f2fs: Some optimizations for background GC in Zoned UFS.

[PATCH 2/2] f2fs: Enhance the subsequent logic of valid_thresh_ratio to prevent unnecessary background GC

Posted by Liao Yuanhong 5 months ago

When the proportion of dirty segments within a section exceeds the
valid_thresh_ratio, the gc_cost of that section is set to UINT_MAX,
indicating that these sections should not be released. However, if all
section costs within the scanning range of get_victim() are UINT_MAX,
background GC will still occur. Add a condition to prevent this situation.

Signed-off-by: Liao Yuanhong <liaoyuanhong@vivo.com>
---
 fs/f2fs/gc.c | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/fs/f2fs/gc.c b/fs/f2fs/gc.c
index 4a8c08f970e3..ffc3188416f4 100644
--- a/fs/f2fs/gc.c
+++ b/fs/f2fs/gc.c
@@ -936,6 +936,11 @@ int f2fs_get_victim(struct f2fs_sb_info *sbi, unsigned int *result,
 		}
 	}
 
+	if (f2fs_sb_has_blkzoned(sbi) && p.min_cost == UINT_MAX) {
+		ret = -ENODATA;
+		goto out;
+	}
+
 	/* get victim for GC_AT/AT_SSR */
 	if (is_atgc) {
 		lookup_victim_by_age(sbi, &p);
-- 
2.34.1

Re: [PATCH 2/2] f2fs: Enhance the subsequent logic of valid_thresh_ratio to prevent unnecessary background GC

Posted by Chao Yu 4 months, 4 weeks ago

On 9/9/25 21:44, Liao Yuanhong wrote:
> When the proportion of dirty segments within a section exceeds the
> valid_thresh_ratio, the gc_cost of that section is set to UINT_MAX,
> indicating that these sections should not be released. However, if all
> section costs within the scanning range of get_victim() are UINT_MAX,
> background GC will still occur. Add a condition to prevent this situation.

For this case, f2fs_get_victim() will return 0, and f2fs_gc() will use unchanged
segno for GC?

Thanks,

> 
> Signed-off-by: Liao Yuanhong <liaoyuanhong@vivo.com>
> ---
>  fs/f2fs/gc.c | 5 +++++
>  1 file changed, 5 insertions(+)
> 
> diff --git a/fs/f2fs/gc.c b/fs/f2fs/gc.c
> index 4a8c08f970e3..ffc3188416f4 100644
> --- a/fs/f2fs/gc.c
> +++ b/fs/f2fs/gc.c
> @@ -936,6 +936,11 @@ int f2fs_get_victim(struct f2fs_sb_info *sbi, unsigned int *result,
>  		}
>  	}
>  
> +	if (f2fs_sb_has_blkzoned(sbi) && p.min_cost == UINT_MAX) {
> +		ret = -ENODATA;
> +		goto out;
> +	}
> +
>  	/* get victim for GC_AT/AT_SSR */
>  	if (is_atgc) {
>  		lookup_victim_by_age(sbi, &p);

Re: [PATCH 2/2] f2fs: Enhance the subsequent logic of valid_thresh_ratio to prevent unnecessary background GC

Posted by Liao Yuanhong 4 months, 3 weeks ago

On 9/15/2025 4:36 PM, Chao Yu wrote:
> On 9/9/25 21:44, Liao Yuanhong wrote:
>> When the proportion of dirty segments within a section exceeds the
>> valid_thresh_ratio, the gc_cost of that section is set to UINT_MAX,
>> indicating that these sections should not be released. However, if all
>> section costs within the scanning range of get_victim() are UINT_MAX,
>> background GC will still occur. Add a condition to prevent this situation.
> For this case, f2fs_get_victim() will return 0, and f2fs_gc() will use unchanged
> segno for GC?
>
> Thanks,

You're right, segno won't update in this scenario, and this patch 
feature is redundant.


