Hi Jaegeuk, On 2017/5/12 2:36, Jaegeuk Kim wrote: > Hi Chao, > > On 05/09, Chao Yu wrote: >> From: Chao Yu <[email protected]> >> >> Serialize data/node IOs by using fifo list instead of mutex lock, >> it will help to enhance concurrency of f2fs, meanwhile keeping LFS >> IO semantics. > > I'm not against to give it a try, but not sure how much we can get a benefit > from this approach frankly. Have you got a trouble on any lock contention from > the below io_rwsem or mutex?
Yes, because submitting IOs can be blocked in block layer since there may be: - limitation of some resources, e.g. request number. - IO throttle Holding a global mutex lock in the path is not good idea, as it may cause potential hungtask or concurrency performance regression. So I add this patch to relief impacting of global mutex. Thanks, > > Thanks, > >> >> Signed-off-by: Chao Yu <[email protected]> >> --- >> fs/f2fs/checkpoint.c | 1 + >> fs/f2fs/data.c | 28 ++++++++++++++++++++++++---- >> fs/f2fs/f2fs.h | 5 ++++- >> fs/f2fs/gc.c | 3 ++- >> fs/f2fs/segment.c | 20 ++++++++++++++------ >> fs/f2fs/segment.h | 3 ++- >> 6 files changed, 47 insertions(+), 13 deletions(-) >> >> diff --git a/fs/f2fs/checkpoint.c b/fs/f2fs/checkpoint.c >> index 2a475e83a092..7b3393474f6b 100644 >> --- a/fs/f2fs/checkpoint.c >> +++ b/fs/f2fs/checkpoint.c >> @@ -162,6 +162,7 @@ int ra_meta_pages(struct f2fs_sb_info *sbi, block_t >> start, int nrpages, >> .op = REQ_OP_READ, >> .op_flags = sync ? (REQ_META | REQ_PRIO) : REQ_RAHEAD, >> .encrypted_page = NULL, >> + .in_list = false, >> }; >> struct blk_plug plug; >> >> diff --git a/fs/f2fs/data.c b/fs/f2fs/data.c >> index 5f001b471252..89eaa8aaa97b 100644 >> --- a/fs/f2fs/data.c >> +++ b/fs/f2fs/data.c >> @@ -393,10 +393,28 @@ int f2fs_submit_page_mbio(struct f2fs_io_info *fio) >> struct f2fs_bio_info *io; >> bool is_read = is_read_io(fio->op); >> struct page *bio_page; >> + struct curseg_info *curseg; >> int err = 0; >> >> + if (fio->in_list) >> + curseg = CURSEG_I(sbi, fio->seg_type); >> + >> io = __get_bio_info(sbi, fio->op, fio->type, fio->seg_type); >> >> + down_write(&io->io_rwsem); >> +next: >> + if (fio->in_list) { >> + spin_lock(&curseg->io_lock); >> + if (list_empty(&curseg->io_list)) { >> + spin_unlock(&curseg->io_lock); >> + goto out_fail; >> + } >> + fio = list_first_entry(&curseg->io_list, >> + struct f2fs_io_info, list); >> + list_del(&fio->list); >> + spin_unlock(&curseg->io_lock); >> + } >> + >> if (fio->old_blkaddr != NEW_ADDR) >> verify_block_addr(sbi, fio->old_blkaddr); >> verify_block_addr(sbi, fio->new_blkaddr); >> @@ -409,8 +427,6 @@ int f2fs_submit_page_mbio(struct f2fs_io_info *fio) >> if (!is_read) >> inc_page_count(sbi, WB_DATA_TYPE(bio_page)); >> >> - down_write(&io->io_rwsem); >> - >> if (io->bio && (io->last_block_in_bio != fio->new_blkaddr - 1 || >> (io->fio.op != fio->op || io->fio.op_flags != fio->op_flags) || >> !