[PATCH] qspinlock: no need to check return value of debugfs_create functions

[PATCH] qspinlock: no need to check return value of debugfs_create functions

[Greg Kroah-Hartman]

When calling debugfs functions, there is no need to ever check the
return value.  The function can work or not, but the code logic should
never do something different based on this.

Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Will Deacon <will.deacon@arm.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
---
 kernel/locking/qspinlock_stat.h | 18 ++++--------------
 1 file changed, 4 insertions(+), 14 deletions(-)

diff --git a/kernel/locking/qspinlock_stat.h b/kernel/locking/qspinlock_stat.h
index 42d3d8dc8f49..766fa0c8c115 100644
--- a/kernel/locking/qspinlock_stat.h
+++ b/kernel/locking/qspinlock_stat.h
@@ -213,9 +213,6 @@ static int __init init_qspinlock_stat(void)
 	struct dentry *d_qstat = debugfs_create_dir("qlockstat", NULL);
 	int i;
 
-	if (!d_qstat)
-		goto out;
-
 	/*
 	 * Create the debugfs files
 	 *
@@ -224,20 +221,13 @@ static int __init init_qspinlock_stat(void)
 	 * performance.
 	 */
 	for (i = 0; i < qstat_num; i++)
-		if (!debugfs_create_file(qstat_names[i], 0400, d_qstat,
-					 (void *)(long)i, &fops_qstat))
-			goto fail_undo;
+		debugfs_create_file(qstat_names[i], 0400, d_qstat,
+				    (void *)(long)i, &fops_qstat);
 
-	if (!debugfs_create_file(qstat_names[qstat_reset_cnts], 0200, d_qstat,
-				 (void *)(long)qstat_reset_cnts, &fops_qstat))
-		goto fail_undo;
+	debugfs_create_file(qstat_names[qstat_reset_cnts], 0200, d_qstat,
+			    (void *)(long)qstat_reset_cnts, &fops_qstat);
 
 	return 0;
-fail_undo:
-	debugfs_remove_recursive(d_qstat);
-out:
-	pr_warn("Could not create 'qlockstat' debugfs entries\n");
-	return -ENOMEM;
 }
 fs_initcall(init_qspinlock_stat);
 
-- 
2.20.1

Re: [PATCH] qspinlock: no need to check return value of debugfs_create functions

[Peter Zijlstra]

On Tue, Jan 22, 2019 at 04:21:43PM +0100, Greg Kroah-Hartman wrote:
> When calling debugfs functions, there is no need to ever check the
> return value.  The function can work or not, but the code logic should
> never do something different based on this.

So I've seen you do a fair number of these patches; but I don't
fully understand.

The existing code rolls back the created files such that we either have
all files or none at all. Why is this wrong?

It for some daft reason one of the debugfs calls fails (imagine this was
a module and we did modprobe while under memory pressure), why should we
present a partial interface to the 'user' ?

> Cc: Peter Zijlstra <peterz@infradead.org>
> Cc: Ingo Molnar <mingo@redhat.com>
> Cc: Will Deacon <will.deacon@arm.com>
> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
> ---
>  kernel/locking/qspinlock_stat.h | 18 ++++--------------
>  1 file changed, 4 insertions(+), 14 deletions(-)
> 
> diff --git a/kernel/locking/qspinlock_stat.h b/kernel/locking/qspinlock_stat.h
> index 42d3d8dc8f49..766fa0c8c115 100644
> --- a/kernel/locking/qspinlock_stat.h
> +++ b/kernel/locking/qspinlock_stat.h
> @@ -213,9 +213,6 @@ static int __init init_qspinlock_stat(void)
>  	struct dentry *d_qstat = debugfs_create_dir("qlockstat", NULL);
>  	int i;
>  
> -	if (!d_qstat)
> -		goto out;
> -
>  	/*
>  	 * Create the debugfs files
>  	 *
> @@ -224,20 +221,13 @@ static int __init init_qspinlock_stat(void)
>  	 * performance.
>  	 */
>  	for (i = 0; i < qstat_num; i++)
> -		if (!debugfs_create_file(qstat_names[i], 0400, d_qstat,
> -					 (void *)(long)i, &fops_qstat))
> -			goto fail_undo;
> +		debugfs_create_file(qstat_names[i], 0400, d_qstat,
> +				    (void *)(long)i, &fops_qstat);
>  
> -	if (!debugfs_create_file(qstat_names[qstat_reset_cnts], 0200, d_qstat,
> -				 (void *)(long)qstat_reset_cnts, &fops_qstat))
> -		goto fail_undo;
> +	debugfs_create_file(qstat_names[qstat_reset_cnts], 0200, d_qstat,
> +			    (void *)(long)qstat_reset_cnts, &fops_qstat);
>  
>  	return 0;
> -fail_undo:
> -	debugfs_remove_recursive(d_qstat);
> -out:
> -	pr_warn("Could not create 'qlockstat' debugfs entries\n");
> -	return -ENOMEM;
>  }
>  fs_initcall(init_qspinlock_stat);
>  
> -- 
> 2.20.1
> 

