Re: taskqgroup_adjust kernel panic

On Mon, Sep 05, 2016 at 02:54:54PM -0700, Mark Johnston wrote:
> On Mon, Sep 05, 2016 at 01:55:38PM -0400, Shawn Webb wrote:
> > Hey all,
> > 
> > I'm at revision 3872750 of the hardened/current/drm-next-4.7 branch in
> > the HardenedBSD/hardenedBSD-playground repo. I've gotten this kernel
> > panic a couple times when booting. I'm using full-disk encryption with
> > ZFS and encrypted swap. The hardware is a Purism 15 2K laptop.
> > 
> > The panic doesn't happen often nor is there a way I can reproduce it
> > 100%.
> > 
> > Here's my `uname -a` output:
> > 
> > FreeBSD hbsd-dev-laptop 12.0-CURRENT-HBSD FreeBSD 12.0-CURRENT-HBSD #0 3872750(hardened/current/drm-next-4.7): Tue Aug 30 17:41:53 EDT 2016     shawn_at_hbsd-dev-laptop:/usr/obj/usr/src/sys/LATT-SEC  amd64
> > 
> > Here's a couple pictures of the panic I took:
> > 
> > https://goo.gl/photos/P5kiwabPYjwQX7Kr8
> > https://goo.gl/photos/BWtvBnq7QLnwgRP28
> 
> Based on the faulting instruction, the panic probably happened because
> qid is uninitialized in the loop that starts with
> 
>     while ((gtask = LIST_FIRST(&gtask_head))) {
> 
> I don't know this code very well, so I'm not sure how that can happen. I
> suspect iflib_irq_alloc_generic() is buggy: it calls
> 
>     taskqgroup_attach_cpu(... CPU_FFS(&cpus) ...);
> 
> and CPU_FFS returns 1-indexed IDs, but taskqgroup_attach_cpu() pretty
> clearly expects 0-indexed CPU IDs. There's a similar bug in find_nth()
> in iflib.c.

I think you hit the nail right on the head. Attached is a patch that
doesn't fix the underlying issue, but at least detects improperly
setting qid. It'll throw a KASSERT if qid isn't set properly.

I'll study this code a bit more within the next couple days and I hope
to have a full patch to address the underlying issue.

Thanks,

-- 
Shawn Webb
Cofounder and Security Engineer
HardenedBSD

GPG Key ID:          0x6A84658F52456EEE
GPG Key Fingerprint: 2ABA B6BD EF6A F486 BE89  3D9E 6A84 658F 5245 6EEE