Re: [Qemu-devel] [PATCH] rdma: Fix qemu crash when IPv6 address is used for migration

[Dr. David Alan Gilbert]

* Juan Quintela (address@hidden) wrote:
> Padmanabh Ratnakar <address@hidden> wrote:
> > Qemu crashes when IPv6 address is specified for migration and access
> > to any RDMA uverbs device available on the system is blocked using cgroups.
> > Fix the crash by checking the return value of ibv_open_device routine.
> >
> > Signed-off-by: Meghana Cheripady <address@hidden>
> > Signed-off-by: Padmanabh Ratnakar <address@hidden>
> > ---
> >  migration/rdma.c |    7 +++++++
> >  1 files changed, 7 insertions(+), 0 deletions(-)
> >
> > diff --git a/migration/rdma.c b/migration/rdma.c
> > index 77e3444..3671903 100644
> > --- a/migration/rdma.c
> > +++ b/migration/rdma.c
> > @@ -790,6 +790,13 @@ static int qemu_rdma_broken_ipv6_kernel(Error **errp, 
> > struct ibv_context *verbs)
> >  
> >          for (x = 0; x < num_devices; x++) {
> >              verbs = ibv_open_device(dev_list[x]);
> > +            if (!verbs) {
> > +                if (errno == EPERM) {
> > +                    continue;
> 
> Why do we want to continue here?

I think the idea is to loop through the set of devices and find if any of them
are 'roce';  if one of the devices iwe don't have permission to use then it's 
probably
ok to skip and move onto the next one?

> > +                } else {
> > +                    return -EINVAL;
> > +                }
> > +            }
> >  
> >              if (ibv_query_port(verbs, 1, &port_attr)) {
> 
> 
> Reading the documentation, my understandig is that if verbs is NULL, we
> don't have a context to do anything useful here, no?

But that's OK because of the if (!verbs) above?

> >                  ibv_close_device(verbs);
> 
> 
> Thanks, Juan.

Dave

> 
--
Dr. David Alan Gilbert / address@hidden / Manchester, UK