2021-07-20 06:53:44

by Zhaoyang Huang

[permalink] [raw]
Subject: [PATCH] driver: zram: fix wrong counting on pages_store

From: Zhaoyang Huang <[email protected]>

It makes the statistic confused when counting pages_store in during writeback.
pages_store should just reflect the swapped pages on ZRAM, while bd_writes be
responsible for writeback things.

Signed-off-by: Zhaoyang Huang <[email protected]>
---
drivers/block/zram/zram_drv.c | 1 -
1 file changed, 1 deletion(-)

diff --git a/drivers/block/zram/zram_drv.c b/drivers/block/zram/zram_drv.c
index cf8deec..6bdaa9d 100644
--- a/drivers/block/zram/zram_drv.c
+++ b/drivers/block/zram/zram_drv.c
@@ -765,7 +765,6 @@ static ssize_t writeback_store(struct device *dev,
zram_set_flag(zram, index, ZRAM_WB);
zram_set_element(zram, index, blk_idx);
blk_idx = 0;
- atomic64_inc(&zram->stats.pages_stored);
spin_lock(&zram->wb_limit_lock);
if (zram->wb_limit_enable && zram->bd_wb_limit > 0)
zram->bd_wb_limit -= 1UL << (PAGE_SHIFT - 12);
--
1.9.1


2021-07-25 01:26:06

by Zhaoyang Huang

[permalink] [raw]
Subject: [Resend PATCH] driver: zram: fix wrong counting on pages_store

Subject: [PATCH] driver: zram: fix wrong counting on pages_store
Date: Tue, 20 Jul 2021 14:47:38 +0800
Message-Id: <1626763658-27748-1-git-se[email protected]>
X-Mailer: git-send-email 1.7.9.5

From: Zhaoyang Huang <[email protected]>

It makes the statistic confused when counting pages_store in during writeback.
pages_store should just reflect the swapped pages on ZRAM, while bd_writes be
responsible for writeback things.

Signed-off-by: Zhaoyang Huang <[email protected]>
---
drivers/block/zram/zram_drv.c | 1 -
1 file changed, 1 deletion(-)

diff --git a/drivers/block/zram/zram_drv.c b/drivers/block/zram/zram_drv.c
index cf8deec..6bdaa9d 100644
--- a/drivers/block/zram/zram_drv.c
+++ b/drivers/block/zram/zram_drv.c
@@ -765,7 +765,6 @@ static ssize_t writeback_store(struct device *dev,
zram_set_flag(zram, index, ZRAM_WB);
zram_set_element(zram, index, blk_idx);
blk_idx = 0;
- atomic64_inc(&zram->stats.pages_stored);
spin_lock(&zram->wb_limit_lock);
if (zram->wb_limit_enable && zram->bd_wb_limit > 0)
zram->bd_wb_limit -= 1UL << (PAGE_SHIFT - 12);
--
1.9.1

2021-07-26 01:29:13

by Sergey Senozhatsky

[permalink] [raw]
Subject: Re: [PATCH] driver: zram: fix wrong counting on pages_store

On (21/07/20 14:47), Huangzhaoyang wrote:
> It makes the statistic confused when counting pages_store in during writeback.
> pages_store should just reflect the swapped pages on ZRAM, while bd_writes be
> responsible for writeback things.
>
> Signed-off-by: Zhaoyang Huang <[email protected]>
> ---
> drivers/block/zram/zram_drv.c | 1 -
> 1 file changed, 1 deletion(-)
>
> diff --git a/drivers/block/zram/zram_drv.c b/drivers/block/zram/zram_drv.c
> index cf8deec..6bdaa9d 100644
> --- a/drivers/block/zram/zram_drv.c
> +++ b/drivers/block/zram/zram_drv.c
> @@ -765,7 +765,6 @@ static ssize_t writeback_store(struct device *dev,
> zram_set_flag(zram, index, ZRAM_WB);
> zram_set_element(zram, index, blk_idx);
> blk_idx = 0;
> - atomic64_inc(&zram->stats.pages_stored);

Yeah, I think we don't need it here.

Minchan, what do you think?

2021-07-26 17:55:46

by Minchan Kim

[permalink] [raw]
Subject: Re: [PATCH] driver: zram: fix wrong counting on pages_store

On Mon, Jul 26, 2021 at 10:27:40AM +0900, Sergey Senozhatsky wrote:
> On (21/07/20 14:47), Huangzhaoyang wrote:
> > It makes the statistic confused when counting pages_store in during writeback.
> > pages_store should just reflect the swapped pages on ZRAM, while bd_writes be
> > responsible for writeback things.
> >
> > Signed-off-by: Zhaoyang Huang <[email protected]>
> > ---
> > drivers/block/zram/zram_drv.c | 1 -
> > 1 file changed, 1 deletion(-)
> >
> > diff --git a/drivers/block/zram/zram_drv.c b/drivers/block/zram/zram_drv.c
> > index cf8deec..6bdaa9d 100644
> > --- a/drivers/block/zram/zram_drv.c
> > +++ b/drivers/block/zram/zram_drv.c
> > @@ -765,7 +765,6 @@ static ssize_t writeback_store(struct device *dev,
> > zram_set_flag(zram, index, ZRAM_WB);
> > zram_set_element(zram, index, blk_idx);
> > blk_idx = 0;
> > - atomic64_inc(&zram->stats.pages_stored);
>
> Yeah, I think we don't need it here.
>
> Minchan, what do you think?

