Re: [PATCH] ocfs2: embed actual values into ocfs2_sysfile_lock_key names

[Heming Zhao <heming.zhao@suse.com>]

On 6/30/25 18:39, Tetsuo Handa wrote:
> On 2025/06/30 11:42, Heming Zhao wrote:
>> I am not familiar with lockdep, and just have two questions regarding
>> the lockdep in ocfs2.
>>
>> 1>
>> there are three global "static struct lock_class_key" definitions:
>> - fs/ocfs2/inode.c  : ocfs2_sysfile_lock_key[NUM_SYSTEM_INODES]
>> - fs/ocfs2/dlmglue.c: lockdep_keys[OCFS2_NUM_LOCK_TYPES]
>> - fs/ocfs2/sysfile.c: ocfs2_sysfile_cluster_lock_key[NUM_SYSTEM_INODES]
>>
>> why did you env only trigger the ocfs2_sysfile_lock_key[] warning?
> 
> Because syzbot is reporting lockdep warning on ocfs2_sysfile_lock_key
> at https://syzkaller.appspot.com/bug?extid=68c788938ba0326046a9 and
> I couldn't find which lock_class_key syzbot is reporting.
> 
> Unless you want me to update all keys within this patch, you can submit
> similar changes on lockdep_keys and ocfs2_sysfile_cluster_lock_key as
> separate patches.
> 
>>
>> 2>
>> It seems the existing CONFIG_DEBUG_LOCK_ALLOC is incorrect, it should be
>> replaced with CONFIG_LOCKDEP.
> 
> I couldn't catch what you mean. There are many modules which declare
> "struct lock_class_key" under CONFIG_DEBUG_LOCK_ALLOC=y.
> 

I mean OCFS2 should use unified kernel configuration option.

- Heming