Skip to contentSkip to conversation

lockdep: fix warning: print_lock_trace defined but not used

PATCH Superseded [email protected] 3 participants 6 messages May 21, 2019 – Jun 18, 2019 Locking primitives

lore
Share & tools
lore
#0 lore raw

cover lockdep: fix warning: print_lock_trace defined but not used

From: John Hubbard <[email protected]>

Hi,

This is a second attempt to send this. The first attempt [1] went via mailgw.nvidia.com, and did not receive a DKIM signature, so I think it probably got discarded by many systems.

This one goes via gmail, which is what I was using before. We'll keep working on getting the NVIDIA outgoing email servers working with mailing lists, but meanwhile I'm back to using "envelope" techniques, I guess. I was hoping to use a "native" email address, but not quite there yet. :)

Anyway, here's the original cover letter:

I ran across this minor warning while building against today's linux.git.

The proposed trivial fix leaves it a little fragile from a "what if someone adds a new call to print_lock_trace()" point of view, but I believe that it makes the current combination of ifdefs accurate, anyway.

[1] https://lkml.org/lkml/2019/5/19/60

Cc Frederic Weisbecker <[email protected]>
Cc Andrew Morton <[email protected]>
Cc Linus Torvalds <[email protected]>
Cc Paul E. McKenney <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc Will Deacon <[email protected]>

John Hubbard (1):
lockdep: fix warning: print_lock_trace defined but not used

 kernel/locking/lockdep.c | 3 +++
 1 file changed, 3 insertions(+)
Signature · 1 line
2.21.0
#1 lore raw

PATCH 1/1 lockdep: fix warning: print_lock_trace defined but not used

From: John Hubbard <[email protected]>

Commit 0d2cc3b34532 ("locking/lockdep: Move valid_state() inside CONFIG_TRACE_IRQFLAGS && CONFIG_PROVE_LOCKING") moved the only usage of print_lock_trace() that was originally outside of the CONFIG_PROVE_LOCKING case. It moved that usage into a different case: CONFIG_PROVE_LOCKING && CONFIG_TRACE_IRQFLAGS. That leaves things not symmetrical, and as a result, the following warning fires on my build, when I have

!CONFIG_TRACE_IRQFLAGS && !CONFIG_PROVE_LOCKING

set:

kernel/locking/lockdep.c:2821:13: warning: ‘print_lock_trace’ defined
but not used [-Wunused-function]

Fix this by only defining print_lock_trace() in cases in which is it called.

Fixes 0d2cc3b34532 ("locking/lockdep: Move valid_state() inside CONFIG_TRACE_IRQFLAGS && CONFIG_PROVE_LOCKING")
Cc Frederic Weisbecker <[email protected]>
Cc Andrew Morton <[email protected]>
Cc Linus Torvalds <[email protected]>
Cc Paul E. McKenney <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc Will Deacon <[email protected]>
---
 kernel/locking/lockdep.c | 3 +++
 1 file changed, 3 insertions(+)
kernel/locking/lockdep.c +3 -0
index d06190fa5082..3065dc36c27a 100644
@@ -2817,11 +2817,14 @@ static inline int validate_chain(struct task_struct *curr,
 	return 1;
 }
 
+#if defined(CONFIG_TRACE_IRQFLAGS)
 static void print_lock_trace(struct lock_trace *trace, unsigned int spaces)
 {
 }
 #endif
 
+#endif
+
 /*
  * We are building curr_chain_key incrementally, so double-check
  * it from scratch, to make sure that it's done correctly:
Signature · 1 line
2.21.0
#2 lore raw
Quoted 37 lines · On Tue, May 21, 2019 at 12:08:08AM -0700, [email protected] wrote:

From: John Hubbard <[email protected]>

Commit 0d2cc3b34532 ("locking/lockdep: Move valid_state() inside CONFIG_TRACE_IRQFLAGS && CONFIG_PROVE_LOCKING") moved the only usage of print_lock_trace() that was originally outside of the CONFIG_PROVE_LOCKING case. It moved that usage into a different case: CONFIG_PROVE_LOCKING && CONFIG_TRACE_IRQFLAGS. That leaves things not symmetrical, and as a result, the following warning fires on my build, when I have

!CONFIG_TRACE_IRQFLAGS && !CONFIG_PROVE_LOCKING

set:

kernel/locking/lockdep.c:2821:13: warning: ‘print_lock_trace’ defined
but not used [-Wunused-function]

Fix this by only defining print_lock_trace() in cases in which is it called.

