linux-kernel - CPU softlockup due to smp_call

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: <smp-call-function-lockup-1@mdm.bga.com>
Date:	Wed, 04 Apr 2012 23:06:54 -0500
From:	Milton Miller <miltonm@....com>
To:	Sasha Levin <levinsasha928@...il.com>
Cc:	Peter Zijlstra <peterz@...radead.org>,
	Thomas Gleixner <tglx@...utronix.de>,
	Ingo Molnar <mingo@...e.hu>, "H. Peter Anvin" <hpa@...or.com>,
	Avi Kivity <avi@...hat.com>, Dave Jones <davej@...hat.com>,
	<kvm@...r.kernel.org>,
	"linux-kernel@...r.kernel.org List" <linux-kernel@...r.kernel.org>
Subject: CPU softlockup due to smp_call_function()

On Wed, 4 Apr 2012 about 22:12:36 +0200, Sasha Levin wrote:
> I've starting seeing soft lockups resulting from smp_call_function()
> calls. I've attached two different backtraces of this happening with
> different code paths.
> 
> This is running inside a KVM guest with the trinity fuzzer, using
> today's linux-next kernel.

Hi Sasha.

You have two different call sites (arch/x86/mm/pageattr.c
cpa_flush_range and net/core/dev.c netdev_run_todo), and both use
call on_each_cpu with wait=1.

I tried a few options but can't get close enough to your compiled
length of 2a0 to know if the code is spinning on the first
csd_lock_wait in csd_lock or in the second csd_lock_wait after the
call to arch_send_call_function_ipi_mask (aka smp_ops + 0x44 in my
x86_64 compile).  Please check your disassembly and report.

If its the first lock, then the current stack is an innocent victim.
In either case we need to find what the cpu(s) holding up the reporting
cpus call function data (cfd_data per_cpu var) is(are) doing.

Since interrupts are on, we could read the time at entry (even jiffies)
and report both the function and mask of cpus that have not processed
the cpus entry if the elapsed time has exceeded some threshold.

I described the call flow of smp_call_function_many and outlined some
debug sanity checks that could be added at [1] if you suspect the function
list is getting corrupted.

Let me know if you need help creating this debug code.

[1] https://lkml.org/lkml/2012/1/13/308

milton
--
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/