Re: [f2fs-dev] [PATCH v6] f2fs: compress: add compress_inode to cache compressed blocks

From: Jaegeuk Kim
Date: Thu Jun 03 2021 - 12:13:57 EST

Next message: Vladimir Oltean: "Re: [RESEND PATCH net-next v4 1/3] net: stmmac: split xPCS setup from mdio register"
Previous message: Luck, Tony: "Re: [PATCH v3] libfs: fix error cast of negative value in simple_attr_write()"
In reply to: Chao Yu: "Re: [f2fs-dev] [PATCH v6] f2fs: compress: add compress_inode to cache compressed blocks"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

On 06/03, Chao Yu wrote:
> On 2021/6/2 23:34, Jaegeuk Kim wrote:
> > On 06/02, Chao Yu wrote:
> > > On 2021/6/2 1:27, Jaegeuk Kim wrote:
> > > > On 06/01, Jaegeuk Kim wrote:
> > > > > On 05/26, Chao Yu wrote:
> > > > > > On 2021/5/26 21:26, Jaegeuk Kim wrote:
> > > > > > > On 05/26, Chao Yu wrote:
> > > > > > > > On 2021/5/25 22:01, Jaegeuk Kim wrote:
> > > > > > > > > On 05/25, Chao Yu wrote:
> > > > > > > > > > On 2021/5/25 21:02, Jaegeuk Kim wrote:
> > > > > > > > > > > On 05/25, Jaegeuk Kim wrote:
> > > > > > > > > > > > On 05/25, Chao Yu wrote:
> > > > > > > > > > > > > Also, and queue this?
> > > > > > > > > > > >
> > > > > > > > > > > > Easy to get this?
> > > > > > > > > > >
> > > > > > > > > > > need GFP_NOFS?
> > > > > > > > > >
> > > > > > > > > > Not sure, I use __GFP_IO intentionally here to avoid __GFP_RECLAIM from
> > > > > > > > > > GFP_NOFS, because in low memory case, I don't want to instead page cache
> > > > > > > > > > of normal file with page cache of sbi->compress_inode.
> > > > > > > > > >
> > > > > > > > > > What is memory size in your vm?
> > > > > > > > >
> > > > > > > > > 4GB. If I set GFP_NOFS, I don't see the error anymore, at least.
> > > > > > > >
> > > > > > > > I applied below patch and don't see the warning message anymore.
> > > > > > > >
> > > > > > > > ---
> > > > > > > > fs/f2fs/compress.c | 2 +-
> > > > > > > > 1 file changed, 1 insertion(+), 1 deletion(-)
> > > > > > > >
> > > > > > > > diff --git a/fs/f2fs/compress.c b/fs/f2fs/compress.c
> > > > > > > > index 701dd0f6f4ec..ed5b7fabc604 100644
> > > > > > > > --- a/fs/f2fs/compress.c
> > > > > > > > +++ b/fs/f2fs/compress.c
> > > > > > > > @@ -1703,7 +1703,7 @@ void f2fs_cache_compressed_page(struct f2fs_sb_info *sbi, struct page *page,
> > > > > > > > avail_ram = si.totalram - si.totalhigh;
> > > > > > > >
> > > > > > > > /* free memory is lower than watermark, deny caching compress page */
> > > > > > > > - if (free_ram <= sbi->compress_watermark / 100 * avail_ram)
> > > > > >
> > > > > > This is buggy, because sbi->compress_watermark equals to 20, so that
> > > > > > sbi->compress_watermark / 100 * avail_ram always be zero...
> > > > > >
> > > > > > After this change, if free ram is lower, we may just skip caching
> > > > > > compressed blocks here.
> > > > >
> > > > > Can we move this in f2fs_available_free_memory()?
> > >
> > > More clean.
> > >
> > > One comment below:
> > >
> > > >
> > > > Testing this.
> > > >
> > > > ---
> > > > fs/f2fs/compress.c | 14 +-------------
> > > > fs/f2fs/node.c | 11 ++++++++++-
> > > > fs/f2fs/node.h | 1 +
> > > > 3 files changed, 12 insertions(+), 14 deletions(-)
> > > >
> > > > diff --git a/fs/f2fs/compress.c b/fs/f2fs/compress.c
> > > > index 9fd62a0a646b..455561826c7d 100644
> > > > --- a/fs/f2fs/compress.c
> > > > +++ b/fs/f2fs/compress.c
> > > > @@ -1688,8 +1688,6 @@ void f2fs_cache_compressed_page(struct f2fs_sb_info *sbi, struct page *page,
> > > > {
> > > > struct page *cpage;
> > > > int ret;
> > > > - struct sysinfo si;
> > > > - unsigned long free_ram, avail_ram;
> > > > if (!test_opt(sbi, COMPRESS_CACHE))
> > > > return;
> > > > @@ -1697,17 +1695,7 @@ void f2fs_cache_compressed_page(struct f2fs_sb_info *sbi, struct page *page,
> > > > if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE_READ))
> > > > return;
> > > > - si_meminfo(&si);
> > > > - free_ram = si.freeram;
> > > > - avail_ram = si.totalram - si.totalhigh;
> > > > -
> > > > - /* free memory is lower than watermark, deny caching compress page */
> > > > - if (free_ram <= sbi->compress_watermark / 100 * avail_ram)
> > > > - return;
> > > > -
> > > > - /* cached page count exceed threshold, deny caching compress page */
> > > > - if (COMPRESS_MAPPING(sbi)->nrpages >=
> > >
> > > Need to cover COMPRESS_MAPPING() with CONFIG_F2FS_FS_COMPRESSION.
> >
> > Added like this.
> >
> > --- a/fs/f2fs/node.c
> > +++ b/fs/f2fs/node.c
> > @@ -99,6 +99,7 @@ bool f2fs_available_free_memory(struct f2fs_sb_info *sbi, int type)
> > sizeof(struct discard_cmd)) >> PAGE_SHIFT;
> > res = mem_size < (avail_ram * nm_i->ram_thresh / 100);
> > } else if (type == COMPRESS_PAGE) {
> > +#ifdef CONFIG_F2FS_FS_COMPRESSION
>
> How about adding free_ram definition and assigment here?
>
> unsigned long free_ram = val.freeram;