Fixes 0d2cc3b34532 ("locking/lockdep: Move valid_state() inside CONFIG_TRACE_IRQFLAGS && CONFIG_PROVE_LOCKING")
Cc Frederic Weisbecker <[email protected]>
Cc Andrew Morton <[email protected]>
Cc Linus Torvalds <[email protected]>
Cc Paul E. McKenney <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc Will Deacon <[email protected]>
---
 kernel/locking/lockdep.c | 3 +++
 1 file changed, 3 insertions(+)
kernel/locking/lockdep.c +0 -0
index d06190fa5082..3065dc36c27a 100644

kernel/locking/lockdep.c static inline int validate_chain(struct task_struct *curr, · line 2821

Quoted 5 lines
kernel/locking/lockdep.c +1 -0
@@ -2817,11 +2817,14 @@ static inline int validate_chain(struct task_struct *curr,
 	return 1;
 }
 
+#if defined(CONFIG_TRACE_IRQFLAGS)
 static void print_lock_trace(struct lock_trace *trace, unsigned int spaces)

This works, but another approach is to put "__maybe_unused" in the above declaration, which avoids the need to have "#if" in a .c file. But this file already has quite a few #if and #ifdef commands, so maybe it is OK here.

Also, "#ifdef CONFIG_TRACE_IRQFLAGS" is a bit more conventional than the above, should the "__maybe_unused" be undesirable.

Yet another approach is to move this function to include/linux/lockdep.h, where #ifdef is considered less objectionable.

But I must defer to the maintainers.

Thanx, Paul

Quoted 12 lines
kernel/locking/lockdep.c +2 -0
 {
 }
 #endif
 
+#endif
+
 /*
  * We are building curr_chain_key incrementally, so double-check
  * it from scratch, to make sure that it's done correctly:
Signature · 1 line
2.21.0
#3 lore raw
Quoted 41 lines · Paul E. McKenney
Quoted 40 lines · On Tue, May 21, 2019 at 12:08:08AM -0700, [email protected] wrote:

From: John Hubbard <[email protected]>

Commit 0d2cc3b34532 ("locking/lockdep: Move valid_state() inside CONFIG_TRACE_IRQFLAGS && CONFIG_PROVE_LOCKING") moved the only usage of print_lock_trace() that was originally outside of the CONFIG_PROVE_LOCKING case. It moved that usage into a different case: CONFIG_PROVE_LOCKING && CONFIG_TRACE_IRQFLAGS. That leaves things not symmetrical, and as a result, the following warning fires on my build, when I have

!CONFIG_TRACE_IRQFLAGS && !CONFIG_PROVE_LOCKING

set:

kernel/locking/lockdep.c:2821:13: warning: ‘print_lock_trace’ defined
but not used [-Wunused-function]

Fix this by only defining print_lock_trace() in cases in which is it called.

Fixes 0d2cc3b34532 ("locking/lockdep: Move valid_state() inside CONFIG_TRACE_IRQFLAGS && CONFIG_PROVE_LOCKING")
Cc Frederic Weisbecker <[email protected]>
Cc Andrew Morton <[email protected]>
Cc Linus Torvalds <[email protected]>
Cc Paul E. McKenney <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc Will Deacon <[email protected]>
---
 kernel/locking/lockdep.c | 3 +++
 1 file changed, 3 insertions(+)
kernel/locking/lockdep.c +0 -0
index d06190fa5082..3065dc36c27a 100644
@@ -2817,11 +2817,14 @@ static inline int validate_chain(struct task_struct *curr,
 	return 1;
 }
 
Quoted 5 lines

kernel/locking/lockdep.c static inline int validate_chain(struct task_struct *curr, · line 2821

Quoted 2 lines
kernel/locking/lockdep.c +1 -0
+#if defined(CONFIG_TRACE_IRQFLAGS)
 static void print_lock_trace(struct lock_trace *trace, unsigned int spaces)

This works, but another approach is to put "__maybe_unused" in the above declaration, which avoids the need to have "#if" in a .c file.

Good idea, that approach appeals to me here, because tracing is a natural fit for "might not be used" types of functions.

Quoted 5 lines

But this file already has quite a few #if and #ifdef commands, so maybe it is OK here.

Also, "#ifdef CONFIG_TRACE_IRQFLAGS" is a bit more conventional than the above, should the "__maybe_unused" be undesirable.

ah, OK, I'll keep that in mind. (The two seemed identical to my mind, but it's good to make things look like surrounding code, of course.)

thanks,

Signature · 2 lines
John Hubbard
NVIDIA
Quoted 7 lines

Yet another approach is to move this function to include/linux/lockdep.h, where #ifdef is considered less objectionable.