What pages_stored currently represents is orig_data_size from mm_stat

================ =============================================================
orig_data_size uncompressed size of data stored in this disk.
Unit: bytes

Thus, if we don't increase it after zram_free_page, it loses what's
the original data size which would be important to calculate
compression ratio along with compr_data_size.

2021-07-27 01:34:35

by Zhaoyang Huang

[permalink] [raw]
Subject: Re: [PATCH] driver: zram: fix wrong counting on pages_store

On Tue, Jul 27, 2021 at 1:53 AM Minchan Kim <[email protected]> wrote:
>
> On Mon, Jul 26, 2021 at 10:27:40AM +0900, Sergey Senozhatsky wrote:
> > On (21/07/20 14:47), Huangzhaoyang wrote:
> > > It makes the statistic confused when counting pages_store in during writeback.
> > > pages_store should just reflect the swapped pages on ZRAM, while bd_writes be
> > > responsible for writeback things.
> > >
> > > Signed-off-by: Zhaoyang Huang <[email protected]>
> > > ---
> > > drivers/block/zram/zram_drv.c | 1 -
> > > 1 file changed, 1 deletion(-)
> > >
> > > diff --git a/drivers/block/zram/zram_drv.c b/drivers/block/zram/zram_drv.c
> > > index cf8deec..6bdaa9d 100644
> > > --- a/drivers/block/zram/zram_drv.c
> > > +++ b/drivers/block/zram/zram_drv.c
> > > @@ -765,7 +765,6 @@ static ssize_t writeback_store(struct device *dev,
> > > zram_set_flag(zram, index, ZRAM_WB);
> > > zram_set_element(zram, index, blk_idx);
> > > blk_idx = 0;
> > > - atomic64_inc(&zram->stats.pages_stored);
> >
> > Yeah, I think we don't need it here.
> >
> > Minchan, what do you think?
>
> What pages_stored currently represents is orig_data_size from mm_stat
>
> ================ =============================================================
> orig_data_size uncompressed size of data stored in this disk.
> Unit: bytes
>
> Thus, if we don't increase it after zram_free_page, it loses what's
> the original data size which would be important to calculate
> compression ratio along with compr_data_size.
zram_free_page?? What the commit try to fix is pages_stored being
increased incorrectly in writeback_store, which will lead to wrong
statistic on orig size of ZRAM.

2021-07-27 05:49:20

by Zhaoyang Huang

[permalink] [raw]
Subject: Re: [PATCH] driver: zram: fix wrong counting on pages_store

On Tue, Jul 27, 2021 at 9:32 AM Zhaoyang Huang <[email protected]> wrote:
>
> On Tue, Jul 27, 2021 at 1:53 AM Minchan Kim <[email protected]> wrote:
> >
> > On Mon, Jul 26, 2021 at 10:27:40AM +0900, Sergey Senozhatsky wrote:
> > > On (21/07/20 14:47), Huangzhaoyang wrote:
> > > > It makes the statistic confused when counting pages_store in during writeback.
> > > > pages_store should just reflect the swapped pages on ZRAM, while bd_writes be
> > > > responsible for writeback things.
> > > >
> > > > Signed-off-by: Zhaoyang Huang <[email protected]>
> > > > ---
> > > > drivers/block/zram/zram_drv.c | 1 -
> > > > 1 file changed, 1 deletion(-)
> > > >
> > > > diff --git a/drivers/block/zram/zram_drv.c b/drivers/block/zram/zram_drv.c
> > > > index cf8deec..6bdaa9d 100644
> > > > --- a/drivers/block/zram/zram_drv.c
> > > > +++ b/drivers/block/zram/zram_drv.c
> > > > @@ -765,7 +765,6 @@ static ssize_t writeback_store(struct device *dev,
> > > > zram_set_flag(zram, index, ZRAM_WB);
> > > > zram_set_element(zram, index, blk_idx);
> > > > blk_idx = 0;
> > > > - atomic64_inc(&zram->stats.pages_stored);
> > >
> > > Yeah, I think we don't need it here.
> > >
> > > Minchan, what do you think?
> >
> > What pages_stored currently represents is orig_data_size from mm_stat
> >
> > ================ =============================================================
> > orig_data_size uncompressed size of data stored in this disk.
> > Unit: bytes
> >
> > Thus, if we don't increase it after zram_free_page, it loses what's
> > the original data size which would be important to calculate
> > compression ratio along with compr_data_size.
> zram_free_page?? What the commit try to fix is pages_stored being
> increased incorrectly in writeback_store, which will lead to wrong
> statistic on orig size of ZRAM.
OK, if I understand right, I think what you mean is the page count
corresponding to SWAP should be still available even if it is written
back to the backing dev.So we need to increase page_stored again after
zram_free_page free it.