On Wed, Feb 21, 2018 at 11:26:34AM -0700, Jason Gunthorpe wrote:
> On Wed, Feb 21, 2018 at 08:18:27PM +0200, Leon Romanovsky wrote:
>
> > > > + } else {
> > > > + /* It is done in _ib_create_qp for other QP types */
> > > > + qp->uobject = &obj->uevent.uobject;
> > > > }
> > >
> > > Steve got rid of this line to make sure the uobject was init'd during
> > > restrack add, so why is adding it back here OK for XRC?
> >
> > Steve got rid of this line to determine kernel vs. user QPs and not
> > because of being XRC. XRC QPs are not participating in restrack
> > and IB_QPT_XRC_TGT actually needs this qp->object to be initialized.
>
> But can we not set it someplace where it doesn't break the restrack
> assumption even though we don't care today?
It is exactly the right place, you can see the following code above:
1520 if (cmd->qp_type == IB_QPT_XRC_TGT)
1521 qp = ib_create_qp(pd, &attr);
1522 else
1523 qp = _ib_create_qp(device, pd, &attr, uhw,
1524 &obj->uevent.uobject);
So it has nothing to do with restrack assumption, but with the fact that
IB_QPT_XRC_TGT QPs are created with ib_create_qp() which is kernel
version and it doesn't have &obj->uevent.uobject.
This area (QP creation) worth cleanup, but this patch is good for now
and fixes real user triggered oops, which is now in Linus's master too.
Thanks
>
> Jason
> --
> To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
> the body of a message to majordomo@xxxxxxxxxxxxxxx
> More majordomo info at http://vger.kernel.org/majordomo-info.html
Attachment:
signature.asc
Description: PGP signature