__same_bdev(sbi, fio->new_blkaddr, io->bio))) >> @@ -437,9 +453,13 @@ int f2fs_submit_page_mbio(struct f2fs_io_info *fio) >> >> io->last_block_in_bio = fio->new_blkaddr; >> f2fs_trace_ios(fio, 0); >> + >> + trace_f2fs_submit_page_mbio(fio->page, fio); >> + >> + if (fio->in_list) >> + goto next; >> out_fail: >> up_write(&io->io_rwsem); >> - trace_f2fs_submit_page_mbio(fio->page, fio); >> return err; >> } >> >> @@ -752,7 +772,7 @@ static int __allocate_data_block(struct dnode_of_data >> *dn) >> set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version); >> >> allocate_data_block(sbi, NULL, dn->data_blkaddr, &dn->data_blkaddr, >> - &sum, CURSEG_WARM_DATA); >> + &sum, CURSEG_WARM_DATA, NULL, false); >> set_data_blkaddr(dn); >> >> /* update i_size */ >> diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h >> index 9129a6229bc8..6b8e9f051aa2 100644 >> --- a/fs/f2fs/f2fs.h >> +++ b/fs/f2fs/f2fs.h >> @@ -802,8 +802,10 @@ struct f2fs_io_info { >> block_t old_blkaddr; /* old block address before Cow */ >> struct page *page; /* page to be written */ >> struct page *encrypted_page; /* encrypted page */ >> + struct list_head list; /* serialize IOs */ >> bool submitted; /* indicate IO submission */ >> bool need_lock; /* indicate we need to lock cp_rwsem */ >> + bool in_list; /* indicate fio is in io_list */ >> }; >> >> #define is_read_io(rw) ((rw) == READ) >> @@ -2274,7 +2276,8 @@ void f2fs_replace_block(struct f2fs_sb_info *sbi, >> struct dnode_of_data *dn, >> bool recover_newaddr); >> void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page, >> block_t old_blkaddr, block_t *new_blkaddr, >> - struct f2fs_summary *sum, int type); >> + struct f2fs_summary *sum, int type, >> + struct f2fs_io_info *fio, bool add_list); >> void f2fs_wait_on_page_writeback(struct page *page, >> enum page_type type, bool ordered); >> void f2fs_wait_on_encrypted_page_writeback(struct f2fs_sb_info *sbi, >> diff --git a/fs/f2fs/gc.c b/fs/f2fs/gc.c >> index 8b267ca30926..ac2f74e40eea 100644 >> --- a/fs/f2fs/gc.c >> +++ b/fs/f2fs/gc.c >> @@ -590,6 +590,7 @@ static void move_encrypted_block(struct inode *inode, >> block_t bidx, >> .op = REQ_OP_READ, >> .op_flags = 0, >> .encrypted_page = NULL, >> + .in_list = false, >> }; >> struct dnode_of_data dn; >> struct f2fs_summary sum; >> @@ -633,7 +634,7 @@ static void move_encrypted_block(struct inode *inode, >> block_t bidx, >> fio.new_blkaddr = fio.old_blkaddr = dn.data_blkaddr; >> >> allocate_data_block(fio.sbi, NULL, fio.old_blkaddr, &newaddr, >> - &sum, fio.seg_type); >> + &sum, fio.seg_type, NULL, false); >> >> fio.encrypted_page = pagecache_get_page(META_MAPPING(fio.sbi), newaddr, >> FGP_LOCK | FGP_CREAT, GFP_NOFS); >> diff --git a/fs/f2fs/segment.c b/fs/f2fs/segment.c >> index c047b5d8b9d3..d4975b8f4620 100644 >> --- a/fs/f2fs/segment.c >> +++ b/fs/f2fs/segment.c >> @@ -2100,7 +2100,8 @@ static int __get_segment_type(struct page *page, enum >> page_type p_type) >> >> void allocate_data_block(struct f2fs_sb_info *sbi, struct page *page, >> block_t old_blkaddr, block_t *new_blkaddr, >> - struct f2fs_summary *sum, int type) >> + struct f2fs_summary *sum, int type, >> + struct f2fs_io_info *fio, bool add_list) >> { >> struct sit_info *sit_i = SIT_I(sbi); >> struct curseg_info *curseg = CURSEG_I(sbi, type); >> @@ -2136,6 +2137,14 @@ void allocate_data_block(struct f2fs_sb_info *sbi, >> struct page *page, >> if (page && IS_NODESEG(type)) >> fill_node_footer_blkaddr(page, NEXT_FREE_BLKADDR(sbi, curseg)); >> >> + if (add_list) { >> + INIT_LIST_HEAD(&fio->list); >> + fio->in_list = true; >> + spin_lock(&curseg->io_lock); >> + list_add_tail(&fio->list, &curseg->io_list); >> + spin_unlock(&curseg->io_lock); >> + } >> + >> mutex_unlock(&curseg->curseg_mutex); >> } >> >> @@ -2145,10 +2154,9 @@ static void do_write_page(struct f2fs_summary *sum, >> struct f2fs_io_info *fio) >> >> fio->seg_type = __get_segment_type(fio->page, fio->type); >> >> - mutex_lock(&CURSEG_I(fio->sbi, fio->seg_type)->wio_mutex); >> reallocate: >> allocate_data_block(fio->sbi, fio->page, fio->old_blkaddr, >> - &fio->new_blkaddr, sum, fio->seg_type); >> + &fio->new_blkaddr, sum, fio->seg_type, fio, true); >> >> /* writeout dirty page into bdev */ >> err = f2fs_submit_page_mbio(fio); >> @@ -2156,8 +2164,6 @@ static void do_write_page(struct f2fs_summary *sum, >> struct f2fs_io_info *fio) >> fio->old_blkaddr = fio->new_blkaddr; >> goto reallocate; >> } >> - >> - mutex_unlock(&CURSEG_I(fio->sbi, fio->seg_type)->wio_mutex); >> } >> >> void write_meta_page(struct f2fs_sb_info *sbi, struct page *page) >> @@ -2171,6 +2177,7 @@ void write_meta_page(struct f2fs_sb_info *sbi, struct >> page *page) >> .new_blkaddr = page->index, >> .page = page, >> .encrypted_page = NULL, >> + .in_list = false, >> }; >> >> if (unlikely(page->index >= MAIN_BLKADDR(sbi))) >> @@ -2978,7 +2985,8 @@ static int build_curseg(struct f2fs_sb_info *sbi) >> array[i].bio_info.sbi = sbi; >> array[i].bio_info.bio = NULL; >> >> - mutex_init(&array[i].wio_mutex); >> + spin_lock_init(&array[i].io_lock); >> + INIT_LIST_HEAD(&array[i].io_list); >> } >> return restore_curseg_summaries(sbi); >> } >> diff --git a/fs/f2fs/segment.h b/fs/f2fs/segment.h >> index 701944b462cd..b6f5dffeaa61 100644 >> --- a/fs/f2fs/segment.h >> +++ b/fs/f2fs/segment.h >> @@ -283,7 +283,8 @@ struct curseg_info { >> struct rw_semaphore journal_rwsem; /* protect journal area */ >> struct f2fs_journal *journal; /* cached journal info */ >> struct f2fs_bio_info bio_info; /* for log bios */ >> - struct mutex wio_mutex; /* serialize DATA/NODE IOs */ >> + spinlock_t io_lock; /* serialize DATA/NODE IOs */ >> + struct list_head io_list; /* tracking fios */ >> unsigned char alloc_type; /* current allocation type */ >> unsigned int segno; /* current segment number */ >> unsigned short next_blkoff; /* next block offset to write */ >> -- >> 2.12.2.575.gb14f27f > > . > ------------------------------------------------------------------------------ Check out the vibrant tech community on one of the world's most engaging tech sites, Slashdot.org! http://sdm.link/slashdot _______________________________________________ Linux-f2fs-devel mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel
