diff mbox series

[v2,2/3] mm/damon/paddr: minor refactor of damon_pa_young()

Message ID 20230303084343.171958-3-wangkefeng.wang@huawei.com
State New
Headers
Series mm/damon/paddr: minor code improvement |

Commit Message

Kefeng Wang March 3, 2023, 8:43 a.m. UTC 
  Omit three lines by unified folio_put(), and make code more clear.

Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>
---
 mm/damon/paddr.c | 11 ++++-------
 1 file changed, 4 insertions(+), 7 deletions(-)

Comments

SeongJae Park March 3, 2023, 6:39 p.m. UTC | #1 
Hi Kefeng,

On Fri, 3 Mar 2023 16:43:42 +0800 Kefeng Wang <wangkefeng.wang@huawei.com> wrote:

> Omit three lines by unified folio_put(), and make code more clear.
> 
> Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>
> ---
>  mm/damon/paddr.c | 11 ++++-------
>  1 file changed, 4 insertions(+), 7 deletions(-)
> 
> diff --git a/mm/damon/paddr.c b/mm/damon/paddr.c
> index 3fda00a0f786..2ef9db0189ca 100644
> --- a/mm/damon/paddr.c
> +++ b/mm/damon/paddr.c
> @@ -130,24 +130,21 @@ static bool damon_pa_young(unsigned long paddr, unsigned long *folio_sz)
>  			accessed = false;
>  		else
>  			accessed = true;
> -		folio_put(folio);
>  		goto out;

Because you moved 'out' label to not include *folio_sz setting, folio_sz will
not set in this case.  It should be set.

>  	}
>  
>  	need_lock = !folio_test_anon(folio) || folio_test_ksm(folio);
> -	if (need_lock && !folio_trylock(folio)) {
> -		folio_put(folio);
> -		return false;
> -	}
> +	if (need_lock && !folio_trylock(folio))
> +		goto out;
>  
>  	rmap_walk(folio, &rwc);
>  
>  	if (need_lock)
>  		folio_unlock(folio);
> -	folio_put(folio);
>  
> -out:
>  	*folio_sz = folio_size(folio);
> +out:
> +	folio_put(folio);

Before this change, folio_size() is called after folio_put().  Shouldn't it be
called before folio_put()?  If so, could we make a separate fix for that first,
and then make this change on top of it, so that it can be easily applied to
relevant stable kernels?