Thanks,

Liao

>> Signed-off-by: Liao Yuanhong <liaoyuanhong@vivo.com>
>> ---
>>   fs/f2fs/gc.c | 5 +++++
>>   1 file changed, 5 insertions(+)
>>
>> diff --git a/fs/f2fs/gc.c b/fs/f2fs/gc.c
>> index 4a8c08f970e3..ffc3188416f4 100644
>> --- a/fs/f2fs/gc.c
>> +++ b/fs/f2fs/gc.c
>> @@ -936,6 +936,11 @@ int f2fs_get_victim(struct f2fs_sb_info *sbi, unsigned int *result,
>>   		}
>>   	}
>>   
>> +	if (f2fs_sb_has_blkzoned(sbi) && p.min_cost == UINT_MAX) {
>> +		ret = -ENODATA;
>> +		goto out;
>> +	}
>> +
>>   	/* get victim for GC_AT/AT_SSR */
>>   	if (is_atgc) {
>>   		lookup_victim_by_age(sbi, &p);

Re: [PATCH 2/2] f2fs: Enhance the subsequent logic of valid_thresh_ratio to prevent unnecessary background GC

Posted by Chao Yu 4 months, 3 weeks ago

On 9/17/25 15:08, Liao Yuanhong wrote:
> 
> On 9/15/2025 4:36 PM, Chao Yu wrote:
>> On 9/9/25 21:44, Liao Yuanhong wrote:
>>> When the proportion of dirty segments within a section exceeds the
>>> valid_thresh_ratio, the gc_cost of that section is set to UINT_MAX,
>>> indicating that these sections should not be released. However, if all
>>> section costs within the scanning range of get_victim() are UINT_MAX,
>>> background GC will still occur. Add a condition to prevent this situation.
>> For this case, f2fs_get_victim() will return 0, and f2fs_gc() will use unchanged
>> segno for GC?
>>
>> Thanks,
> 
> You're right, segno won't update in this scenario, and this patch feature is redundant.

Oh, I meant, if f2fs_get_victim() fails to select a valid victim due to the reason you
described, f2fs_get_victim() will return 0, and f2fs_gc() will migrate segment #NULL_SEGNO?
Or am I missing something?

Thanks,

> 
> 
> Thanks,
> 
> Liao
> 
>>> Signed-off-by: Liao Yuanhong <liaoyuanhong@vivo.com>
>>> ---
>>>   fs/f2fs/gc.c | 5 +++++
>>>   1 file changed, 5 insertions(+)
>>>
>>> diff --git a/fs/f2fs/gc.c b/fs/f2fs/gc.c
>>> index 4a8c08f970e3..ffc3188416f4 100644
>>> --- a/fs/f2fs/gc.c
>>> +++ b/fs/f2fs/gc.c
>>> @@ -936,6 +936,11 @@ int f2fs_get_victim(struct f2fs_sb_info *sbi, unsigned int *result,
>>>           }
>>>       }
>>>   +    if (f2fs_sb_has_blkzoned(sbi) && p.min_cost == UINT_MAX) {
>>> +        ret = -ENODATA;
>>> +        goto out;
>>> +    }
>>> +
>>>       /* get victim for GC_AT/AT_SSR */
>>>       if (is_atgc) {
>>>           lookup_victim_by_age(sbi, &p);

Re: [PATCH 2/2] f2fs: Enhance the subsequent logic of valid_thresh_ratio to prevent unnecessary background GC

