On 2/15/22 09:05, Dave Chinner wrote:
> On Tue, Feb 15, 2022 at 07:49:27AM +0900, Damien Le Moal wrote:
>> On 2/15/22 06:35, Dave Chinner wrote:
>>> On Thu, Feb 10, 2022 at 02:59:04PM +0900, Naohiro Aota wrote:
>>>> Add an assert function sb_assert_write_started() to check if
>>>> sb_start_write() is properly called. It is used in the next commit.
>>>>
>>>> Also, add the assert functions for sb_start_pagefault() and
>>>> sb_start_intwrite().
>>>>
>>>> Signed-off-by: Naohiro Aota <naohiro.aota@xxxxxxx>
>>>> ---
>>>> include/linux/fs.h | 20 ++++++++++++++++++++
>>>> 1 file changed, 20 insertions(+)
>>>>
>>>> diff --git a/include/linux/fs.h b/include/linux/fs.h
>>>> index bbf812ce89a8..5d5dc9a276d9 100644
>>>> --- a/include/linux/fs.h
>>>> +++ b/include/linux/fs.h
>>>> @@ -1820,6 +1820,11 @@ static inline bool __sb_start_write_trylock(struct super_block *sb, int level)
>>>> #define __sb_writers_release(sb, lev) \
>>>> percpu_rwsem_release(&(sb)->s_writers.rw_sem[(lev)-1], 1, _THIS_IP_)
>>>>
>>>> +static inline void __sb_assert_write_started(struct super_block *sb, int level)
>>>> +{
>>>> + lockdep_assert_held_read(sb->s_writers.rw_sem + level - 1);
>>>> +}
>>>> +
>>>
>>> So this isn't an assert, it's a WARN_ON(). Asserts stop execution
>>> (i.e. kill the task) rather than just issue a warning, so let's not
>>> name a function that issues a warning "assert"...
>>>
>>> Hence I'd much rather see this implemented as:
>>>
>>> static inline bool __sb_write_held(struct super_block *sb, int level)
>>> {
>>> return lockdep_is_held_type(sb->s_writers.rw_sem + level - 1, 1);
>>> }
>>
>> Since this would be true when called in between __sb_start_write() and
>> __sb_end_write(), what about calling it __sb_write_started() ? That
>> disconnects from the fact that the implementation uses a sem.
>
> Makes no difference to me; I initially was going to suggest
> *_inprogress() but that seemed a bit verbose. We don't need to
> bikeshed this to death - all I want is it to be a check that can be
> used for generic purposes rather than being an explicit assert.
agree.
>
> Cheers,
>
> Dave.
--
Damien Le Moal
Western Digital Research
