f2fs: allocate HOT_DATA for IPU writes

[PATCH] f2fs: allocate HOT_DATA for IPU writes

Posted by Jaegeuk Kim 1 month ago

Let's split IPU writes in hot data area to improve the GC efficiency.

Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
---
 fs/f2fs/segment.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/fs/f2fs/segment.c b/fs/f2fs/segment.c
index e0f6589c6a1c..bb79dad55ac9 100644
--- a/fs/f2fs/segment.c
+++ b/fs/f2fs/segment.c
@@ -3688,7 +3688,8 @@ static int __get_segment_type_6(struct f2fs_io_info *fio)
 
 		if (file_is_hot(inode) ||
 				is_inode_flag_set(inode, FI_HOT_DATA) ||
-				f2fs_is_cow_file(inode))
+				f2fs_is_cow_file(inode) ||
+				is_inode_flag_set(inode, FI_NEED_IPU))
 			return CURSEG_HOT_DATA;
 		return f2fs_rw_hint_to_seg_type(F2FS_I_SB(inode),
 						inode->i_write_hint);
-- 
2.51.0.318.gd7df087d1a-goog

Re: [f2fs-dev] [PATCH] f2fs: allocate HOT_DATA for IPU writes

Posted by Chao Yu 1 month ago

On 8/28/25 05:52, Jaegeuk Kim via Linux-f2fs-devel wrote:
> Let's split IPU writes in hot data area to improve the GC efficiency.

I didn't get it, IPU may be not friendly for migrating data along w/ write in
hot area?

Thanks,

> 
> Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
> ---
>  fs/f2fs/segment.c | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/fs/f2fs/segment.c b/fs/f2fs/segment.c
> index e0f6589c6a1c..bb79dad55ac9 100644
> --- a/fs/f2fs/segment.c
> +++ b/fs/f2fs/segment.c
> @@ -3688,7 +3688,8 @@ static int __get_segment_type_6(struct f2fs_io_info *fio)
>  
>  		if (file_is_hot(inode) ||
>  				is_inode_flag_set(inode, FI_HOT_DATA) ||
> -				f2fs_is_cow_file(inode))
> +				f2fs_is_cow_file(inode) ||
> +				is_inode_flag_set(inode, FI_NEED_IPU))
>  			return CURSEG_HOT_DATA;
>  		return f2fs_rw_hint_to_seg_type(F2FS_I_SB(inode),
>  						inode->i_write_hint);

Re: [f2fs-dev] [PATCH] f2fs: allocate HOT_DATA for IPU writes

Posted by Jaegeuk Kim 1 month ago

On 08/28, Chao Yu wrote:
> On 8/28/25 05:52, Jaegeuk Kim via Linux-f2fs-devel wrote:
> > Let's split IPU writes in hot data area to improve the GC efficiency.
> 
> I didn't get it, IPU may be not friendly for migrating data along w/ write in
> hot area?

This deals with IPU in the LFS mode.

> 
> Thanks,
> 
> > 
> > Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
> > ---
> >  fs/f2fs/segment.c | 3 ++-
> >  1 file changed, 2 insertions(+), 1 deletion(-)
> > 
> > diff --git a/fs/f2fs/segment.c b/fs/f2fs/segment.c
> > index e0f6589c6a1c..bb79dad55ac9 100644
> > --- a/fs/f2fs/segment.c
> > +++ b/fs/f2fs/segment.c
> > @@ -3688,7 +3688,8 @@ static int __get_segment_type_6(struct f2fs_io_info *fio)
> >  
> >  		if (file_is_hot(inode) ||
> >  				is_inode_flag_set(inode, FI_HOT_DATA) ||
> > -				f2fs_is_cow_file(inode))
> > +				f2fs_is_cow_file(inode) ||
> > +				is_inode_flag_set(inode, FI_NEED_IPU))
> >  			return CURSEG_HOT_DATA;
> >  		return f2fs_rw_hint_to_seg_type(F2FS_I_SB(inode),
> >  						inode->i_write_hint);