Done.

>
> Thanks,
>
> > /*
> > * free memory is lower than watermark or cached page count
> > * exceed threshold, deny caching compress page.
> > @@ -106,6 +107,9 @@ bool f2fs_available_free_memory(struct f2fs_sb_info *sbi, int type)
> > res = (free_ram > avail_ram * sbi->compress_watermark / 100) &&
> > (COMPRESS_MAPPING(sbi)->nrpages <
> > free_ram * sbi->compress_percent / 100);
> > +#else
> > + res = false;
> > +#endif
> > } else {
> > if (!sbi->sb->s_bdi->wb.dirty_exceeded)
> > return true;
> >
> > >
> > > Thanks,
> > >
> > > > - free_ram / 100 * sbi->compress_percent)
> > > > + if (!f2fs_available_free_memory(sbi, COMPRESS_PAGE))
> > > > return;
> > > > cpage = find_get_page(COMPRESS_MAPPING(sbi), blkaddr);
> > > > diff --git a/fs/f2fs/node.c b/fs/f2fs/node.c
> > > > index 3a8f7afa5059..67093416ce9c 100644
> > > > --- a/fs/f2fs/node.c
> > > > +++ b/fs/f2fs/node.c
> > > > @@ -45,7 +45,7 @@ bool f2fs_available_free_memory(struct f2fs_sb_info *sbi, int type)
> > > > struct f2fs_nm_info *nm_i = NM_I(sbi);
> > > > struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
> > > > struct sysinfo val;
> > > > - unsigned long avail_ram;
> > > > + unsigned long avail_ram, free_ram;
> > > > unsigned long mem_size = 0;
> > > > bool res = false;
> > > > @@ -56,6 +56,7 @@ bool f2fs_available_free_memory(struct f2fs_sb_info *sbi, int type)
> > > > /* only uses low memory */
> > > > avail_ram = val.totalram - val.totalhigh;
> > > > + free_ram = val.freeram;
> > > > /*
> > > > * give 25%, 25%, 50%, 50%, 50% memory for each components respectively
> > > > @@ -97,6 +98,14 @@ bool f2fs_available_free_memory(struct f2fs_sb_info *sbi, int type)
> > > > mem_size = (atomic_read(&dcc->discard_cmd_cnt) *
> > > > sizeof(struct discard_cmd)) >> PAGE_SHIFT;
> > > > res = mem_size < (avail_ram * nm_i->ram_thresh / 100);
> > > > + } else if (type == COMPRESS_PAGE) {
> > > > + /*
> > > > + * free memory is lower than watermark or cached page count
> > > > + * exceed threshold, deny caching compress page.
> > > > + */
> > > > + res = (free_ram > avail_ram * sbi->compress_watermark / 100) &&
> > > > + (COMPRESS_MAPPING(sbi)->nrpages <
> > > > + free_ram * sbi->compress_percent / 100);
> > > > } else {
> > > > if (!sbi->sb->s_bdi->wb.dirty_exceeded)
> > > > return true;
> > > > diff --git a/fs/f2fs/node.h b/fs/f2fs/node.h
> > > > index d85e8659cfda..84d45385d1f2 100644
> > > > --- a/fs/f2fs/node.h
> > > > +++ b/fs/f2fs/node.h
> > > > @@ -148,6 +148,7 @@ enum mem_type {
> > > > EXTENT_CACHE, /* indicates extent cache */
> > > > INMEM_PAGES, /* indicates inmemory pages */
> > > > DISCARD_CACHE, /* indicates memory of cached discard cmds */
> > > > + COMPRESS_PAGE, /* indicates memory of cached compressed pages */
> > > > BASE_CHECK, /* check kernel status */
> > > > };
> > > >

Next message: Vladimir Oltean: "Re: [RESEND PATCH net-next v4 1/3] net: stmmac: split xPCS setup from mdio register"
Previous message: Luck, Tony: "Re: [PATCH v3] libfs: fix error cast of negative value in simple_attr_write()"
In reply to: Chao Yu: "Re: [f2fs-dev] [PATCH v6] f2fs: compress: add compress_inode to cache compressed blocks"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]