Re: 2.6.35-rc2-git2: Reported regressions from 2.6.34

[Ingo Molnar]

* Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:

> > Bug-Entry	: http://bugzilla.kernel.org/show_bug.cgi?id=16129
> > Subject	: BUG: using smp_processor_id() in preemptible [00000000] code: jbd2/sda2
> > Submitter	: Jan Kreuzer <kontrollator@xxxxxx>
> > Date		: 2010-06-05 06:15 (4 days old)
> 
> This seems to have been introduced by
> 
> 	commit 7cbaef9c83e58bbd4bdd534b09052b6c5ec457d5
> 	Author: Ingo Molnar <mingo@xxxxxxx>
> 	Date:   Sat Nov 8 17:05:38 2008 +0100
> 
> 	    sched: optimize sched_clock() a bit
>     
> 	    sched_clock() uses cycles_2_ns() needlessly - which is an irq-disabling
> 	    variant of __cycles_2_ns().
>     
> 	    Most of the time sched_clock() is called with irqs disabled already.
> 	    The few places that call it with irqs enabled need to be updated.
>     
> 	    Signed-off-by: Ingo Molnar <mingo@xxxxxxx>
> 
> and this seems to be one of those calling cases that need to be updated.

That's a commit from 2008.

> Ingo? The call trace is:
> 
> 	BUG: using smp_processor_id() in preemptible [00000000] code: jbd2/sda2-8/337
> 	caller is native_sched_clock+0x3c/0x68
> 	Pid: 337, comm: jbd2/sda2-8 Not tainted 2.6.35-rc1jan+ #4
> 	Call Trace:
> 	[<ffffffff812362c5>] debug_smp_processor_id+0xc9/0xe4
> 	[<ffffffff8101059d>] native_sched_clock+0x3c/0x68
> 	[<ffffffff8101043d>] sched_clock+0x9/0xd
> 	[<ffffffff81212d7a>] blk_rq_init+0x97/0xa3
> 	[<ffffffff81214d71>] get_request+0x1c4/0x2d0
> 	[<ffffffff81214ea6>] get_request_wait+0x29/0x1a6
> 	[<ffffffff81215537>] __make_request+0x338/0x45b
> 	[<ffffffff812147c2>] generic_make_request+0x2bb/0x330
> 	[<ffffffff81214909>] submit_bio+0xd2/0xef
> 	[<ffffffff811413cb>] submit_bh+0xf4/0x116
> 	[<ffffffff81144853>] block_write_full_page_endio+0x89/0x96
> 	[<ffffffff81144875>] block_write_full_page+0x15/0x17
> 	[<ffffffff8119b00a>] ext4_writepage+0x356/0x36b
> 	[<ffffffff810e1f91>] __writepage+0x1a/0x39
> 	[<ffffffff810e32a6>] write_cache_pages+0x20d/0x346
> 	[<ffffffff810e3406>] generic_writepages+0x27/0x29
> 	[<ffffffff811ca279>] journal_submit_data_buffers+0x110/0x17d
> 	[<ffffffff811ca986>] jbd2_journal_commit_transaction+0x4cb/0x156d
> 	[<ffffffff811d0cba>] kjournald2+0x147/0x37a
> 
> (from the bugzilla thing)

The warning was introduced by this fresh commit (and a followup commit) merged 
in the .35 merge window:

 | commit 9195291e5f05e01d67f9a09c756b8aca8f009089
 | Author: Divyesh Shah <dpshah@xxxxxxxxxx>
 | Date:   Thu Apr 1 15:01:41 2010 -0700
 | 
 |     blkio: Increment the blkio cgroup stats for real now

IIRC Jens posted a fix for the regression. Jens, what's the status of that?

As the code there started using a raw sched_clock() call for block statistics 
purposes, which was a poorly thought out (and buggy) approach:

 - it takes timestamps on different cpus and then compares then, but doesnt 
   consider that sched_clock() is not comparable between CPUs without extra 
   care

 - it doesnt consider the possibility for the sched_clock() result going
   backwards on certain platforms (such as x86)

 - it doesnt consider preemptability

(There's work ongoing to add a clock variant that can be used for such 
purposes, but that's .36 fodder.)

Thanks,

	Ingo
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/