linux-kernel - Re: [PATCH] rcu: Fix soft lockup for rcu_nocb

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 for Android: free password hash cracker in your pocket

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

Date:   Sat, 19 Nov 2016 15:17:37 +0800
From:   Ding Tianhong <dingtianhong@...wei.com>
To:     <paulmck@...ux.vnet.ibm.com>
CC:     <josh@...htriplett.org>, <rostedt@...dmis.org>,
        <mathieu.desnoyers@...icios.com>, <jiangshanlai@...il.com>,
        "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH] rcu: Fix soft lockup for rcu_nocb_kthread



On 2016/11/18 20:56, Paul E. McKenney wrote:
> On Fri, Nov 18, 2016 at 08:37:28PM +0800, Ding Tianhong wrote:
>>
>>
>> On 2016/8/10 9:59, Paul E. McKenney wrote:
>>> On Wed, Aug 10, 2016 at 09:13:14AM +0800, Ding Tianhong wrote:
>>>> On 2016/6/16 22:19, Paul E. McKenney wrote:
>>>>> On Thu, Jun 16, 2016 at 02:09:47PM +0800, Ding Tianhong wrote:
>>>>>> On 2016/6/15 23:49, Paul E. McKenney wrote:
>>>>>>> On Wed, Jun 15, 2016 at 03:27:36PM +0800, Ding Tianhong wrote:
>>>>>>>> I met this problem when using the Testgine to send package to ixgbevf nic
>>>>>>>> by this steps:
>>>>>>>> 1. Connect to ixgbevf, and set the speed to 10Gb/s, it could work fine.
>>>>>>>> 2. Then use ifconfig to down the nic and up again, loop for several times.
>>>>>>>> 3. The system panic by soft lockup.
>>>>>>>
>>>>>>> Good catch, queued for review and testing.  But what .config was your
>>>>>>> kernel built with?
>>>>>>>
>>>>>>
>>>>>> I use the redhat7.1 defconfig to build my kernel, and the RCU config is this:
>>>>>>  120 #
>>>>>>  121 # RCU Subsystem
>>>>>>  122 #
>>>>>>  123 CONFIG_TREE_RCU=y
>>>>>>  124 # CONFIG_PREEMPT_RCU is not set
>>>>>>  125 CONFIG_RCU_STALL_COMMON=y
>>>>>>  126 CONFIG_CONTEXT_TRACKING=y
>>>>>>  127 CONFIG_RCU_USER_QS=y
>>>>>>  128 # CONFIG_CONTEXT_TRACKING_FORCE is not set
>>>>>>  129 CONFIG_RCU_FANOUT=64
>>>>>>  130 CONFIG_RCU_FANOUT_LEAF=16
>>>>>>  131 # CONFIG_RCU_FANOUT_EXACT is not set
>>>>>>  132 # CONFIG_RCU_FAST_NO_HZ is not set
>>>>>>  133 # CONFIG_TREE_RCU_TRACE is not set
>>>>>>  134 CONFIG_RCU_NOCB_CPU=y
>>>>>>  135 CONFIG_RCU_NOCB_CPU_ALL=y
>>>>>>  136 CONFIG_BUILD_BIN2C=y
>>>>>
>>>>> Thank you!  You were running with preemption disabled, so your system
>>>>> would indeed be very susceptible to this problem.
>>>>>
>>>>>>> Also, I did tweak both the commit log and the patch.  Your cond_resched()
>>>>>>> would prevent soft lockups, but not RCU stalls, so I substituted
>>>>>>> cond_resched_rcu_qs().  Please let me know if either of those changes
>>>>>>> causes problems at your end.
>>>>>>
>>>>>> Looks fine to me, I will apply this to my branch and test it, thanks.
>>>>>
>>>>> Please let me know how it goes!
>>>>>
>>>>> 							Thanx, Paul
>>>>>
>>>>
>>>> Hi Paul:
>>>>
>>>> It has been a long time after applying this patch, and didn't found any problem, I believe this patch is fine, thanks.
>>>
>>> Very good!  I will push this one upstream during the next merge window.
>>>
>>> 							Thanx, Paul
>>>
>>
>> Hi Paul:
>>
>> Sorry to say that I have found this patch will introduce an OOM problem, it will be triggered by huge IP abnormal packet
>> arrived, it looks that avoid process any pending softirqs in the rcuos kthread is the best way to fix this problem, I will
>> send a new patch to revert this and fix the problem.
> 
> Interesting...
> 
> Could you please let me know exactly how the added cond_resched_rcu_qs()
> leads to an OOM?  Is it that the softirqs prevent the grace-period kthread
> from making progress?
> 

Ok, reply and discuss on other patch, thanks.

Ding

> 							Thanx, Paul
> 
> 
> .
>