2020-11-07 09:21:24

by Tiezhu Yang

[permalink] [raw]
Subject: [PATCH] locking/lock_events: no need to check return value of debugfs_create functions

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.

Signed-off-by: Tiezhu Yang <[email protected]>
---
kernel/locking/lock_events.c | 19 ++++---------------
1 file changed, 4 insertions(+), 15 deletions(-)

diff --git a/kernel/locking/lock_events.c b/kernel/locking/lock_events.c
index fa2c2f9..bac77a1 100644
--- a/kernel/locking/lock_events.c
+++ b/kernel/locking/lock_events.c
@@ -146,9 +146,6 @@ static int __init init_lockevent_counts(void)
struct dentry *d_counts = debugfs_create_dir(LOCK_EVENTS_DIR, NULL);
int i;

- if (!d_counts)
- goto out;
-
/*
* Create the debugfs files
*
@@ -159,21 +156,13 @@ static int __init init_lockevent_counts(void)
for (i = 0; i < lockevent_num; i++) {
if (skip_lockevent(lockevent_names[i]))
continue;
- if (!debugfs_create_file(lockevent_names[i], 0400, d_counts,
- (void *)(long)i, &fops_lockevent))
- goto fail_undo;
+ debugfs_create_file(lockevent_names[i], 0400, d_counts,
+ (void *)(long)i, &fops_lockevent);
}

- if (!debugfs_create_file(lockevent_names[LOCKEVENT_reset_cnts], 0200,
- d_counts, (void *)(long)LOCKEVENT_reset_cnts,
- &fops_lockevent))
- goto fail_undo;
+ debugfs_create_file(lockevent_names[LOCKEVENT_reset_cnts], 0200, d_counts,
+ (void *)(long)LOCKEVENT_reset_cnts, &fops_lockevent);

return 0;
-fail_undo:
- debugfs_remove_recursive(d_counts);
-out:
- pr_warn("Could not create '%s' debugfs entries\n", LOCK_EVENTS_DIR);
- return -ENOMEM;
}
fs_initcall(init_lockevent_counts);
--
2.1.0


2020-11-09 08:35:09

by Peter Zijlstra

[permalink] [raw]
Subject: Re: [PATCH] locking/lock_events: no need to check return value of debugfs_create functions

On Sat, Nov 07, 2020 at 05:19:13PM +0800, Tiezhu Yang 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.

I strongly disagree and have told this to Greg before. Having half a
debug interface is weird at best, so upon failure we remove the whole
thing, which is consistent.

> Signed-off-by: Tiezhu Yang <[email protected]>
> ---
> kernel/locking/lock_events.c | 19 ++++---------------
> 1 file changed, 4 insertions(+), 15 deletions(-)
>
> diff --git a/kernel/locking/lock_events.c b/kernel/locking/lock_events.c
> index fa2c2f9..bac77a1 100644
> --- a/kernel/locking/lock_events.c
> +++ b/kernel/locking/lock_events.c
> @@ -146,9 +146,6 @@ static int __init init_lockevent_counts(void)
> struct dentry *d_counts = debugfs_create_dir(LOCK_EVENTS_DIR, NULL);
> int i;
>
> - if (!d_counts)
> - goto out;
> -
> /*
> * Create the debugfs files
> *
> @@ -159,21 +156,13 @@ static int __init init_lockevent_counts(void)
> for (i = 0; i < lockevent_num; i++) {
> if (skip_lockevent(lockevent_names[i]))
> continue;
> - if (!debugfs_create_file(lockevent_names[i], 0400, d_counts,
> - (void *)(long)i, &fops_lockevent))
> - goto fail_undo;
> + debugfs_create_file(lockevent_names[i], 0400, d_counts,
> + (void *)(long)i, &fops_lockevent);
> }
>
> - if (!debugfs_create_file(lockevent_names[LOCKEVENT_reset_cnts], 0200,
> - d_counts, (void *)(long)LOCKEVENT_reset_cnts,
> - &fops_lockevent))
> - goto fail_undo;
> + debugfs_create_file(lockevent_names[LOCKEVENT_reset_cnts], 0200, d_counts,
> + (void *)(long)LOCKEVENT_reset_cnts, &fops_lockevent);
>
> return 0;
> -fail_undo:
> - debugfs_remove_recursive(d_counts);
> -out:
> - pr_warn("Could not create '%s' debugfs entries\n", LOCK_EVENTS_DIR);
> - return -ENOMEM;
> }
> fs_initcall(init_lockevent_counts);
> --
> 2.1.0
>

2020-11-09 09:55:01

by Tiezhu Yang

[permalink] [raw]
Subject: Re: [PATCH] locking/lock_events: no need to check return value of debugfs_create functions

On 11/09/2020 04:32 PM, Peter Zijlstra wrote:
> On Sat, Nov 07, 2020 at 05:19:13PM +0800, Tiezhu Yang 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.
> I strongly disagree and have told this to Greg before. Having half a
> debug interface is weird at best, so upon failure we remove the whole
> thing, which is consistent.

