Linus, please pull the latest scheduler git tree from:
   git://git.kernel.org/pub/scm/linux/kernel/git/mingo/linux-2.6-sched.git
It contains lots of scheduler updates from lots of people - hopefully

the last big one for quite some time. Most of the focus was on

performance (both micro-performance and scalability/balancing), but

there's the fair-scheduling feature now Kconfig selectable too. Find the

shortlog below.
Code that is touched outside of the scheduler: the KVM bits were acked

by Avi, the net/unix change is trivial and only affects sync wakeups,

ditto the fs/pipe.c changes - but i can push those separately if it

needs an ack from David first.
ABI/API changes:
 - new CONFIG_FAIR_USER_SCHED and /sys/kernel/uids/ + uevent API.

 - /proc/stat and /proc/<pid>/stat changes for guest-CPU usage [KVM]

 - /proc/sched_debug formats changed/enhanced
Testing status: the changes are chronological and all the

interactivity-impacting changes are near the head of the queue and most

of them were done weeks ago, and were thus part of the CFS-v22 backport

series - which was tested by many people. There are no known regressions

at the moment. It's all fully bisectable.
Thanks,
	Ingo
------------------>

Alexey Dobriyan (1):

      sched: uninline scheduler
Andi Kleen (4):

      sched: cleanup: remove unnecessary gotos

      sched: cleanup: refactor common code of sleep_on / wait_for_completion

      sched: cleanup: refactor normalize_rt_tasks

      sched: remove stale comment from sched_group_set_shares()
Arjan van de Ven (1):

      Make scheduler debug file operations const
Dhaval Giani (1):

      sched: group scheduling, sysfs tunables
Dmitry Adamushko (14):

      sched: clean up struct load_stat

      sched: clean up schedstat block in dequeue_entity()

      sched: sched_setscheduler() fix

      sched: add set_curr_task() calls

      sched: do not keep current in the tree and get rid of sched_entity::fair_key

      sched: optimize task_ne...
[ message continues ]

How does this one compare to the v22 you released earlier ?
I'm thinking of backporting any fixes/optimizations to 2.6.22

(and possibly 2.6.23)
--

Thomas

-

i have already backported it as v22.1 - will release it within a few

days. (once the currently open regressions have been fixed)
	Ingo

-

i've uploaded what i have at the moment, to:
   http://people.redhat.com/mingo/cfs-scheduler/devel/sched-cfs-v2.6.23.1-v...
	Ingo

-

Big thanks for your work...
Now I just have to see if I can get it to work with the -hrt series and

I'm really happy ;-)
--

Thomas

-

Nice work...
However it's a pity all the balancing stuff got wildly changed

in 2.6.23 and then somewhat changed back again now.
Despite appearances, a lot of those things weren't actually

*completely* arbitrary values. I fear that it will make finding

performance regressions harder than it should have...
Anyway.
-

so i dropped them and re-pushed. New shortlog below.
	Ingo
------------------>

Alexey Dobriyan (1):

      sched: uninline scheduler
Andi Kleen (4):

      sched: cleanup: remove unnecessary gotos

      sched: cleanup: refactor common code of sleep_on / wait_for_completion

      sched: cleanup: refactor normalize_rt_tasks

      sched: remove stale comment from sched_group_set_shares()
Arjan van de Ven (1):

      Make scheduler debug file operations const
Dhaval Giani (1):

      sched: group scheduling, sysfs tunables
Dmitry Adamushko (14):

      sched: clean up struct load_stat

      sched: clean up schedstat block in dequeue_entity()

      sched: sched_setscheduler() fix

      sched: add set_curr_task() calls

      sched: do not keep current in the tree and get rid of sched_entity::fair_key

      sched: optimize task_new_fair()

      sched: simplify sched_class::yield_task()

      sched: rework enqueue/dequeue_entity() to get rid of set_curr_task()

      sched: yield fix

      sched: fix __pick_next_entity()

      sched: tidy up SCHED_RR

      sched: cleanup, remove calc_weighted()

      sched: cleanup, make dequeue_entity() and update_stats_wait_end() similar

      sched: fix group scheduling for SCHED_BATCH
Gautham R Shenoy (1):

