kmemleak complaints in nvme PM

Andy Lutomirski luto at kernel.org
Thu Apr 20 09:24:43 PDT 2017


On Thu, Apr 20, 2017 at 4:34 AM, Sagi Grimberg <sagi at grimberg.me> wrote:
>
>>>> Greg and/or Rafael, what's the right way to fix this?  I would imagine
>>>> that the driver core should free pm_qos requests when the device goes
>>>> away, no?
>>>
>>>
>>> Yes, it should.
>>>
>>
>> I don't see the bug.  dev_pm_qos_constraints_destroy() should take care of
>> it.
>
>
> I see it all the time, where is dev_pm_qos_constraints_destroy() called?
>
> Not in nvme.

What I meant was: I don't see the bug in the code.

dev_pm_qos_constraints_destroy() is called from dpm_sysfs_remove(),
which is called from device_del().

Anyway, I tried to reproduce it and I failed:

# nvmetcli
/> cd /ports/
/ports> create 1
module 'nvmet' has no attribute 'Port'

I can manually mkdir '1' in /sys/kernel/config/nvmet/ports, but I have
no idea what I'm supposed to do from there.

Help?  Is there some trivial script I can run to reproduce the problem?



More information about the Linux-nvme mailing list