Thanks,
SJ

>  	return accessed;
>  }
>  
> -- 
> 2.35.3
> 
>
Kefeng Wang March 6, 2023, 1:10 a.m. UTC | #2 
On 2023/3/4 2:39, SeongJae Park wrote:
> Hi Kefeng,
> 
> On Fri, 3 Mar 2023 16:43:42 +0800 Kefeng Wang <wangkefeng.wang@huawei.com> wrote:
> 
>> Omit three lines by unified folio_put(), and make code more clear.
>>
>> Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>
>> ---
>>   mm/damon/paddr.c | 11 ++++-------
>>   1 file changed, 4 insertions(+), 7 deletions(-)
>>
>> diff --git a/mm/damon/paddr.c b/mm/damon/paddr.c
>> index 3fda00a0f786..2ef9db0189ca 100644
>> --- a/mm/damon/paddr.c
>> +++ b/mm/damon/paddr.c
>> @@ -130,24 +130,21 @@ static bool damon_pa_young(unsigned long paddr, unsigned long *folio_sz)
>>   			accessed = false;
>>   		else
>>   			accessed = true;
>> -		folio_put(folio);
>>   		goto out;
> 
> Because you moved 'out' label to not include *folio_sz setting, folio_sz will
> not set in this case.  It should be set.
oh, it should be fixed.
> 
>>   	}
>>   
>>   	need_lock = !folio_test_anon(folio) || folio_test_ksm(folio);
>> -	if (need_lock && !folio_trylock(folio)) {
>> -		folio_put(folio);
>> -		return false;
>> -	}
>> +	if (need_lock && !folio_trylock(folio))
>> +		goto out;
>>   
>>   	rmap_walk(folio, &rwc);
>>   
>>   	if (need_lock)
>>   		folio_unlock(folio);
>> -	folio_put(folio);
>>   
>> -out:
>>   	*folio_sz = folio_size(folio);
>> +out:
>> +	folio_put(folio);
> 
> Before this change, folio_size() is called after folio_put().  Shouldn't it be
> called before folio_put()?  If so, could we make a separate fix for that first,
> and then make this change on top of it, so that it can be easily applied to
> relevant stable kernels?
> 
Yes， I could separate it, after folio_put(), the folio could be 
re-allocated and the folio_size calculation is not right.
> 
> Thanks,
> SJ
> 
>>   	return accessed;
>>   }
>>   
>> -- 
>> 2.35.3
>>
>>
Kefeng Wang March 6, 2023, 1:56 a.m. UTC | #3 
On 2023/3/6 9:10, Kefeng Wang wrote:
> 
> 
> On 2023/3/4 2:39, SeongJae Park wrote:
>> Hi Kefeng,
>>
>> On Fri, 3 Mar 2023 16:43:42 +0800 Kefeng Wang 
>> <wangkefeng.wang@huawei.com> wrote:
>>
>>> Omit three lines by unified folio_put(), and make code more clear.
>>>
>>> Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>
>>> ---
>>>   mm/damon/paddr.c | 11 ++++-------
>>>   1 file changed, 4 insertions(+), 7 deletions(-)
>>>
>>> diff --git a/mm/damon/paddr.c b/mm/damon/paddr.c
>>> index 3fda00a0f786..2ef9db0189ca 100644
>>> --- a/mm/damon/paddr.c
>>> +++ b/mm/damon/paddr.c
>>> @@ -130,24 +130,21 @@ static bool damon_pa_young(unsigned long paddr, 
>>> unsigned long *folio_sz)
>>>               accessed = false;
>>>           else
>>>               accessed = true;
>>> -        folio_put(folio);
>>>           goto out;
>>
>> Because you moved 'out' label to not include *folio_sz setting, 
>> folio_sz will
>> not set in this case.  It should be set.
> oh, it should be fixed.
>>
>>>       }
>>>       need_lock = !folio_test_anon(folio) || folio_test_ksm(folio);
>>> -    if (need_lock && !folio_trylock(folio)) {
>>> -        folio_put(folio);
>>> -        return false;
>>> -    }

Hi SJ,  apart from above issue, it looks that this branch need the 
folio_size() setting, right?

Thanks

>>> +    if (need_lock && !folio_trylock(folio))
>>> +        goto out;
>>>       rmap_walk(folio, &rwc);
>>>       if (need_lock)
>>>           folio_unlock(folio);
>>> -    folio_put(folio);
>>> -out:
>>>       *folio_sz = folio_size(folio);
>>> +out:
>>> +    folio_put(folio);
>>
>> Before this change, folio_size() is called after folio_put().  
>> Shouldn't it be
>> called before folio_put()?  If so, could we make a separate fix for 
>> that first,
>> and then make this change on top of it, so that it can be easily 
>> applied to
>> relevant stable kernels?
>>
> Yes， I could separate it, after folio_put(), the folio could be 
> re-allocated and the folio_size calculation is not right.
>>
>> Thanks,
>> SJ
>>
>>>       return accessed;
>>>   }
>>> -- 
>>> 2.35.3
>>>
>>>
SeongJae Park March 6, 2023, 9:27 p.m. UTC | #4 
Hi Kefeng,

On Mon, 6 Mar 2023 09:56:49 +0800 Kefeng Wang <wangkefeng.wang@huawei.com> wrote:

> 
> 
> On 2023/3/6 9:10, Kefeng Wang wrote:
> > 
> > 
> > On 2023/3/4 2:39, SeongJae Park wrote:
> >> Hi Kefeng,
> >>
> >> On Fri, 3 Mar 2023 16:43:42 +0800 Kefeng Wang 
> >> <wangkefeng.wang@huawei.com> wrote:
> >>
> >>> Omit three lines by unified folio_put(), and make code more clear.
> >>>
> >>> Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>
> >>> ---
> >>>   mm/damon/paddr.c | 11 ++++-------
> >>>   1 file changed, 4 insertions(+), 7 deletions(-)
> >>>
> >>> diff --git a/mm/damon/paddr.c b/mm/damon/paddr.c
> >>> index 3fda00a0f786..2ef9db0189ca 100644
> >>> --- a/mm/damon/paddr.c
> >>> +++ b/mm/damon/paddr.c
> >>> @@ -130,24 +130,21 @@ static bool damon_pa_young(unsigned long paddr, 
> >>> unsigned long *folio_sz)
> >>>               accessed = false;
> >>>           else
> >>>               accessed = true;
> >>> -        folio_put(folio);
> >>>           goto out;
> >>
> >> Because you moved 'out' label to not include *folio_sz setting, 
> >> folio_sz will
> >> not set in this case.  It should be set.
> > oh, it should be fixed.
> >>
> >>>       }
> >>>       need_lock = !folio_test_anon(folio) || folio_test_ksm(folio);
> >>> -    if (need_lock && !folio_trylock(folio)) {
> >>> -        folio_put(folio);
> >>> -        return false;
> >>> -    }
> 
> Hi SJ,  apart from above issue, it looks that this branch need the 
> folio_size() setting, right?

folio_sz is effectively used by caller of damon_pa_young() only if this
function returns true, so this branch doesn't need to set folio_sz.


Thanks,
SJ

> 
> Thanks
> 
> >>> +    if (need_lock && !folio_trylock(folio))
> >>> +        goto out;
> >>>       rmap_walk(folio, &rwc);
> >>>       if (need_lock)
> >>>           folio_unlock(folio);
> >>> -    folio_put(folio);
> >>> -out:
> >>>       *folio_sz = folio_size(folio);
> >>> +out:
> >>> +    folio_put(folio);
> >>
> >> Before this change, folio_size() is called after folio_put().  
> >> Shouldn't it be
> >> called before folio_put()?  If so, could we make a separate fix for 
> >> that first,
> >> and then make this change on top of it, so that it can be easily 
> >> applied to
> >> relevant stable kernels?
> >>
> > Yes， I could separate it, after folio_put(), the folio could be 
> > re-allocated and the folio_size calculation is not right.
> >>
> >> Thanks,
> >> SJ
> >>
> >>>       return accessed;
> >>>   }
> >>> -- 
> >>> 2.35.3
> >>>
> >>>
>
Kefeng Wang March 7, 2023, 1:22 a.m. UTC | #5 
On 2023/3/7 5:27, SeongJae Park wrote:
> Hi Kefeng,
> 
> On Mon, 6 Mar 2023 09:56:49 +0800 Kefeng Wang <wangkefeng.wang@huawei.com> wrote:
> 
>>
>>
>> On 2023/3/6 9:10, Kefeng Wang wrote:
>>>
>>>
>>> On 2023/3/4 2:39, SeongJae Park wrote:
>>>> Hi Kefeng,
>>>>
>>>> On Fri, 3 Mar 2023 16:43:42 +0800 Kefeng Wang
>>>> <wangkefeng.wang@huawei.com> wrote:
>>>>
>>>>> Omit three lines by unified folio_put(), and make code more clear.
>>>>>
>>>>> Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>
>>>>> ---
>>>>>    mm/damon/paddr.c | 11 ++++-------
>>>>>    1 file changed, 4 insertions(+), 7 deletions(-)
>>>>>
>>>>> diff --git a/mm/damon/paddr.c b/mm/damon/paddr.c
>>>>> index 3fda00a0f786..2ef9db0189ca 100644
>>>>> --- a/mm/damon/paddr.c
>>>>> +++ b/mm/damon/paddr.c
>>>>> @@ -130,24 +130,21 @@ static bool damon_pa_young(unsigned long paddr,
>>>>> unsigned long *folio_sz)
>>>>>                accessed = false;
>>>>>            else
>>>>>                accessed = true;
>>>>> -        folio_put(folio);
>>>>>            goto out;
>>>>
>>>> Because you moved 'out' label to not include *folio_sz setting,
>>>> folio_sz will
>>>> not set in this case.  It should be set.
>>> oh, it should be fixed.
>>>>
>>>>>        }
>>>>>        need_lock = !folio_test_anon(folio) || folio_test_ksm(folio);
>>>>> -    if (need_lock && !folio_trylock(folio)) {
>>>>> -        folio_put(folio);
>>>>> -        return false;
>>>>> -    }
>>
>> Hi SJ,  apart from above issue, it looks that this branch need the
>> folio_size() setting, right?
> 
> folio_sz is effectively used by caller of damon_pa_young() only if this
> function returns true, so this branch doesn't need to set folio_sz.