Hi Peter,

Thanks for your reply.

I find the early discussion and see the following opinion by Greg:

https://lore.kernel.org/patchwork/patch/1290162/

[ For debugfs, this isn't an issue, what can a user do with something like
"debugfs isn't working? What does that mean???"

And if we _really_ want warnings like this, it should go into the
debugfs core, not require this to be done for every debugfs user, right?

debugfs is just there for kernel developers to help debug things, it's
not a dependancy on any userspace functionality, so if it works or not
should not be an issue for any user.

Unless that user is a kernel developer of course :)

thanks,

greg k-h ]

Anyway, if this patch is meaningless after discussion, please ignore it.

Thanks,
Tiezhu

>
>> Signed-off-by: Tiezhu Yang <[email protected]>
>> ---
>> kernel/locking/lock_events.c | 19 ++++---------------
>> 1 file changed, 4 insertions(+), 15 deletions(-)
>>
>> diff --git a/kernel/locking/lock_events.c b/kernel/locking/lock_events.c
>> index fa2c2f9..bac77a1 100644
>> --- a/kernel/locking/lock_events.c
>> +++ b/kernel/locking/lock_events.c
>> @@ -146,9 +146,6 @@ static int __init init_lockevent_counts(void)
>> struct dentry *d_counts = debugfs_create_dir(LOCK_EVENTS_DIR, NULL);
>> int i;
>>
>> - if (!d_counts)
>> - goto out;
>> -
>> /*
>> * Create the debugfs files
>> *
>> @@ -159,21 +156,13 @@ static int __init init_lockevent_counts(void)
>> for (i = 0; i < lockevent_num; i++) {
>> if (skip_lockevent(lockevent_names[i]))
>> continue;
>> - if (!debugfs_create_file(lockevent_names[i], 0400, d_counts,
>> - (void *)(long)i, &fops_lockevent))
>> - goto fail_undo;
>> + debugfs_create_file(lockevent_names[i], 0400, d_counts,
>> + (void *)(long)i, &fops_lockevent);
>> }
>>
>> - if (!debugfs_create_file(lockevent_names[LOCKEVENT_reset_cnts], 0200,
>> - d_counts, (void *)(long)LOCKEVENT_reset_cnts,
>> - &fops_lockevent))
>> - goto fail_undo;
>> + debugfs_create_file(lockevent_names[LOCKEVENT_reset_cnts], 0200, d_counts,
>> + (void *)(long)LOCKEVENT_reset_cnts, &fops_lockevent);
>>
>> return 0;
>> -fail_undo:
>> - debugfs_remove_recursive(d_counts);
>> -out:
>> - pr_warn("Could not create '%s' debugfs entries\n", LOCK_EVENTS_DIR);
>> - return -ENOMEM;
>> }
>> fs_initcall(init_lockevent_counts);
>> --
>> 2.1.0
>>

2020-11-09 10:05:14

by Greg Kroah-Hartman

[permalink] [raw]
Subject: Re: [PATCH] locking/lock_events: no need to check return value of debugfs_create functions

On Mon, Nov 09, 2020 at 05:51:56PM +0800, Tiezhu Yang wrote:
> On 11/09/2020 04:32 PM, Peter Zijlstra wrote:
> > On Sat, Nov 07, 2020 at 05:19:13PM +0800, Tiezhu Yang 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.
> > I strongly disagree and have told this to Greg before. Having half a
> > debug interface is weird at best, so upon failure we remove the whole
> > thing, which is consistent.
>
> Hi Peter,
>
> Thanks for your reply.
>
> I find the early discussion and see the following opinion by Greg:
>
> https://lore.kernel.org/patchwork/patch/1290162/
>
> [ For debugfs, this isn't an issue, what can a user do with something like
> "debugfs isn't working? What does that mean???"
>
> And if we _really_ want warnings like this, it should go into the
> debugfs core, not require this to be done for every debugfs user, right?

The debugfs core does spit out a warning when this happens, so no need
to duplicate it in your code as well.

And for subsystems that _really_ want to check this, that's fine, it's
the minority for the whole tree, but please, document it well with a
comment on the check so that it doesn't get "cleanup" patches sent for
it in the future.

thanks,

greg k-h