Re: [f2fs-dev] [PATCH] f2fs: allocate HOT_DATA for IPU writes

Posted by Chao Yu 1 month ago

On 8/28/2025 11:23 PM, Jaegeuk Kim wrote:
> On 08/28, Chao Yu wrote:
>> On 8/28/25 05:52, Jaegeuk Kim via Linux-f2fs-devel wrote:
>>> Let's split IPU writes in hot data area to improve the GC efficiency.
>>
>> I didn't get it, IPU may be not friendly for migrating data along w/ write in
>> hot area?
> 
> This deals with IPU in the LFS mode.

Okay, so, we missed to add f2fs_lfs_mode() condition, right?

Thanks,

> 
>>
>> Thanks,
>>
>>>
>>> Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
>>> ---
>>>   fs/f2fs/segment.c | 3 ++-
>>>   1 file changed, 2 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/fs/f2fs/segment.c b/fs/f2fs/segment.c
>>> index e0f6589c6a1c..bb79dad55ac9 100644
>>> --- a/fs/f2fs/segment.c
>>> +++ b/fs/f2fs/segment.c
>>> @@ -3688,7 +3688,8 @@ static int __get_segment_type_6(struct f2fs_io_info *fio)
>>>   
>>>   		if (file_is_hot(inode) ||
>>>   				is_inode_flag_set(inode, FI_HOT_DATA) ||
>>> -				f2fs_is_cow_file(inode))
>>> +				f2fs_is_cow_file(inode) ||
>>> +				is_inode_flag_set(inode, FI_NEED_IPU))
>>>   			return CURSEG_HOT_DATA;
>>>   		return f2fs_rw_hint_to_seg_type(F2FS_I_SB(inode),
>>>   						inode->i_write_hint);

Re: [f2fs-dev] [PATCH] f2fs: allocate HOT_DATA for IPU writes

Posted by Jaegeuk Kim 1 month ago

On 08/29, Chao Yu wrote:
> On 8/28/2025 11:23 PM, Jaegeuk Kim wrote:
> > On 08/28, Chao Yu wrote:
> > > On 8/28/25 05:52, Jaegeuk Kim via Linux-f2fs-devel wrote:
> > > > Let's split IPU writes in hot data area to improve the GC efficiency.
> > > 
> > > I didn't get it, IPU may be not friendly for migrating data along w/ write in
> > > hot area?
> > 
> > This deals with IPU in the LFS mode.
> 
> Okay, so, we missed to add f2fs_lfs_mode() condition, right?

Why do we need that?

> 
> Thanks,
> 
> > 
> > > 
> > > Thanks,
> > > 
> > > > 
> > > > Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
> > > > ---
> > > >   fs/f2fs/segment.c | 3 ++-
> > > >   1 file changed, 2 insertions(+), 1 deletion(-)
> > > > 
> > > > diff --git a/fs/f2fs/segment.c b/fs/f2fs/segment.c
> > > > index e0f6589c6a1c..bb79dad55ac9 100644
> > > > --- a/fs/f2fs/segment.c
> > > > +++ b/fs/f2fs/segment.c
> > > > @@ -3688,7 +3688,8 @@ static int __get_segment_type_6(struct f2fs_io_info *fio)
> > > >   		if (file_is_hot(inode) ||
> > > >   				is_inode_flag_set(inode, FI_HOT_DATA) ||
> > > > -				f2fs_is_cow_file(inode))
> > > > +				f2fs_is_cow_file(inode) ||
> > > > +				is_inode_flag_set(inode, FI_NEED_IPU))
> > > >   			return CURSEG_HOT_DATA;
> > > >   		return f2fs_rw_hint_to_seg_type(F2FS_I_SB(inode),
> > > >   						inode->i_write_hint);

