Re: [Xen-devel] PVops domain 0 crash on NUMA system only Node==1 present (Was: Re: Bug#603632: linux-image-2.6.32-5-xen-amd64: Linux kernel 2.6.32/xen/amd64 booting fine on bare metal, but not as dom0 with Xen 4.0.1 (Dell R410))

[Ian Campbell <Ian.Campbell@xxxxxxxxxxxxx>]

On Thu, 2010-12-02 at 23:47 +0000, Vincent Caron wrote:
> On Tue, 2010-11-23 at 22:18 +0000, Ian Campbell wrote:
> > On Tue, 2010-11-23 at 22:12 +0000, Ian Campbell wrote: 
> > > On Tue, 2010-11-23 at 18:52 +0000, Ian Campbell wrote: 
> > > > On Tue, 2010-11-23 at 11:51 +0000, Ian Campbell wrote:
> > > > > 
> > > > > Perhaps we should be passing numa_node_id() (e.g. current node)
> > > > > instead of node 0? 
> > > > 
> > > > I've just kicked off a build of the 2.6.32-27 Debian kernel with the
> > > > following additional patch, I will hopefully post the binaries tomorrow.
> > > 
> > > Build was quicker than I thought... Vincent, Cris if you get a chance
> > > please can you test the kernel from:
> > > http://xenbits.xen.org/people/ianc/2.6.32-27+numa1/
> > 
> > Also, please can you try adding "numa=noacpi" to your kernel command
> > line when running with the standard Debian kernel (not the one above).
> > 
> > Thanks!
> 
>   It just happens that your kernel above (2.6.32-27+numa1) boots fine
> under hypervisor _when_ passed 'numa=noacpi'. Yeah !
> 
>   I then tried again with Debian Squeeze's latest 2.6.32-28, which
> crashes as -27 under hypervisor (and changelog show no xen or
> numa-related thingies). Then I added 'numa=noacpi', and it boots fine
> too. I got my 8 cores, networking, etc.
> 
>   Enclosed is the dmesg for the latter, Debian, kernel.
> 
>   Is the 'numa=noacpi' a "production acceptable" workaround ?

Yes and in fact I think the actual fix is simply to have Xen fake out
the behaviour of numa=noacpi as below. I'll send this plus the other fix
out after I've given it a bit of proper testing.

Ian.

xen: disable ACPI NUMA for PV guests

Xen does not currently expose PV-NUMA information to PV
guests. Therefore disable NUMA for the time being.

Signed-off-by: Ian Campbell <ian.campbell@xxxxxxxxxx>
diff --git a/arch/x86/xen/enlighten.c b/arch/x86/xen/enlighten.c
index 02c710b..5c55e1b 100644
--- a/arch/x86/xen/enlighten.c
+++ b/arch/x86/xen/enlighten.c
@@ -1178,6 +1178,10 @@ asmlinkage void __init xen_start_kernel(void)
 
        xen_smp_init();
 
+#ifdef CONFIG_ACPI_NUMA
+       acpi_numa = -1;
+#endif
+
        pgd = (pgd_t *)xen_start_info->pt_base;
 
        if (!xen_initial_domain())



_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel