs->ip = ip;
@@ -1713,4 +1714,4 @@ void perf_arch_fetch_caller_regs(struct pt_regs *regs, unsigned long ip, int ski
 	regs->cs = __KERNEL_CS;
 	local_save_flags(regs->flags);
 }
-EXPORT_SYMBOL_GPL(perf_arch_fetch_caller_regs);
+#endif
kernel/perf_event.c +2 -0
index fb3031c..574ee58 100644
@@ -2786,10 +2786,12 @@ __weak struct perf_callchain_entry *perf_callchain(struct pt_regs *regs)
 	return NULL;
 }
 
+#ifdef CONFIG_EVENT_TRACING
 __weak
 void perf_arch_fetch_caller_regs(struct pt_regs *regs, unsigned long ip, int skip)
 {
 }
+#endif
 
 /*
  * Output
kernel/trace/trace_event_perf.c +2 -0
index 0709e4f..7d79a10 100644
@@ -12,6 +12,8 @@
 DEFINE_PER_CPU(struct pt_regs, perf_trace_regs);
 EXPORT_PER_CPU_SYMBOL_GPL(perf_trace_regs);
 
+EXPORT_SYMBOL_GPL(perf_arch_fetch_caller_regs);
+
 static char *perf_trace_buf;
 static char *perf_trace_buf_nmi;
 
Signature · 1 line
1.6.2.3
#9 lore raw

[tip:perf/core] perf: Fix unexported generic perf_arch_fetch_caller_regs

Commit-ID:  1d199b1ad606ae8b88acebd295b101c4e1cf2a57
Gitweb:     http://git.kernel.org/tip/1d199b1ad606ae8b88acebd295b101c4e1cf2a57
Author:     Frederic Weisbecker <[email protected]>
AuthorDate: Tue, 16 Mar 2010 01:05:02 +0100
Committer:  Ingo Molnar <[email protected]>
CommitDate: Tue, 16 Mar 2010 09:27:27 +0100

perf: Fix unexported generic perf_arch_fetch_caller_regs

perf_arch_fetch_caller_regs() is exported for the overriden x86 version, but not for the generic weak version.

As a general rule, weak functions should not have their symbol exported in the same file they are defined.

So let's export it on trace_event_perf.c as it is used by trace events only.

This fixes:

ERROR: ".perf_arch_fetch_caller_regs" [fs/xfs/xfs.ko] undefined!
ERROR: ".perf_arch_fetch_caller_regs" [arch/powerpc/platforms/cell/spufs/spufs.ko] undefined!

-v2: And also only build it if trace events are enabled. -v3: Fix changelog mistake

LKML-Reference: <[email protected]>

---
 arch/x86/kernel/cpu/perf_event.c |    3 ++-
 kernel/perf_event.c              |    2 ++
 kernel/trace/trace_event_perf.c  |    2 ++
 3 files changed, 6 insertions(+), 1 deletions(-)
arch/x86/kernel/cpu/perf_event.c +2 -1
index 978d297..0d3466c 100644
@@ -1695,6 +1695,7 @@ struct perf_callchain_entry *perf_callchain(struct pt_regs *regs)
 	return entry;
 }
 
+#ifdef CONFIG_EVENT_TRACING
 void perf_arch_fetch_caller_regs(struct pt_regs *regs, unsigned long ip, int skip)
 {
 	regs->ip = ip;
@@ -1706,4 +1707,4 @@ void perf_arch_fetch_caller_regs(struct pt_regs *regs, unsigned long ip, int ski
 	regs->cs = __KERNEL_CS;
 	local_save_flags(regs->flags);
 }
-EXPORT_SYMBOL_GPL(perf_arch_fetch_caller_regs);
+#endif
kernel/perf_event.c +2 -0
index 8bf6127..455393e 100644
@@ -2790,10 +2790,12 @@ __weak struct perf_callchain_entry *perf_callchain(struct pt_regs *regs)
 	return NULL;
 }
 
+#ifdef CONFIG_EVENT_TRACING
 __weak
 void perf_arch_fetch_caller_regs(struct pt_regs *regs, unsigned long ip, int skip)
 {
 }
+#endif
 
 /*
  * Output
kernel/trace/trace_event_perf.c +2 -0
index 0709e4f..7d79a10 100644
@@ -12,6 +12,8 @@
 DEFINE_PER_CPU(struct pt_regs, perf_trace_regs);
 EXPORT_PER_CPU_SYMBOL_GPL(perf_trace_regs);
 