But I must defer to the maintainers.

Thanx, Paul

#4 lore raw

PATCH v2 1/1 lockdep: fix warning: print_lock_trace defined but not used

In reply to Paul E. McKenney

From: John Hubbard <[email protected]>

Commit 0d2cc3b34532 ("locking/lockdep: Move valid_state() inside CONFIG_TRACE_IRQFLAGS && CONFIG_PROVE_LOCKING") moved the only usage of print_lock_trace() that was originally outside of the CONFIG_PROVE_LOCKING case. It moved that usage into a different case: CONFIG_PROVE_LOCKING && CONFIG_TRACE_IRQFLAGS. That leaves things not symmetrical, and as a result, the following warning fires on my build, when I have

!CONFIG_TRACE_IRQFLAGS && !CONFIG_PROVE_LOCKING

set:

kernel/locking/lockdep.c:2821:13: warning: ‘print_lock_trace’ defined
but not used [-Wunused-function]

Fix this by annotating print_lock_trace() with "__maybe_unused". Thanks to Paul E. McKenney for suggesting this less intrusive fix, as compared to adding more ifdef noise.

Fixes 0d2cc3b34532 ("locking/lockdep: Move valid_state() inside CONFIG_TRACE_IRQFLAGS && CONFIG_PROVE_LOCKING")
Cc Frederic Weisbecker <[email protected]>
Cc Andrew Morton <[email protected]>
Cc Linus Torvalds <[email protected]>
Cc Paul E. McKenney <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc Will Deacon <[email protected]>
---
 kernel/locking/lockdep.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
kernel/locking/lockdep.c +2 -1
index c47788fa85f9..2726dafdb29b 100644
@@ -2818,7 +2818,8 @@ static inline int validate_chain(struct task_struct *curr,
 	return 1;
 }
 
-static void print_lock_trace(struct lock_trace *trace, unsigned int spaces)
+static void __maybe_unused print_lock_trace(struct lock_trace *trace,
+					    unsigned int spaces)
 {
 }
 #endif
Signature · 1 line
2.21.0
Tested-by on v2 1/1
#5 lore raw
Quoted 22 lines · On Sun, Jun 09, 2019 at 05:09:33PM -0700, [email protected] wrote:

From: John Hubbard <[email protected]>

Commit 0d2cc3b34532 ("locking/lockdep: Move valid_state() inside CONFIG_TRACE_IRQFLAGS && CONFIG_PROVE_LOCKING") moved the only usage of print_lock_trace() that was originally outside of the CONFIG_PROVE_LOCKING case. It moved that usage into a different case: CONFIG_PROVE_LOCKING && CONFIG_TRACE_IRQFLAGS. That leaves things not symmetrical, and as a result, the following warning fires on my build, when I have

!CONFIG_TRACE_IRQFLAGS && !CONFIG_PROVE_LOCKING

set:

kernel/locking/lockdep.c:2821:13: warning: ‘print_lock_trace’ defined
but not used [-Wunused-function]

Fix this by annotating print_lock_trace() with "__maybe_unused". Thanks to Paul E. McKenney for suggesting this less intrusive fix, as compared to adding more ifdef noise.

Fixes 0d2cc3b34532 ("locking/lockdep: Move valid_state() inside CONFIG_TRACE_IRQFLAGS && CONFIG_PROVE_LOCKING")
Quoted 8 lines
Cc Frederic Weisbecker <[email protected]>
Cc Andrew Morton <[email protected]>
Cc Linus Torvalds <[email protected]>
Cc Paul E. McKenney <[email protected]>
Cc Peter Zijlstra <[email protected]>
Cc Thomas Gleixner <[email protected]>
Cc Will Deacon <[email protected]>
Quoted 21 lines
---
 kernel/locking/lockdep.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
kernel/locking/lockdep.c +2 -1
index c47788fa85f9..2726dafdb29b 100644
@@ -2818,7 +2818,8 @@ static inline int validate_chain(struct task_struct *curr,
 	return 1;
 }
 
-static void print_lock_trace(struct lock_trace *trace, unsigned int spaces)
+static void __maybe_unused print_lock_trace(struct lock_trace *trace,
+					    unsigned int spaces)
 {
 }
 #endif
Signature · 1 line
2.21.0