Page: 1, 2  Next

scheduler
zika
Status: Interested
Joined: 02 Apr 2012
Posts: 28
Reply Quote
Is it me or...?
Liquorix:
:: Code ::
:~$ uname -a
Linux **** 3.18-2.dmz.1-liquorix-amd64 #1 ZEN SMP PREEMPT Debian 3.18-2 (2015-01-10) x86_64 x86_64 x86_64 GNU/Linux
:: Code ::
:~$ cat /sys/block/sda/queue/scheduler
none

3.19-999:
:: Code ::
:~$ uname -a
Linux **** 3.19.0-999-generic #201501170205 SMP Sat Jan 17 02:06:48 UTC 2015 x86_64 x86_64 x86_64 GNU/Linux
:: Code ::
:~$ cat /sys/block/sda/queue/scheduler
noop [deadline] cfq

Back to top
damentz
Status: Assistant
Joined: 09 Sep 2008
Posts: 1122
Reply Quote
Yes, this is strange. I believe it's something in the BFQ patches (barring anything else we added to the zen kernel).

Fortunately, the IO scheduler shows up in the dmesg logs:
:: Code ::
$ dmesg | grep -i bfq
[    1.345291] io scheduler bfq registered (default)
[    1.345291] BFQ I/O-scheduler version: v7r7


Can you confirm?
Back to top
zika
Status: Interested
Joined: 02 Apr 2012
Posts: 28
Reply Quote
:: damentz wrote ::
Yes, this is strange. I believe it's something in the BFQ patches (barring anything else we added to the zen kernel).
Fortunately, the IO scheduler shows up in the dmesg logs:
:: Code ::
$ dmesg | grep -i bfq
[    1.345291] io scheduler bfq registered (default)
[    1.345291] BFQ I/O-scheduler version: v7r7
Can you confirm?
:: Code ::
:~$ dmesg | grep -i bfq
[    0.037703] Initializing cgroup subsys bfqio
[    0.593596] io scheduler bfq registered
[    0.593598] BFQ I/O-scheduler version: v7r7

Back to top
zika
Status: Interested
Joined: 02 Apr 2012
Posts: 28
Reply Quote
This morning, before going to work, I stole a minute or two to check new edition of 3.18 (3) and it hit kernel panic very early in boot. I've tried even reinstall but...
Update₁:
:: Code ::
:~$ uname -a
Linux miha 3.18-3.dmz.2-liquorix-amd64 #1 ZEN SMP PREEMPT Debian 3.18-4 (2015-01-20) x86_64 x86_64 x86_64 GNU/Linux
:~$ cat /sys/block/sda/queue/scheduler
none

Not getting kernel panic...

< Edited by zika :: Feb 8, 15, 12:02 >

Back to top
damentz
Status: Assistant
Joined: 09 Sep 2008
Posts: 1122
Reply Quote
Yes, sorry about that, I didn't properly clean the 3.dmz.1 kernel package before building.

I haven't had time to investigate the IO scheduler properly. I did a few tests, removed the BFQ patch, tried different combinations of IO schedulers - nothing seemed to make a difference. I'll continue to try different things and get to the bottom of it.
Back to top
braveheartleo
Status: Interested
Joined: 23 Jan 2012
Posts: 43
Reply Quote
Hi damentz,

Has there been any progress regarding this?

It still shows none for 3.19-2.dmz.1-liquorix-amd64. However, stock kernels from Ubuntu or Debian are not affected by this.
Back to top
arclance
Status: Interested
Joined: 26 May 2012
Posts: 41
Reply Quote
I would like to know the progress on it as well.
This must be the reason I get "n/a" for what scheduler each disk is using in conky.
Back to top
damentz
Status: Assistant
Joined: 09 Sep 2008
Posts: 1122
Reply Quote
Sorry guys, I haven't spent much time debugging this defect the past 2 months. My intention was to release 3.19 and cross my fingers that the scheduler would list properly in /sys/. That doesn't appear to have happened.

I don't have that much time to spend on troubleshooting this issue for the next 2 or 3 weeks. So, instead what I'll do until I do get a chance is upload the archive of past builds to liquorix.net/archive if you guys want to check for yourselves.

I think it's possible for us to at least bisect back to the last known good kernel package. From there, the change that introduced the defect should be discoverable through diffing the kernel package folders and files.
Back to top
braveheartleo
Status: Interested
Joined: 23 Jan 2012
Posts: 43
Reply Quote
It looks like the problem lies with the use of CONFIG_SCSI_MQ_DEFAULT=y in liquorix, whereas stock kernels default to 'not set'.

From this presentation, on slide 23 tells that one of the missing features is I/O scheduler support:
:: Quote ::
Missing features:
I/O scheduler support in blk-mq

Blk-mq support for SCSI was merged in 3.17, and can be turned on using scsi_mod.use_blk_mq=y boot option. But since Liquorix turns this on by default (CONFIG_SCSI_MQ_DEFAULT=y), if need to have the ability to change schedulers once again, one can pass a 'n' to this boot option.
Back to top
damentz
Status: Assistant
Joined: 09 Sep 2008
Posts: 1122
Reply Quote
braveheartleo,

Great find! I disabled this in the configuration and will roll out updates to both the 3.19 and 3.18 kernels.
Back to top
Display posts from previous:   
Page: 1, 2  Next
All times are GMT - 8 Hours