2009-12-30 12:34:53

by Con Kolivas

[permalink] [raw]
Subject: BFS v0.313 CPU scheduler for 2.6.32

A stupid mismerge in __might_sleep meant that with the right config
options dmesg would have been swamped with false warnings. This also
was making BFS consume a lot of CPU under certain conditions in
combination with preempt RCU.

Thanks to Jayson R King for picking it up!

There are no other changes in this version compared to 312

full patch:
http://ck.kolivas.org/patches/bfs/2.6.32-sched-bfs-313.patch

incremental:
http://ck.kolivas.org/patches/bfs/2.6.32-bfs312-313.patch

I highly recommend BFS users upgrade.

Regards,
--
-ck


2009-12-30 13:55:33

by Thomas Backlund

[permalink] [raw]
Subject: Re: BFS v0.313 CPU scheduler for 2.6.32

Con Kolivas skrev 30.12.2009 14:34:
> A stupid mismerge in __might_sleep meant that with the right config
> options dmesg would have been swamped with false warnings. This also
> was making BFS consume a lot of CPU under certain conditions in
> combination with preempt RCU.
>
> Thanks to Jayson R King for picking it up!
>
> There are no other changes in this version compared to 312
>
> full patch:
> http://ck.kolivas.org/patches/bfs/2.6.32-sched-bfs-313.patch
>
> incremental:
> http://ck.kolivas.org/patches/bfs/2.6.32-bfs312-313.patch
>
> I highly recommend BFS users upgrade.
>

Nice.

any timeline for a version for 2.6.33 series kernels ?

--
Thomas

2009-12-30 14:41:50

by Con Kolivas

[permalink] [raw]
Subject: Re: BFS v0.313 CPU scheduler for 2.6.32

2009/12/31 Thomas Backlund <[email protected]>:
> any timeline for a version for 2.6.33 series kernels ?

To minimise my workload I won't start syncing up till an end is in
sight to the rc kernels I'm sorry to say. That's usually around rc8
these days...

Regards,
Con

2009-12-30 18:01:27

by Miguel Botón

[permalink] [raw]
Subject: Re: BFS v0.313 CPU scheduler for 2.6.32

On Wed, Dec 30, 2009 at 2:55 PM, Thomas Backlund <[email protected]> wrote:
> Nice.
>
> any timeline for a version for 2.6.33 series kernels ?

If you want to try BFS on a 2.6.33 kernel, you can try this 'unofficial' port:
http://git.zen-kernel.org/?p=kernel/zen.git;a=shortlog;h=refs/heads/sched-bfs

I'm currently using it with no issues so far.

Greetings,

--
Miguel Botón

2009-12-30 19:55:57

by Heinz Diehl

[permalink] [raw]
Subject: Re: BFS v0.313 CPU scheduler for 2.6.32

On 30.12.2009, Miguel Bot?n wrote:

> If you want to try BFS on a 2.6.33 kernel, you can try this 'unofficial' port:
> http://git.zen-kernel.org/?p=kernel/zen.git;a=shortlog;h=refs/heads/sched-bfs

> I'm currently using it with no issues so far.

Your patch applies cleanly to 2.6.33-rc2, but compilation fails:

[....]
In file included from kernel/sched.c:2:
kernel/sched_bfs.c: In function '__sched_setscheduler':
kernel/sched_bfs.c:3484: warning: passing argument 1 of 'spinlock_check' from incompatible pointer type
kernel/sched_bfs.c:3493: warning: passing argument 1 of 'spin_unlock_irqrestore' from incompatible pointer type
kernel/sched_bfs.c:3509: warning: passing argument 1 of 'spin_unlock_irqrestore' from incompatible pointer type
kernel/sched_bfs.c: In function 'sys_sched_yield':
kernel/sched_bfs.c:3846: warning: passing argument 1 of '_raw_spin_unlock' from incompatible pointer type
kernel/sched_bfs.c: At top level:
kernel/sched_bfs.c:6018: error: conflicting types for 'partition_sched_domains'
include/linux/sched.h:1025: error: previous declaration of 'partition_sched_domains' was here
kernel/sched_bfs.c: In function 'sched_init':
kernel/sched_bfs.c:6425: warning: passing argument 2 of 'plist_head_init' from incompatible pointer type
kernel/sched_bfs.c: In function 'normalize_rt_tasks':
kernel/sched_bfs.c:6499: warning: passing argument 1 of 'spinlock_check' from incompatible pointer type
kernel/sched_bfs.c:6513: warning: passing argument 1 of 'spin_unlock_irqrestore' from incompatible pointer type
make[1]: *** [kernel/sched.o] Error 1
make[1]: *** Waiting for unfinished jobs....
CC arch/x86/kernel/alternative.o
CC fs/open.o
[....]

[....]
CC arch/x86/kernel/init_task.o
LDS arch/x86/kernel/vmlinux.lds
CC [M] arch/x86/kernel/msr.o
CC [M] arch/x86/kernel/cpuid.o
CC [M] arch/x86/kernel/microcode_core.o
CC [M] arch/x86/kernel/microcode_amd.o
arch/x86/kernel/init_task.c:31: warning: missing braces around initializer
arch/x86/kernel/init_task.c:31: warning: (near initialization for 'init_task.pi_lock')
arch/x86/kernel/init_task.c:31: error: incompatible types in initialization
arch/x86/kernel/init_task.c:31: error: initializer element is not constant
arch/x86/kernel/init_task.c:31: error: (near initialization for 'init_task.dirties.lock')
arch/x86/kernel/init_task.c:31: error: initializer element is not constant
arch/x86/kernel/init_task.c:31: error: (near initialization for 'init_task.perf_event_mutex.wait_lock')
make[2]: *** [arch/x86/kernel/init_task.o] Error 1
make[2]: *** Waiting for unfinished jobs....
make[1]: *** [arch/x86/kernel] Error 2
make: *** [arch/x86] Error 2
liesel:/usr/src/linux-2.6.33-rc2 #

2009-12-30 20:58:37

by Miguel Botón

[permalink] [raw]
Subject: Re: BFS v0.313 CPU scheduler for 2.6.32

On Wed, Dec 30, 2009 at 8:55 PM, Heinz Diehl <[email protected]> wrote:
> On 30.12.2009, Miguel Bot?n wrote:
>
>> If you want to try BFS on a 2.6.33 kernel, you can try this 'unofficial' port:
>> http://git.zen-kernel.org/?p=kernel/zen.git;a=shortlog;h=refs/heads/sched-bfs
>
>> I'm currently using it with no issues so far.
>
> Your patch applies cleanly to 2.6.33-rc2, but compilation fails:
>
> [....]

Well, looks like you didn't apply the patch correctly. Your
'sched_bfs.c' is different.

Try this patch: http://omploader.org/vMzRieg/2.6.33-rc2-sched-bfs-313.patch

Greetings,

--
Miguel Bot?n