Re: [f2fs-dev] [PATCH] f2fs: allocate HOT_DATA for IPU writes

Posted by Chao Yu 1 month ago

On 8/30/2025 4:47 AM, Jaegeuk Kim wrote:
> On 08/29, Chao Yu wrote:
>> On 8/28/2025 11:23 PM, Jaegeuk Kim wrote:
>>> On 08/28, Chao Yu wrote:
>>>> On 8/28/25 05:52, Jaegeuk Kim via Linux-f2fs-devel wrote:
>>>>> Let's split IPU writes in hot data area to improve the GC efficiency.
>>>>
>>>> I didn't get it, IPU may be not friendly for migrating data along w/ write in
>>>> hot area?
>>>
>>> This deals with IPU in the LFS mode.
>>
>> Okay, so, we missed to add f2fs_lfs_mode() condition, right?
> 
> Why do we need that?

As you mentioned above, this deals w/ IPU in the LFS mode. Am I missing
something?

Thanks,

> 
>>
>> Thanks,
>>
>>>
>>>>
>>>> Thanks,
>>>>
>>>>>
>>>>> Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
>>>>> ---
>>>>>    fs/f2fs/segment.c | 3 ++-
>>>>>    1 file changed, 2 insertions(+), 1 deletion(-)
>>>>>
>>>>> diff --git a/fs/f2fs/segment.c b/fs/f2fs/segment.c
>>>>> index e0f6589c6a1c..bb79dad55ac9 100644
>>>>> --- a/fs/f2fs/segment.c
>>>>> +++ b/fs/f2fs/segment.c
>>>>> @@ -3688,7 +3688,8 @@ static int __get_segment_type_6(struct f2fs_io_info *fio)
>>>>>    		if (file_is_hot(inode) ||
>>>>>    				is_inode_flag_set(inode, FI_HOT_DATA) ||
>>>>> -				f2fs_is_cow_file(inode))
>>>>> +				f2fs_is_cow_file(inode) ||
>>>>> +				is_inode_flag_set(inode, FI_NEED_IPU))
>>>>>    			return CURSEG_HOT_DATA;
>>>>>    		return f2fs_rw_hint_to_seg_type(F2FS_I_SB(inode),
>>>>>    						inode->i_write_hint);

Re: [f2fs-dev] [PATCH] f2fs: allocate HOT_DATA for IPU writes

Posted by Jaegeuk Kim 1 month ago

On 08/30, Chao Yu wrote:
> On 8/30/2025 4:47 AM, Jaegeuk Kim wrote:
> > On 08/29, Chao Yu wrote:
> > > On 8/28/2025 11:23 PM, Jaegeuk Kim wrote:
> > > > On 08/28, Chao Yu wrote:
> > > > > On 8/28/25 05:52, Jaegeuk Kim via Linux-f2fs-devel wrote:
> > > > > > Let's split IPU writes in hot data area to improve the GC efficiency.
> > > > > 
> > > > > I didn't get it, IPU may be not friendly for migrating data along w/ write in
> > > > > hot area?
> > > > 
> > > > This deals with IPU in the LFS mode.
> > > 
> > > Okay, so, we missed to add f2fs_lfs_mode() condition, right?
> > 
> > Why do we need that?
> 
> As you mentioned above, this deals w/ IPU in the LFS mode. Am I missing
> something?

The IPU flag indicates the data is hot. In lfs_mode, we missed to allocate
it into HOT_DATA. In normal mode, IPU will update the data in place, which
doesn't affect anything, while newly create data is also good to store in
HOT_DATA as well. So, I think we don't need other condition here.

