Re: backport requests for 4.x-testing

[Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>]

> >> > Applied 23225 and 24013. The other, toolstack-related, patches I will leave
> >> > for a tools maintainer to ack or apply.
> >>
> > Hey Teck,
> >
> > Thanks for reporting!
> >
> >> With the two backport patches committed in xen-4.1-testing (changeset
> >> 23271:13741fd6253b), xl list or xl create domU will cause 100% CPU and
> >
> > xl list?
> 
> After a reboot with no domU running, xl list is fine but if I start a
> hvm domU will be stuck and caused high load then open another ssh
> terminal to issue xl list will stuck as well.

This fix fixes it for me:

diff -r 13741fd6253b xen/arch/x86/domain.c
--- a/xen/arch/x86/domain.c	Thu Mar 29 10:20:58 2012 +0100
+++ b/xen/arch/x86/domain.c	Thu Mar 29 11:44:54 2012 -0400
@@ -558,9 +558,9 @@ int arch_domain_create(struct domain *d,
         d->arch.is_32bit_pv = d->arch.has_32bit_shinfo =
             (CONFIG_PAGING_LEVELS != 4);
 
-        spin_lock_init(&d->arch.e820_lock);
     }
 
+    spin_lock_init(&d->arch.e820_lock);
     memset(d->arch.cpuids, 0, sizeof(d->arch.cpuids));
     for ( i = 0; i < MAX_CPUID_INPUT; i++ )
     {
@@ -605,8 +605,8 @@ void arch_domain_destroy(struct domain *
 
     if ( is_hvm_domain(d) )
         hvm_domain_destroy(d);
-    else
-        xfree(d->arch.e820);
+
+    xfree(d->arch.e820);
 
     vmce_destroy_msr(d);
     free_domain_pirqs(d);


The issue is that upstream we have two 'domain structs' - one for PV and
one for HVM. In 4.1 it is just 'arch_domain' and the calls to create
the guests are going through the same interface (at least using xl, with
xm they are seperate). And I only initialized the spinlock in the PV case,
but not in the HVM case. This fix to the backport resolves the problem.

Keir, please apply this to my botched back-port of  23225.