      sched: fix rt ptracer monopolizing CPU
Hiroshi Shimamoto (1):

      sched: clean up sched_fork()
Ingo Molnar (71):

      sched: fix sysctl_sched_child_runs_first flag

      sched: resched task in task_new_fair()

      sched: small sched_debug cleanup

      sched: debug: track maximum 'slice'

      sched: uniform tunings

      sched: use constants if !CONFIG_SCHED_DEBUG

      sched: remove stat_gran

      sched: remove precise CPU load

      sched: remove precise CPU load calculations #2

      sched: track cfs_rq->curr on !group-scheduling too

      sched: cleanup: simplify cfs_rq_curr() methods

      sched: uninline __enqueue_entity()/__dequeue_entity()

      sched: speed up update_load_add/_sub()

      sc...
[ message continues ]

On Mon, 15 Oct 2007 16:17:23 +0200
Did Paul Jackson's crash get fixed?

-

yes - that crash was a showstopper that was holding up the pull request

for 2 days. Paul bisected it down to the culprit and the fix was to do

this in wake_up_new_task():
-       if (!p->sched_class->task_new || !current->se.on_rq) {

+       if (!p->sched_class->task_new || !current->se.on_rq || !rq->cfs.curr) {
(during early bootup the cfs_rq has no curr pointer yet.) It's not clear

why this race did not trigger earlier. (and the two checks can probably

be consolidated into a single "!rq->cfs.curr" condition.)
	Ingo

-

an update on this issue:
shortly, SD_BALANCE_FORK is required to trigger this problem and

hence, only NUMA machines could have been affected by it (and only

ia64 and x86 have SD_BALANCE_FORK in SD_NODE_INIT).
more details:
it's perfectly legitimate for 'rq->cfs.curr' to be NULL in

task_new_fair() in the case when this_cpu != task_cpu(p) (p -- is a

newly created task).
why this_cpu != task_cpu(p) :
do_fork() --> copy_process() --> sched_fork() -->

cpu = sched_balance_self(this_cpu, SD_BALANCE_FORK)
chose a different cpu for the new task and there is _no_

'class_sched_fair' task running on this cpu at the moment (that's why

rq->cfs.curr == NULL).
[ thanks a lot to Paul for providing debugging information ]
btw., it's not the 'curr->vruntime < se->vruntime' part in

task_new_fair() that gave us the oops (it's only executed in the case

of this_cpu == task_cpu(p)) _but_ it's rather:
2 checks are required as 'current' and rq->cfs.curr are not the same :-)

It also should work if we just get rid of [*] or add an adiitional

(curr != NULL) check there.
just as a additional observation:
there are lots of per-cpu threads (like events/cpu, ksoftirq/cpu,

etc.) being created on start-up (x NUMBER_OF_CPUS) and SD_SCHED_FORK

(actually, sched_balance_self() from sched_fork()) is just an overhead

in this case...

although, sched_balance_self() is likely to be responsible for a minor

% of the time taken to create a new context so optimizing it away

(esp. for some corner cases) won't improve the start-up time
--

Best regards,

Dmitry Adamushko

-

Maybe not related to that but now my box is killed after this merge.
When I do not much on the box I get maybe 6h uptime , by doing some work ( compiling etc ) is random freeze.
I was able to capture the OOps finally :
...
[15692.917111] BUG: unable to handle kernel NULL pointer dereference at virtual address 00000044

[15692.917159]  printing eip:

[15692.917174] c0111f90

[15692.917185] *pde = 00000000

[15692.917200] Oops: 0000 [#1]

[15692.917208] PREEMPT SMP

[15692.917240] Modules linked in: fuse netconsole configfs pc87360 hwmon_vid eeprom adm1021 uhci_hcd sr_mod shpchp pci_hotplug ohci_hcd iTCO_wdt iTCO_vendor_support intel_agp i82860_edac i2c_i801 ehci_hcd usbcore edac_core cdrom agpgart 3c59x mii ext4dev jbd2 capability commoncap loop lp parport_pc parport evdev