__damon_pa_check_access() store last_addr, last_accessed and 
last_folio_sz, even damon_pa_young() return false, the following check 
still use last_folio_sz,

   ALIGN_DOWN(last_addr, last_folio_sz) == ALIGN_DOWN(r->sampling_addr, 
last_folio_sz)

but last_folio_sz is not up to date, so I think it need to update, and 
update last_folio_sz is harmless, which could let's unify the return 
path, correct me if I am wrong.
SeongJae Park March 7, 2023, 6 p.m. UTC | #6 
On Tue, 7 Mar 2023 09:22:33 +0800 Kefeng Wang <wangkefeng.wang@huawei.com> wrote:

> 
> 
> On 2023/3/7 5:27, SeongJae Park wrote:
> > Hi Kefeng,
> > 
> > On Mon, 6 Mar 2023 09:56:49 +0800 Kefeng Wang <wangkefeng.wang@huawei.com> wrote:
> > 
> >>
> >>
> >> On 2023/3/6 9:10, Kefeng Wang wrote:
> >>>
> >>>
> >>> On 2023/3/4 2:39, SeongJae Park wrote:
> >>>> Hi Kefeng,
> >>>>
> >>>> On Fri, 3 Mar 2023 16:43:42 +0800 Kefeng Wang
> >>>> <wangkefeng.wang@huawei.com> wrote:
> >>>>
> >>>>> Omit three lines by unified folio_put(), and make code more clear.
> >>>>>
> >>>>> Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>
> >>>>> ---
> >>>>>    mm/damon/paddr.c | 11 ++++-------
> >>>>>    1 file changed, 4 insertions(+), 7 deletions(-)
> >>>>>
> >>>>> diff --git a/mm/damon/paddr.c b/mm/damon/paddr.c
> >>>>> index 3fda00a0f786..2ef9db0189ca 100644
> >>>>> --- a/mm/damon/paddr.c
> >>>>> +++ b/mm/damon/paddr.c
> >>>>> @@ -130,24 +130,21 @@ static bool damon_pa_young(unsigned long paddr,
> >>>>> unsigned long *folio_sz)
> >>>>>                accessed = false;
> >>>>>            else
> >>>>>                accessed = true;
> >>>>> -        folio_put(folio);
> >>>>>            goto out;
> >>>>
> >>>> Because you moved 'out' label to not include *folio_sz setting,
> >>>> folio_sz will
> >>>> not set in this case.  It should be set.
> >>> oh, it should be fixed.
> >>>>
> >>>>>        }
> >>>>>        need_lock = !folio_test_anon(folio) || folio_test_ksm(folio);
> >>>>> -    if (need_lock && !folio_trylock(folio)) {
> >>>>> -        folio_put(folio);
> >>>>> -        return false;
> >>>>> -    }
> >>
> >> Hi SJ,  apart from above issue, it looks that this branch need the
> >> folio_size() setting, right?
> > 
> > folio_sz is effectively used by caller of damon_pa_young() only if this
> > function returns true, so this branch doesn't need to set folio_sz.
> 
> __damon_pa_check_access() store last_addr, last_accessed and 
> last_folio_sz, even damon_pa_young() return false, the following check 
> still use last_folio_sz,
> 
>    ALIGN_DOWN(last_addr, last_folio_sz) == ALIGN_DOWN(r->sampling_addr, 
> last_folio_sz)
> 
> but last_folio_sz is not up to date, so I think it need to update, and 
> update last_folio_sz is harmless, which could let's unify the return 
> path, correct me if I am wrong.

Ah, you're right.  Thank you for kind explanation.  I was out of my mind for
some reason.  Maybe we could just do 'goto out' in the branch.