Posted by Liao Yuanhong 4 months, 3 weeks ago

On 9/17/2025 3:57 PM, Chao Yu wrote:
> On 9/17/25 15:08, Liao Yuanhong wrote:
>> On 9/15/2025 4:36 PM, Chao Yu wrote:
>>> On 9/9/25 21:44, Liao Yuanhong wrote:
>>>> When the proportion of dirty segments within a section exceeds the
>>>> valid_thresh_ratio, the gc_cost of that section is set to UINT_MAX,
>>>> indicating that these sections should not be released. However, if all
>>>> section costs within the scanning range of get_victim() are UINT_MAX,
>>>> background GC will still occur. Add a condition to prevent this situation.
>>> For this case, f2fs_get_victim() will return 0, and f2fs_gc() will use unchanged
>>> segno for GC?
>>>
>>> Thanks,
>> You're right, segno won't update in this scenario, and this patch feature is redundant.
> Oh, I meant, if f2fs_get_victim() fails to select a valid victim due to the reason you
> described, f2fs_get_victim() will return 0, and f2fs_gc() will migrate segment #NULL_SEGNO?
> Or am I missing something?
>
> Thanks,

Yes. In this scenario, since it won't enter the|p.min_cost > 
cost|condition,|p.min_segno|will retain its initial value|||NULL_SEGNO|. 
This is consistent with what you described.


Thanks,

Liao

>>
>> Thanks,
>>
>> Liao
>>
>>>> Signed-off-by: Liao Yuanhong <liaoyuanhong@vivo.com>
>>>> ---
>>>>    fs/f2fs/gc.c | 5 +++++
>>>>    1 file changed, 5 insertions(+)
>>>>
>>>> diff --git a/fs/f2fs/gc.c b/fs/f2fs/gc.c
>>>> index 4a8c08f970e3..ffc3188416f4 100644
>>>> --- a/fs/f2fs/gc.c
>>>> +++ b/fs/f2fs/gc.c
>>>> @@ -936,6 +936,11 @@ int f2fs_get_victim(struct f2fs_sb_info *sbi, unsigned int *result,
>>>>            }
>>>>        }
>>>>    +    if (f2fs_sb_has_blkzoned(sbi) && p.min_cost == UINT_MAX) {
>>>> +        ret = -ENODATA;
>>>> +        goto out;
>>>> +    }
>>>> +
>>>>        /* get victim for GC_AT/AT_SSR */
>>>>        if (is_atgc) {
>>>>            lookup_victim_by_age(sbi, &p);

Re: [PATCH 2/2] f2fs: Enhance the subsequent logic of valid_thresh_ratio to prevent unnecessary background GC

Posted by Chao Yu 4 months, 3 weeks ago

On 9/17/25 16:13, Liao Yuanhong wrote:
> 
> On 9/17/2025 3:57 PM, Chao Yu wrote:
>> On 9/17/25 15:08, Liao Yuanhong wrote:
>>> On 9/15/2025 4:36 PM, Chao Yu wrote:
>>>> On 9/9/25 21:44, Liao Yuanhong wrote:
>>>>> When the proportion of dirty segments within a section exceeds the
>>>>> valid_thresh_ratio, the gc_cost of that section is set to UINT_MAX,
>>>>> indicating that these sections should not be released. However, if all
>>>>> section costs within the scanning range of get_victim() are UINT_MAX,
>>>>> background GC will still occur. Add a condition to prevent this situation.
>>>> For this case, f2fs_get_victim() will return 0, and f2fs_gc() will use unchanged
>>>> segno for GC?
>>>>
>>>> Thanks,
>>> You're right, segno won't update in this scenario, and this patch feature is redundant.
>> Oh, I meant, if f2fs_get_victim() fails to select a valid victim due to the reason you
>> described, f2fs_get_victim() will return 0, and f2fs_gc() will migrate segment #NULL_SEGNO?
>> Or am I missing something?
>>
>> Thanks,
> 
> Yes. In this scenario, since it won't enter the|p.min_cost > cost|condition,|p.min_segno|will retain its initial value|||NULL_SEGNO|. This is consistent with what you described.

Do you have a script to reproduce this bug?

