On 5/8/23 16:38, Dan Carpenter wrote:
> On Mon, May 08, 2023 at 09:40:41PM +0800, Dongliang Mu wrote:
>>>>> diff --git a/drivers/scsi/mpt3sas/mpt3sas_debugfs.c b/drivers/scsi/mpt3sas/mpt3sas_debugfs.c
>>>>> index a6ab1db81167..c92e08c130b9 100644
>>>>> --- a/drivers/scsi/mpt3sas/mpt3sas_debugfs.c
>>>>> +++ b/drivers/scsi/mpt3sas/mpt3sas_debugfs.c
>>>>> @@ -99,8 +99,6 @@ static const struct file_operations mpt3sas_debugfs_iocdump_fops = {
>>>>> void mpt3sas_init_debugfs(void)
>>>>> {
>>>>> mpt3sas_debugfs_root = debugfs_create_dir("mpt3sas", NULL);
>>>>> - if (!mpt3sas_debugfs_root)
>>>>> - pr_info("mpt3sas: Cannot create debugfs root\n");
>>>> Hi Jing,
>>>> most drivers just ignore the return value but here the author wanted to
>>>> have the information logged.
>>>> Can you instead of removing the message modify the 'if' condition so it
>>>> suits the author's intention?
>>>
>>> This code was always just wrong.
>>>
>>> The history of this is slightly complicated and boring. These days it's
>>> harmless dead code so I guess it's less bad than before.
>>
>> Hi Dan and Tomas,
>>
>> Any conclusion about this patch? The student Jing Xu is not sure about how
>> to revise this patch.
>
> The correct fix is to delete the code.
>
> Debugfs code has error checking built in and was never supposed to be
> checked for errors in normal driver code.
>
> Originally, debugfs returned a mix of error pointers and NULL. In the
> kernel, when you have a mix of error pointers and NULL, then the NULL
> means that the feature has been disabled deliberately. It's not an
> error, we should not print a message.
>
> So a different, correct-ish way to write write debugfs error handling
> was to say:
>
> mpt3sas_debugfs_root = debugfs_create_dir("mpt3sas", NULL);
> if (IS_ERR(mpt3sas_debugfs_root))
> return PTR_ERR(mpt3sas_debugfs_root);
I'm fine with this as well, I could wish we get a fix for the exact same
case of debugfs_create_dir in mpt3sas_setup_debugfs and ideally all the
debugfs_create* in mpt3sas_debugfs.c in a single patch. But this patch
is ok even if that wasn't possible.
tomash
>
> However, in those days, a lot of people didn't understand error pointers
> and thought that "if (IS_ERR_OR_NULL(mpt3sas_debugfs_root)) {" was a
> super secure way to check for errors. Or they just got it wrong and
> checked for NULL instead of error pointers. Any of the checks are
> wrong, but if (IS_ERR()) check was at least correct-ish.
>
> I dealt with this a lot because of my work with Smatch. I used to be
> happy if I could persuade someone to write at least correct-ish code,
> but it was pretty painful to try explain this over and over and very few
> people deleted the checks.
>
> Eventually Greg changed the code to never return NULL and mass deleted
> the IS_ERR() checks. Not returning NULL makes it simpler to understand.
> And it makes it impossible to check in the correct-ish way so it kind of
> forces people to just delete the error handling.
>
> regards,
> dan carpenter
>