Re: [PATCH] qspinlock: no need to check return value of debugfs_create functions

[Will Deacon]

On Tue, Jan 22, 2019 at 04:21:43PM +0100, Greg Kroah-Hartman wrote:
> When calling debugfs functions, there is no need to ever check the
> return value.  The function can work or not, but the code logic should
> never do something different based on this.
> 
> Cc: Peter Zijlstra <peterz@infradead.org>
> Cc: Ingo Molnar <mingo@redhat.com>
> Cc: Will Deacon <will.deacon@arm.com>
> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
> ---
>  kernel/locking/qspinlock_stat.h | 18 ++++--------------
>  1 file changed, 4 insertions(+), 14 deletions(-)

Curious, but are you planning to make these functions return void after
you've modified all of the callers?

Will

Re: [PATCH] qspinlock: no need to check return value of debugfs_create functions

[Greg Kroah-Hartman]

On Wed, Jan 23, 2019 at 09:10:18AM +0000, Will Deacon wrote:
> On Tue, Jan 22, 2019 at 04:21:43PM +0100, Greg Kroah-Hartman wrote:
> > When calling debugfs functions, there is no need to ever check the
> > return value.  The function can work or not, but the code logic should
> > never do something different based on this.
> > 
> > Cc: Peter Zijlstra <peterz@infradead.org>
> > Cc: Ingo Molnar <mingo@redhat.com>
> > Cc: Will Deacon <will.deacon@arm.com>
> > Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
> > ---
> >  kernel/locking/qspinlock_stat.h | 18 ++++--------------
> >  1 file changed, 4 insertions(+), 14 deletions(-)
> 
> Curious, but are you planning to make these functions return void after
> you've modified all of the callers?

I wish, but I can't, as those values are needed for some other debugfs
calls.

thanks,

greg k-h

Re: [PATCH] qspinlock: no need to check return value of debugfs_create functions

[Greg Kroah-Hartman]

Sorry for the delay, got stuck with other things...

On Wed, Jan 23, 2019 at 09:14:41AM +0100, Peter Zijlstra wrote:
> On Tue, Jan 22, 2019 at 04:21:43PM +0100, Greg Kroah-Hartman wrote:
> > When calling debugfs functions, there is no need to ever check the
> > return value.  The function can work or not, but the code logic should
> > never do something different based on this.
> 
> So I've seen you do a fair number of these patches; but I don't
> fully understand.
> 
> The existing code rolls back the created files such that we either have
> all files or none at all. Why is this wrong?

Normally you should not care if debugfs files can, or can not, be
created.  So just call the debugfs code and move on.  The worst case is
where valid kernel code aborts just because debugging files could not be
created for some reason, which is not ok.

> It for some daft reason one of the debugfs calls fails (imagine this was
> a module and we did modprobe while under memory pressure), why should we
> present a partial interface to the 'user' ?

Because it's not an interface anyone should ever depend on :)

But, for some code, that only uses debugfs, maybe it does make sense to
just abort the whole thing and keep going on.  If this code should work
that way, my mistake, let's leave it alone.

The overall goal with these patches is to make the code surrounding
using debugfs easier and simpler.  You should not need to handle any
errors from debugfs, as it should never be used for anything that
actually matters to the system, except when trying to debug it.

thanks,

greg k-h