+EXPORT_SYMBOL_GPL(perf_arch_fetch_caller_regs);
+
 static char *perf_trace_buf;
 static char *perf_trace_buf_nmi;
 
#10 lore raw

[tip:perf/urgent] perf: Fix unexported generic perf_arch_fetch_caller_regs

In reply to Frederic Weisbecker
Commit-ID:  a6b84574eed7e4fd8cb8dac2d0926fe2cf34b941
Gitweb:     http://git.kernel.org/tip/a6b84574eed7e4fd8cb8dac2d0926fe2cf34b941
Author:     Frederic Weisbecker <[email protected]>
AuthorDate: Tue, 16 Mar 2010 01:05:02 +0100
Committer:  Ingo Molnar <[email protected]>
CommitDate: Wed, 17 Mar 2010 10:44:42 +0100

perf: Fix unexported generic perf_arch_fetch_caller_regs

perf_arch_fetch_caller_regs() is exported for the overriden x86 version, but not for the generic weak version.

As a general rule, weak functions should not have their symbol exported in the same file they are defined.

So let's export it on trace_event_perf.c as it is used by trace events only.

This fixes:

ERROR: ".perf_arch_fetch_caller_regs" [fs/xfs/xfs.ko] undefined!
ERROR: ".perf_arch_fetch_caller_regs" [arch/powerpc/platforms/cell/spufs/spufs.ko] undefined!

-v2: And also only build it if trace events are enabled. -v3: Fix changelog mistake

LKML-Reference: <[email protected]>

---
 arch/x86/kernel/cpu/perf_event.c |    3 ++-
 kernel/perf_event.c              |    2 ++
 kernel/trace/trace_event_perf.c  |    2 ++
 3 files changed, 6 insertions(+), 1 deletions(-)
arch/x86/kernel/cpu/perf_event.c +2 -1
index 7645fae..60398a0 100644
@@ -1702,6 +1702,7 @@ struct perf_callchain_entry *perf_callchain(struct pt_regs *regs)
 	return entry;
 }
 
+#ifdef CONFIG_EVENT_TRACING
 void perf_arch_fetch_caller_regs(struct pt_regs *regs, unsigned long ip, int skip)
 {
 	regs->ip = ip;
@@ -1713,4 +1714,4 @@ void perf_arch_fetch_caller_regs(struct pt_regs *regs, unsigned long ip, int ski
 	regs->cs = __KERNEL_CS;
 	local_save_flags(regs->flags);
 }
-EXPORT_SYMBOL_GPL(perf_arch_fetch_caller_regs);
+#endif
kernel/perf_event.c +2 -0
index fb3031c..574ee58 100644
@@ -2786,10 +2786,12 @@ __weak struct perf_callchain_entry *perf_callchain(struct pt_regs *regs)
 	return NULL;
 }
 
+#ifdef CONFIG_EVENT_TRACING
 __weak
 void perf_arch_fetch_caller_regs(struct pt_regs *regs, unsigned long ip, int skip)
 {
 }
+#endif
 
 /*
  * Output
kernel/trace/trace_event_perf.c +2 -0
index 0709e4f..7d79a10 100644
@@ -12,6 +12,8 @@
 DEFINE_PER_CPU(struct pt_regs, perf_trace_regs);
 EXPORT_PER_CPU_SYMBOL_GPL(perf_trace_regs);
 
+EXPORT_SYMBOL_GPL(perf_arch_fetch_caller_regs);
+
 static char *perf_trace_buf;
 static char *perf_trace_buf_nmi;
 
#11 lore raw

Re: [PATCH] perf: Also export the weak version of perf_arch_fetch_caller_regs

In reply to Stephen Rothwell
Quoted 9 lines · Stephen Rothwell

Hi Frederic,

Quoted 2 lines · On Fri, 12 Mar 2010 06:50:44 +0100 Frederic Weisbecker <[email protected]>…

Does that fixes the issue? Sorry I can not test for now...

I can't test it right now, sorry, but I have heard that EXPORT_SYMBOL's for weak symbols must be in a different file from the actual symbol definition.

Hmm. Ok, kernel/trace/trace_event_perf.c must be the right place then, as the only users of this helper are trace events.

Will send a fix once I get this tested, thanks for the pointer!