Help debugging stable/10

[Hans Petter Selasky]

On 06/17/15 09:24, Shane Ambler wrote:
> On 24/03/2015 01:55, Shane Ambler wrote:
>> On 25/12/2014 23:03, Andriy Gapon wrote:
>>> On 25/12/2014 11:29, Hans Petter Selasky wrote:
>>>> The cam_sim_free() is stuck, blocking the rest of that controller from
>>>> enumerating. It might look like a non-USB stack issue.
>>>>
>>>> MAV: Do you have some ideas where to start looking, now we have a
>>>> dump? Any
>>>> refcounts to check in particular?
>>>
>>> Apparently sim->refcount > 0.
>>> Not sure how to check who has the reference(s).
>>>
>
> Can anyone think of something I can try?
>
> To recap to save you going back through history -
> After running 9.0 - 9.2 for 3 years I upgraded to 10.1RC3 and started
> getting a locking issue, most new processes fail to start, top and ps
> failing being indicators, the most info I have got is a back trace
> using kgdb, there are 4 instances I got output from procstat -kk -a
>
> On several occasions I have found that after inserting a usb memstick
> the device failed to be created, leaving me unable to mount the
> filesystem without a restart.
>
> I then switched to stable/10 in hopes of a fix finding it's way in.
>
> The back traces I have been able to collect (and a dmesg) are listed at
> http://shaneware.biz/freebsddebugdata/
>
> This is my everyday desktop machine. I am now running
>
> FreeBSD leader.local 10.1-STABLE FreeBSD 10.1-STABLE #11 r283839: Thu
> Jun  4 17:41:28 ACST 2015     root at leader.local:/usr/obj/usr/src/sys
> /GENERIC  amd64
>
> I can only say it appears to be getting worse, though I may just be
> getting sick of having to restart nearly every day. Lately it seems that
> the less I do the quicker it locks up. I have restarted twice this week
> and then let it sit while I have gone out, after returning I get maybe
> 10-15 mins then have to restart, one of them was less than an hour uptime.
>
> While running poudriere I have got past 1 day uptime but it locks up
> harder and I don't usually get a chance to record any data.
>
>

Hi,

One solution is to use fuse instead of the native fs, until the CAM/SCSI 
refcount issues are resolved.

--HPS