Thanks,

> 
> 
> Thanks,
> 
> Liao
> 
>>>
>>> Thanks,
>>>
>>> Liao
>>>
>>>>> Signed-off-by: Liao Yuanhong <liaoyuanhong@vivo.com>
>>>>> ---
>>>>>    fs/f2fs/gc.c | 5 +++++
>>>>>    1 file changed, 5 insertions(+)
>>>>>
>>>>> diff --git a/fs/f2fs/gc.c b/fs/f2fs/gc.c
>>>>> index 4a8c08f970e3..ffc3188416f4 100644
>>>>> --- a/fs/f2fs/gc.c
>>>>> +++ b/fs/f2fs/gc.c
>>>>> @@ -936,6 +936,11 @@ int f2fs_get_victim(struct f2fs_sb_info *sbi, unsigned int *result,
>>>>>            }
>>>>>        }
>>>>>    +    if (f2fs_sb_has_blkzoned(sbi) && p.min_cost == UINT_MAX) {
>>>>> +        ret = -ENODATA;
>>>>> +        goto out;
>>>>> +    }
>>>>> +
>>>>>        /* get victim for GC_AT/AT_SSR */
>>>>>        if (is_atgc) {
>>>>>            lookup_victim_by_age(sbi, &p);

Re: [PATCH 2/2] f2fs: Enhance the subsequent logic of valid_thresh_ratio to prevent unnecessary background GC

Posted by Liao Yuanhong 4 months, 3 weeks ago

On 9/18/2025 10:16 AM, Chao Yu wrote:
> On 9/17/25 16:13, Liao Yuanhong wrote:
>> On 9/17/2025 3:57 PM, Chao Yu wrote:
>>> On 9/17/25 15:08, Liao Yuanhong wrote:
>>>> On 9/15/2025 4:36 PM, Chao Yu wrote:
>>>>> On 9/9/25 21:44, Liao Yuanhong wrote:
>>>>>> When the proportion of dirty segments within a section exceeds the
>>>>>> valid_thresh_ratio, the gc_cost of that section is set to UINT_MAX,
>>>>>> indicating that these sections should not be released. However, if all
>>>>>> section costs within the scanning range of get_victim() are UINT_MAX,
>>>>>> background GC will still occur. Add a condition to prevent this situation.
>>>>> For this case, f2fs_get_victim() will return 0, and f2fs_gc() will use unchanged
>>>>> segno for GC?
>>>>>
>>>>> Thanks,
>>>> You're right, segno won't update in this scenario, and this patch feature is redundant.
>>> Oh, I meant, if f2fs_get_victim() fails to select a valid victim due to the reason you
>>> described, f2fs_get_victim() will return 0, and f2fs_gc() will migrate segment #NULL_SEGNO?
>>> Or am I missing something?
>>>
>>> Thanks,
>> Yes. In this scenario, since it won't enter the|p.min_cost > cost|condition,|p.min_segno|will retain its initial value|||NULL_SEGNO|. This is consistent with what you described.
> Do you have a script to reproduce this bug?
>
> Thanks,

I didn't explain it clearly. What I mean is that this patch is 
redundant, the original code is fine.


Thanks,

Liao

>>
>> Thanks,
>>
>> Liao
>>
>>>> Thanks,
>>>>
>>>> Liao
>>>>
>>>>>> Signed-off-by: Liao Yuanhong <liaoyuanhong@vivo.com>
>>>>>> ---
>>>>>>     fs/f2fs/gc.c | 5 +++++
>>>>>>     1 file changed, 5 insertions(+)
>>>>>>
>>>>>> diff --git a/fs/f2fs/gc.c b/fs/f2fs/gc.c
>>>>>> index 4a8c08f970e3..ffc3188416f4 100644
>>>>>> --- a/fs/f2fs/gc.c
>>>>>> +++ b/fs/f2fs/gc.c
>>>>>> @@ -936,6 +936,11 @@ int f2fs_get_victim(struct f2fs_sb_info *sbi, unsigned int *result,
>>>>>>             }
>>>>>>         }
>>>>>>     +    if (f2fs_sb_has_blkzoned(sbi) && p.min_cost == UINT_MAX) {
>>>>>> +        ret = -ENODATA;
>>>>>> +        goto out;
>>>>>> +    }
>>>>>> +
>>>>>>         /* get victim for GC_AT/AT_SSR */
>>>>>>         if (is_atgc) {
>>>>>>             lookup_victim_by_age(sbi, &p);

