Re: BUG: NULL pointer dereference in shmem_evict_inode()

[John Stultz <john.stultz@linaro.org>]

On 08/20/2012 06:31 PM, Fengguang Wu wrote:
> On Mon, Aug 20, 2012 at 06:10:57PM -0700, John Stultz wrote:
>> On 08/20/2012 06:04 PM, Fengguang Wu wrote:
>>> Hi John,
>>>
>>> The below oops happens in v3.5..v3.6-rc2 and it's bisected down to commit
>>> 2a8c0883c ("time: Move xtime_nsec adjustment underflow handling timekeeping_adjust").
>>>
>>> However linux-next is working fine. Do you have any fixes not yet sent to Linus?
>> Yea, there's a fix pending in tip/timers/urgent
>> (4e8b14526ca7fb046a81c94002c1c43b6fdf0e9b) to catch crazy values
>> from settimeofday or the cmos clock that might overflow a ktime_t.
> That's great!
>
>> Out of curiosity, how are you triggering/reproducing this?
> I boot test lots of randconfig kernels in kvm, and this oops shows up
> several times in one ranconfig and some of the test boxes. I find it
> pretty hard to reproduce, but managed to bisect it down by counting
> 1000 good boots as bisect success and running dozens of KVM instances
> in parallel in several test boxes to speed up the progress. Here is one step:

Oof.  That's an really impressive setup!

That said, if this happens only at boot up, and you don't have systems 
with crazy cmos values, I'm not sure I see how commit 
4e8b14526ca7fb046a81c94002c1c43b6fdf0e9b might fix this.  So that's not 
very reassuring.

As a tangent, I think this sort of big-data style testing is a really 
great contribution, so thank you for setting up and doing all this work.
-john