Skip to contentSkip to conversation

f2fs causes system stall

DISCUSSION Marc Dietrich 2 participants 6 messages Apr 23, 2014 – May 5, 2014 F2FS file system

lore
Share & tools
lore
#0 lore raw

Hi,

it took me a while to bisect a problem which causes my ARM (tegra2) netbook to frequently stall for 5-10 seconds when I enable EXA acceleration (opentegra experimental ddx). I can only guess that the offending commit 8618b88 "f2fs:
fix not to write data pages on the page reclaiming path" somehow causes high kernel latencies which may interrupt/disturb the command stream to the gpu in this particular case. Reverting this commit fixes the problem. Any hints?

Regards,

Marc

#1 lore raw

Hi,

Thank you for the report.
When I veiwed the patch at a glance, it seems that reverting is not a good way.

Instead, could you share the f2fs commit point that you're using? Then, since I have a suspect for this issue, I'd like to write a patch to test.

Thanks,

2014-04-24 (목), 00:00 +0200, Marc Dietrich:

Quoted 13 lines

Hi,

it took me a while to bisect a problem which causes my ARM (tegra2) netbook to frequently stall for 5-10 seconds when I enable EXA acceleration (opentegra experimental ddx). I can only guess that the offending commit 8618b88 "f2fs:
fix not to write data pages on the page reclaiming path" somehow causes high kernel latencies which may interrupt/disturb the command stream to the gpu in this particular case. Reverting this commit fixes the problem. Any hints?

Regards,

Marc

Signature · 2 lines
Jaegeuk Kim
Samsung
#2 lore raw

Hi,

Quoted 7 lines · On Thu, 24 Apr 2014 08:54:30 +0900 Jaegeuk Kim <[email protected]> wrote:

Thank you for the report.
When I veiwed the patch at a glance, it seems that reverting is not a good way.

Instead, could you share the f2fs commit point that you're using? Then, since I have a suspect for this issue, I'd like to write a patch to test.

patch against linus master is ok for me. I'm happy to test your patch.

Quoted 9 lines

2014-04-24 (목), 00:00 +0200, Marc Dietrich:

Quoted 8 lines

Hi,

it took me a while to bisect a problem which causes my ARM (tegra2) netbook to frequently stall for 5-10 seconds when I enable EXA acceleration (opentegra experimental ddx). I can only guess that the offending commit 8618b88 "f2fs:
fix not to write data pages on the page reclaiming path" somehow causes high kernel latencies which may interrupt/disturb the command stream to the gpu in this particular case. Reverting this commit fixes the problem. Any hints?

Regards,

Marc

#3 lore raw

Thank you very much.

Could you test the following patch?

---
 fs/f2fs/data.c | 2 ++
 1 file changed, 2 insertions(+)
fs/f2fs/data.c +2 -0
index 150c12a..0147de7 100644
@@ -833,6 +833,8 @@ out:
 	unlock_page(page);
 	if (need_balance_fs)
 		f2fs_balance_fs(sbi);
+	if (wbc->for_reclaim)
+		f2fs_submit_merged_bio(sbi, DATA, WRITE);
 	return 0;
 
 redirty_out:
Signature · 6 lines
1.8.4.474.g128a96c


-- 
Jaegeuk Kim
Samsung
#4 lore raw
Quoted 3 lines · On Fri, 25 Apr 2014 08:33:56 +0900 Jaegeuk Kim <[email protected]> wrote:

Thank you very much.

Could you test the following patch?

seems to be fixed now. Thanks!

Marc

Quoted 25 lines
---
 fs/f2fs/data.c | 2 ++
 1 file changed, 2 insertions(+)
fs/f2fs/data.c +2 -0
index 150c12a..0147de7 100644
@@ -833,6 +833,8 @@ out:
 	unlock_page(page);
 	if (need_balance_fs)
 		f2fs_balance_fs(sbi);
+	if (wbc->for_reclaim)
+		f2fs_submit_merged_bio(sbi, DATA, WRITE);
 	return 0;
 
 redirty_out:
Signature · 6 lines
1.8.4.474.g128a96c


-- 
Jaegeuk Kim
Samsung
Signature · 1 line
Marc Dietrich <[email protected]>
#5 lore raw

Jaeguek,

Am Freitag, 25. April 2014, 14:39:00 schrieb Marc Dietrich:

Quoted 8 lines

On Fri, 25 Apr 2014 08:33:56 +0900

Quoted 3 lines · Jaegeuk Kim

Thank you very much.

Could you test the following patch?

seems to be fixed now. Thanks!

are you planing to submit this fix ("f2fs: submit bio at the reclaim path") to 3.15?

Thanks,

Marc

Quoted 23 lines
Quoted 23 lines

---

 fs/f2fs/data.c | 2 ++
 1 file changed, 2 insertions(+)
fs/f2fs/data.c +2 -0
index 150c12a..0147de7 100644
@@ -833,6 +833,8 @@ out:
 	unlock_page(page);
 	if (need_balance_fs)
 
 		f2fs_balance_fs(sbi);
 
+	if (wbc->for_reclaim)
+		f2fs_submit_merged_bio(sbi, DATA, WRITE);
 

return 0;

redirty_out: