[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <9c5db535-f10e-4743-a8c7-e74f2680f2d0@vivo.com>
Date: Wed, 17 Sep 2025 15:15:38 +0800
From: Liao Yuanhong <liaoyuanhong@...o.com>
To: Jaegeuk Kim <jaegeuk@...nel.org>
Cc: Chao Yu <chao@...nel.org>,
"open list:F2FS FILE SYSTEM" <linux-f2fs-devel@...ts.sourceforge.net>,
open list <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH 1/2] f2fs: Optimize excessive write operations caused by
continuous background garbage collection in Zoned UFS
On 9/16/2025 10:28 AM, Jaegeuk Kim wrote:
> Could you please share some trends of relation between has_enough_free_blocks()
> vs. has_enough_dirty_blocks()? I'm wondering whethere there's a missing case
> where has_enough_free_blocks() is not enough.
Sure. I will find some time to test the data and create a table to see
if there are any omissions.
Thanks,
Liao
> On 09/09, Liao Yuanhong wrote:
>> Incorporate a check using has_enough_dirty_blocks() to prevent redundant
>> background GC in Zoned UFS. When there are insufficient dirty segments,
>> continuous execution of background GC should be avoided, as it results in
>> unnecessary write operations and impacts device lifespan. The initial
>> threshold is set to 3 * section size (since f2fs data uses three write
>> pointers).
>>
>> Signed-off-by: Liao Yuanhong <liaoyuanhong@...o.com>
>> ---
>> fs/f2fs/gc.c | 8 ++++++--
>> fs/f2fs/gc.h | 10 +++++++++-
>> 2 files changed, 15 insertions(+), 3 deletions(-)
>>
>> diff --git a/fs/f2fs/gc.c b/fs/f2fs/gc.c
>> index ed3acbfc83ca..4a8c08f970e3 100644
>> --- a/fs/f2fs/gc.c
>> +++ b/fs/f2fs/gc.c
>> @@ -120,7 +120,9 @@ static int gc_thread_func(void *data)
>>
>> if (f2fs_sb_has_blkzoned(sbi)) {
>> if (has_enough_free_blocks(sbi,
>> - gc_th->no_zoned_gc_percent)) {
>> + gc_th->no_zoned_gc_percent) ||
>> + !has_enough_dirty_blocks(sbi,
>> + LIMIT_GC_DIRTY_SECTION_NUM)) {
>> wait_ms = gc_th->no_gc_sleep_time;
>> f2fs_up_write(&sbi->gc_lock);
>> goto next;
>> @@ -1750,7 +1752,9 @@ static int do_garbage_collect(struct f2fs_sb_info *sbi,
>>
>> if (f2fs_sb_has_blkzoned(sbi) &&
>> !has_enough_free_blocks(sbi,
>> - sbi->gc_thread->boost_zoned_gc_percent))
>> + sbi->gc_thread->boost_zoned_gc_percent) &&
>> + has_enough_dirty_blocks(sbi,
>> + LIMIT_GC_DIRTY_SECTION_NUM))
>> window_granularity *=
>> sbi->gc_thread->boost_gc_multiple;
>>
>> diff --git a/fs/f2fs/gc.h b/fs/f2fs/gc.h
>> index 24e8b1c27acc..1ef234c2702b 100644
>> --- a/fs/f2fs/gc.h
>> +++ b/fs/f2fs/gc.h
>> @@ -36,6 +36,7 @@
>> #define DEF_MIGRATION_WINDOW_GRANULARITY_ZONED 3
>> #define BOOST_GC_MULTIPLE 5
>> #define ZONED_PIN_SEC_REQUIRED_COUNT 1
>> +#define LIMIT_GC_DIRTY_SECTION_NUM 3
>>
>> #define DEF_GC_FAILED_PINNED_FILES 2048
>> #define MAX_GC_FAILED_PINNED_FILES USHRT_MAX
>> @@ -177,6 +178,12 @@ static inline bool has_enough_free_blocks(struct f2fs_sb_info *sbi,
>> return free_sections(sbi) > ((sbi->total_sections * limit_perc) / 100);
>> }
>>
>> +static inline bool has_enough_dirty_blocks(struct f2fs_sb_info *sbi,
>> + unsigned int limit_num)
>> +{
>> + return dirty_segments(sbi) > limit_num * SEGS_PER_SEC(sbi);
>> +}
>> +
>> static inline bool has_enough_invalid_blocks(struct f2fs_sb_info *sbi)
>> {
>> block_t user_block_count = sbi->user_block_count;
>> @@ -197,6 +204,7 @@ static inline bool need_to_boost_gc(struct f2fs_sb_info *sbi)
>> {
>> if (f2fs_sb_has_blkzoned(sbi))
>> return !has_enough_free_blocks(sbi,
>> - sbi->gc_thread->boost_zoned_gc_percent);
>> + sbi->gc_thread->boost_zoned_gc_percent) &&
>> + has_enough_dirty_blocks(sbi, LIMIT_GC_DIRTY_SECTION_NUM);
>> return has_enough_invalid_blocks(sbi);
>> }
>> --
>> 2.34.1
Powered by blists - more mailing lists