linux-kernel - oprofile can cause an NMI to schedule (was: [RT] scheduling and oprofile)

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-Id: <1161656674.13276.17.camel@localhost.localdomain>
Date:	Mon, 23 Oct 2006 22:24:34 -0400
From:	Steven Rostedt <rostedt@...dmis.org>
To:	Mike Kravetz <kravetz@...ibm.com>
Cc:	linux-kernel@...r.kernel.org, phil.el@...adoo.fr,
	oprofile-list@...ts.sourceforge.net, Ingo Molnar <mingo@...e.hu>
Subject: oprofile can cause an NMI to schedule (was: [RT] scheduling and
	oprofile)

On Mon, 2006-10-23 at 14:23 -0700, Mike Kravetz wrote:
> I've been trying to use oprofile on an RT kernel to look at some
> performance issues.  While running I notice the following sent to
> the console:
> 
> BUG: scheduling with irqs disabled: java/0x00000000/4521
> caller is rt_mutex_slowlock+0x156/0x1dd
>  [<c032051a>] schedule+0x65/0xd2 (8)
>  [<c0321338>] rt_mutex_slowlock+0x156/0x1dd (12)
>  [<c032142a>] rt_mutex_lock+0x24/0x28 (72)
>  [<c0134904>] rt_down_read+0x38/0x3b (20)
>  [<c0322a89>] do_page_fault+0xe3/0x52d (12)
>  [<c03229a6>] do_page_fault+0x0/0x52d (76)
>  [<c01033bb>] error_code+0x4f/0x54 (8)
>  [<c01ce6d0>] __copy_from_user_ll+0x55/0x7c (44)
>  [<f89be7ef>] dump_user_backtrace+0x2e/0x56 [oprofile] (24)
>  [<c0134869>] rt_up_read+0x3e/0x41 (20)
>  [<f89be864>] x86_backtrace+0x4a/0x5a [oprofile] (20)
>  [<f89bd53a>] oprofile_add_sample+0x73/0x89 [oprofile] (20)
>  [<f89beea3>] athlon_check_ctrs+0x22/0x4a [oprofile] (32)
>  [<f89be8c5>] nmi_callback+0x18/0x1b [oprofile] (28)
>  [<c01041ff>] do_nmi+0x24/0x33 (12)
>  [<c0103462>] nmi_stack_correct+0x1d/0x22 (16)
> 
> It seems strange to me that oprofile would be calling
> '__copy_from_user_ll' in this context.  I can see why the
> changes made for RT locking expose this.  But, doesn't this
> issue also exist on non-RT (default) kernels?  What happens
> when we generate a page fault in this context on non-RT kernels?
> 

As Mike has pointed out here, oprofile _can_ cause the nmi to schedule.
Here's the path: (looking at vanilla 2.6.18).

arch/i386/oprofile/nmi_int.c: nmi_callback

	return model->check_ctrs(regs, &cpu_msrs[cpu]);

if model == &op_athlon_spec
 (could be a problem with others, but I'm only looking here).
   
op_athlon_spec.check_ctrs =  &athlon_check_ctrs


Here's the calling path:

  athlon_check_ctrs

   ==> oprofile_add_sample

   ==> oprofile_add_ext_sample

   ==> oprofile_ops.backtrace
           == x86_backtrace

   ==> dump_user_backtrace

   ==> __copy_from_user_inatomic

   Don't let the name fool you, this _can_ schedule! (and says so in the
comments above it).

Now perhaps on a vanilla kernel opfile_add_ext_sample is not likely to
have log_sample fail. I don't know, but this path exits, so we can
indeed schedule in a NMI interrupt.

Mike, thanks for pointing this out.

-- Steve



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