[15692.917623] CPU:    0

[15692.917625] EIP:    0060:[<c0111f90>]    Not tainted VLI

[15692.917629] EFLAGS: 00010046   (2.6.23-g65a6ec0d #330)

[15692.917661] EIP is at pick_next_task_fair+0x1f/0x2d

[15692.917672] eax: c150a7b8   ebx: 00000000   ecx: 00000000   edx: 00000000

[15692.917689] esi: c1507a48   edi: 00000000   ebp: 00eaaf7a   esp: cb1fdf14

[15692.917701] ds: 007b   es: 007b   fs: 00d8  gs: 0000  ss: 0068

[15692.917715] Process sed (pid: 28999, ti=cb1fc000 task=cfdc3500 task.ti=cb1fc000)

[15692.917725] Stack: c02f8268 c02ef7b5 00000002 cb1fdf58 cb1fdf50 00000000 c0400f38 c0403780

[15692.917833]        cfdc3500 cfdc3634 c150a780 00000000 c011a8e7 00000000 c1077aa0 000000ff

[15692.917942]        00000000 00000000 00000000 cb1fdf8c 00000010 cfdc3500 cb1fdf8c c011ace5

[15692.918048] Call Trace:

[15692.918072]  [<c02ef7b5>] schedule+0x321/0x58f

[15692.918109]  [<c011a8e7>] do_exit+0x293/0x6c6

[15692.918143]  [<c011ace5>] do_exit+0x691/0x6c6

[15692.918169]  [<c011ad87>] sys_exit_group+0x0/0xd

[15692.918195]  [<c01026e6>] sysenter_past_esp+0x5f/0x85

[15692.918232]  =======================

[15692.918244] Code: 8b 53 28 89 43 34 89 53 38 5b 5e c3 53 31 d2 83 78 40 00 74 20 83 c...
[ message continues ]

[ cc'ed Srivatsa ]
Gabriel, could you please post a disassembled code for pick_next_task_fair()?

(objdump -d kernel/sched.o and then search for pick_next_task_fair --

copy_and_past)
anyway, my guess is that it's :
                se = pick_next_entity(cfs_rq);

                cfs_rq = group_cfs_rq(se);
'se' _happens_ to be NULL and group_cf_rq(se) does se->my_q and

(according to my calculations) offset(my_q) == 68 (0x44) for x86 32bit

system with CONFIG_SCHEDSTATS=n and CONFIG_FAIR_GROUP_SCHED=y

(according to the config).
that might take place provided put_prev_task_fair() failed for some

reason to insert 'current' (or its corresponding group element) back

into the tree in schedule()... say, due to some inconsistency in

cfs_rq's data.
Srivatsa, that's somewhat similar to another issue that has been

posted earlier today (crash in put_prev_task_fair() -->

__enqueue_task() --> rb_insert_color()) that you are already aware of

...  (/me will continue tomorrow).
--

Best regards,

Dmitry Adamushko

-

Sure here it is :
00000e49 <pick_next_task_fair>:

     e49:       53                      push   %ebx

     e4a:       31 d2                   xor    %edx,%edx

     e4c:       83 78 40 00             cmpl   $0x0,0x40(%eax)

     e50:       74 20                   je     e72 <pick_next_task_fair+0x29>

     e52:       83 c0 38                add    $0x38,%eax

     e55:       8b 50 20                mov    0x20(%eax),%edx

     e58:       31 db                   xor    %ebx,%ebx

     e5a:       85 d2                   test   %edx,%edx

     e5c:       74 0a                   je     e68 <pick_next_task_fair+0x1f>

     e5e:       8d 5a f8                lea    -0x8(%edx),%ebx

     e61:       89 da                   mov    %ebx,%edx

     e63:       e8 a9 ff ff ff          call   e11 <set_next_entity>

     e68:       8b 43 44                mov    0x44(%ebx),%eax

     e6b:       85 c0                   test   %eax,%eax

     e6d:       75 e6                   jne    e55 <pick_next_task_fair+0xc>

     e6f:       8d 53 d0                lea    -0x30(%ebx),%edx

     e72:       89 d0                   mov    %edx,%eax

     e74:       5b                      pop    %ebx

-