svn commit: r331606 - in head/sys: amd64/include i386/include x86/x86 x86/xen

[John Baldwin]

On Tuesday, March 27, 2018 11:07:22 AM Jeff Roberson wrote:
> On Tue, 27 Mar 2018, Li-Wen Hsu wrote:
> 
> > On Mon, Mar 26, 2018 at 20:35:12 -1000, Jeff Roberson wrote:
> >> The patch has been on my branch for weeks and has been tested by a half
> >> dozen people.  I'm sorry it does not work for you.  If you reverted 331605
> >> the change that followed should not have built properly.  Did you build
> >> cleanly?  Can you share your kernel config?
> >>
> >> I tried with and without EARLY_AP_STARTUP and with and without NUMA.  I'm
> >> not having any trouble booting.  Did you make cleandepend && make depend?
> >
> > It also hangs in our testing system:
> >
> > https://ci.freebsd.org/job/FreeBSD-head-amd64-test/6817/console
> > https://ci.freebsd.org/job/FreeBSD-head-i386-test/1000/console
> >
> > It is built cleanly and uses unmodified GENERIC config.
> >
> > The artifacts used are available here:
> >
> > https://artifact.ci.freebsd.org/snapshot/head/r331606/amd64/amd64/
> > https://artifact.ci.freebsd.org/snapshot/head/r331606/i386/i386/
> >
> > Hope these information help.
> 
> Could someone who was experiencing the hang try the enclosed patch?  I can 
> only verify that it continues to boot for me but I believe this fixes the 
> bug on other systems.
> 
> I believe the issue was that cpuset_domain[0] was initialized too late on 
> some systems.  It depended on what kind of hardware was present and which 
> sysinit with SI_ORDER_ANY ran first.

Hmm, does this work if you don't use EARLY_AP_STARTUP?  I think all_cpus
isn't set yet (or only contains the BSP) when cpuset_thread0 is run if you
don't have EARLY_AP_STARTUP (and EARLY_AP_STARTUP is only supported on
x86 currently)

-- 
John Baldwin