> 
> Thanks,
> 
> > 
> > > 
> > > Thanks,
> > > 
> > > > 
> > > > > 
> > > > > Thanks,
> > > > > 
> > > > > > 
> > > > > > Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
> > > > > > ---
> > > > > >    fs/f2fs/segment.c | 3 ++-
> > > > > >    1 file changed, 2 insertions(+), 1 deletion(-)
> > > > > > 
> > > > > > diff --git a/fs/f2fs/segment.c b/fs/f2fs/segment.c
> > > > > > index e0f6589c6a1c..bb79dad55ac9 100644
> > > > > > --- a/fs/f2fs/segment.c
> > > > > > +++ b/fs/f2fs/segment.c
> > > > > > @@ -3688,7 +3688,8 @@ static int __get_segment_type_6(struct f2fs_io_info *fio)
> > > > > >    		if (file_is_hot(inode) ||
> > > > > >    				is_inode_flag_set(inode, FI_HOT_DATA) ||
> > > > > > -				f2fs_is_cow_file(inode))
> > > > > > +				f2fs_is_cow_file(inode) ||
> > > > > > +				is_inode_flag_set(inode, FI_NEED_IPU))
> > > > > >    			return CURSEG_HOT_DATA;
> > > > > >    		return f2fs_rw_hint_to_seg_type(F2FS_I_SB(inode),
> > > > > >    						inode->i_write_hint);

Re: [f2fs-dev] [PATCH] f2fs: allocate HOT_DATA for IPU writes

Posted by Chao Yu 1 month ago

On 8/31/25 07:54, Jaegeuk Kim wrote:
> On 08/30, Chao Yu wrote:
>> On 8/30/2025 4:47 AM, Jaegeuk Kim wrote:
>>> On 08/29, Chao Yu wrote:
>>>> On 8/28/2025 11:23 PM, Jaegeuk Kim wrote:
>>>>> On 08/28, Chao Yu wrote:
>>>>>> On 8/28/25 05:52, Jaegeuk Kim via Linux-f2fs-devel wrote:
>>>>>>> Let's split IPU writes in hot data area to improve the GC efficiency.
>>>>>>
>>>>>> I didn't get it, IPU may be not friendly for migrating data along w/ write in
>>>>>> hot area?
>>>>>
>>>>> This deals with IPU in the LFS mode.
>>>>
>>>> Okay, so, we missed to add f2fs_lfs_mode() condition, right?
>>>
>>> Why do we need that?
>>
>> As you mentioned above, this deals w/ IPU in the LFS mode. Am I missing
>> something?
> 
> The IPU flag indicates the data is hot. In lfs_mode, we missed to allocate
> it into HOT_DATA. In normal mode, IPU will update the data in place, which
> doesn't affect anything, while newly create data is also good to store in
> HOT_DATA as well. So, I think we don't need other condition here.

It makes sense, thanks for the explanation.

Reviewed-by: Chao Yu <chao@kernel.org>

Thanks,

> 
>>
>> Thanks,
>>
>>>
>>>>
>>>> Thanks,
>>>>
>>>>>
>>>>>>
>>>>>> Thanks,
>>>>>>
>>>>>>>
>>>>>>> Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
>>>>>>> ---
>>>>>>>    fs/f2fs/segment.c | 3 ++-
>>>>>>>    1 file changed, 2 insertions(+), 1 deletion(-)
>>>>>>>
>>>>>>> diff --git a/fs/f2fs/segment.c b/fs/f2fs/segment.c
>>>>>>> index e0f6589c6a1c..bb79dad55ac9 100644
>>>>>>> --- a/fs/f2fs/segment.c
>>>>>>> +++ b/fs/f2fs/segment.c
>>>>>>> @@ -3688,7 +3688,8 @@ static int __get_segment_type_6(struct f2fs_io_info *fio)
>>>>>>>    		if (file_is_hot(inode) ||
>>>>>>>    				is_inode_flag_set(inode, FI_HOT_DATA) ||
>>>>>>> -				f2fs_is_cow_file(inode))
>>>>>>> +				f2fs_is_cow_file(inode) ||
>>>>>>> +				is_inode_flag_set(inode, FI_NEED_IPU))
>>>>>>>    			return CURSEG_HOT_DATA;
>>>>>>>    		return f2fs_rw_hint_to_seg_type(F2FS_I_SB(inode),
>>>>>>>    						inode->i_write_hint);