Thanks,
SJ
Kefeng Wang March 8, 2023, 1:03 a.m. UTC | #7 
On 2023/3/8 2:00, SeongJae Park wrote:
> On Tue, 7 Mar 2023 09:22:33 +0800 Kefeng Wang <wangkefeng.wang@huawei.com> wrote:
> 
>>
>>
>> On 2023/3/7 5:27, SeongJae Park wrote:
>>> Hi Kefeng,
>>>
>>> On Mon, 6 Mar 2023 09:56:49 +0800 Kefeng Wang <wangkefeng.wang@huawei.com> wrote:
>>>
>>>>
>>>>
>>>> On 2023/3/6 9:10, Kefeng Wang wrote:
>>>>>
>>>>>
>>>>> On 2023/3/4 2:39, SeongJae Park wrote:
>>>>>> Hi Kefeng,
>>>>>>
>>>>>> On Fri, 3 Mar 2023 16:43:42 +0800 Kefeng Wang
>>>>>> <wangkefeng.wang@huawei.com> wrote:
>>>>>>
>>>>>>> Omit three lines by unified folio_put(), and make code more clear.
>>>>>>>
>>>>>>> Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>
>>>>>>> ---
>>>>>>>     mm/damon/paddr.c | 11 ++++-------
>>>>>>>     1 file changed, 4 insertions(+), 7 deletions(-)
>>>>>>>
>>>>>>> diff --git a/mm/damon/paddr.c b/mm/damon/paddr.c
>>>>>>> index 3fda00a0f786..2ef9db0189ca 100644
>>>>>>> --- a/mm/damon/paddr.c
>>>>>>> +++ b/mm/damon/paddr.c
>>>>>>> @@ -130,24 +130,21 @@ static bool damon_pa_young(unsigned long paddr,
>>>>>>> unsigned long *folio_sz)
>>>>>>>                 accessed = false;
>>>>>>>             else
>>>>>>>                 accessed = true;
>>>>>>> -        folio_put(folio);
>>>>>>>             goto out;
>>>>>>
>>>>>> Because you moved 'out' label to not include *folio_sz setting,
>>>>>> folio_sz will
>>>>>> not set in this case.  It should be set.
>>>>> oh, it should be fixed.
>>>>>>
>>>>>>>         }
>>>>>>>         need_lock = !folio_test_anon(folio) || folio_test_ksm(folio);
>>>>>>> -    if (need_lock && !folio_trylock(folio)) {
>>>>>>> -        folio_put(folio);
>>>>>>> -        return false;
>>>>>>> -    }
>>>>
>>>> Hi SJ,  apart from above issue, it looks that this branch need the
>>>> folio_size() setting, right?
>>>
>>> folio_sz is effectively used by caller of damon_pa_young() only if this
>>> function returns true, so this branch doesn't need to set folio_sz.
>>
>> __damon_pa_check_access() store last_addr, last_accessed and
>> last_folio_sz, even damon_pa_young() return false, the following check
>> still use last_folio_sz,
>>
>>     ALIGN_DOWN(last_addr, last_folio_sz) == ALIGN_DOWN(r->sampling_addr,
>> last_folio_sz)
>>
>> but last_folio_sz is not up to date, so I think it need to update, and
>> update last_folio_sz is harmless, which could let's unify the return
>> path, correct me if I am wrong.
> 
> Ah, you're right.  Thank you for kind explanation.  I was out of my mind for
> some reason.  Maybe we could just do 'goto out' in the branch.

Yes, will update this patchset with this change.
> 
> 
> Thanks,
> SJ
diff mbox series

Patch

diff --git a/mm/damon/paddr.c b/mm/damon/paddr.c
index 3fda00a0f786..2ef9db0189ca 100644
--- a/mm/damon/paddr.c
+++ b/mm/damon/paddr.c
@@ -130,24 +130,21 @@  static bool damon_pa_young(unsigned long paddr, unsigned long *folio_sz)
 			accessed = false;
 		else
 			accessed = true;
-		folio_put(folio);
 		goto out;
 	}
 
 	need_lock = !folio_test_anon(folio) || folio_test_ksm(folio);
-	if (need_lock && !folio_trylock(folio)) {
-		folio_put(folio);
-		return false;
-	}
+	if (need_lock && !folio_trylock(folio))
+		goto out;
 
 	rmap_walk(folio, &rwc);
 
 	if (need_lock)
 		folio_unlock(folio);
-	folio_put(folio);
 
-out:
 	*folio_sz = folio_size(folio);
+out:
+	folio_put(folio);
 	return accessed;
 }