Re: [PATCH 2/2] f2fs: Enhance the subsequent logic of valid_thresh_ratio to prevent unnecessary background GC

Posted by Chao Yu 4 months, 2 weeks ago

On 9/20/25 10:49, Liao Yuanhong wrote:
> 
> On 9/18/2025 10:16 AM, Chao Yu wrote:
>> On 9/17/25 16:13, Liao Yuanhong wrote:
>>> On 9/17/2025 3:57 PM, Chao Yu wrote:
>>>> On 9/17/25 15:08, Liao Yuanhong wrote:
>>>>> On 9/15/2025 4:36 PM, Chao Yu wrote:
>>>>>> On 9/9/25 21:44, Liao Yuanhong wrote:
>>>>>>> When the proportion of dirty segments within a section exceeds the
>>>>>>> valid_thresh_ratio, the gc_cost of that section is set to UINT_MAX,
>>>>>>> indicating that these sections should not be released. However, if all
>>>>>>> section costs within the scanning range of get_victim() are UINT_MAX,
>>>>>>> background GC will still occur. Add a condition to prevent this situation.
>>>>>> For this case, f2fs_get_victim() will return 0, and f2fs_gc() will use unchanged
>>>>>> segno for GC?
>>>>>>
>>>>>> Thanks,
>>>>> You're right, segno won't update in this scenario, and this patch feature is redundant.
>>>> Oh, I meant, if f2fs_get_victim() fails to select a valid victim due to the reason you
>>>> described, f2fs_get_victim() will return 0, and f2fs_gc() will migrate segment #NULL_SEGNO?
>>>> Or am I missing something?
>>>>
>>>> Thanks,
>>> Yes. In this scenario, since it won't enter the|p.min_cost > cost|condition,|p.min_segno|will retain its initial value|||NULL_SEGNO|. This is consistent with what you described.
>> Do you have a script to reproduce this bug?
>>
>> Thanks,
> 
> I didn't explain it clearly. What I mean is that this patch is redundant, the original code is fine.

Oh, I see. I don't see any problem in f2fs_get_victim() as well.

Thanks,

> 
> 
> Thanks,
> 
> Liao
> 
>>>
>>> Thanks,
>>>
>>> Liao
>>>
>>>>> Thanks,
>>>>>
>>>>> Liao
>>>>>
>>>>>>> Signed-off-by: Liao Yuanhong <liaoyuanhong@vivo.com>
>>>>>>> ---
>>>>>>>     fs/f2fs/gc.c | 5 +++++
>>>>>>>     1 file changed, 5 insertions(+)
>>>>>>>
>>>>>>> diff --git a/fs/f2fs/gc.c b/fs/f2fs/gc.c
>>>>>>> index 4a8c08f970e3..ffc3188416f4 100644
>>>>>>> --- a/fs/f2fs/gc.c
>>>>>>> +++ b/fs/f2fs/gc.c
>>>>>>> @@ -936,6 +936,11 @@ int f2fs_get_victim(struct f2fs_sb_info *sbi, unsigned int *result,
>>>>>>>             }
>>>>>>>         }
>>>>>>>     +    if (f2fs_sb_has_blkzoned(sbi) && p.min_cost == UINT_MAX) {
>>>>>>> +        ret = -ENODATA;
>>>>>>> +        goto out;
>>>>>>> +    }
>>>>>>> +
>>>>>>>         /* get victim for GC_AT/AT_SSR */
>>>>>>>         if (is_atgc) {
>>>>>>>             lookup_victim_by_age(sbi, &p);