Re: [rdma-next v3 02/24] IB/uverbs: Introduce and use helper functions to copy ah attributes

[Leon Romanovsky <leon-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org>]

[-- Attachment #1: Type: text/plain, Size: 5026 bytes --]

On Mon, Aug 28, 2017 at 12:32:04PM -0400, Doug Ledford wrote:
> On Mon, 2017-08-28 at 18:55 +0300, Leon Romanovsky wrote:
> > On Mon, Aug 28, 2017 at 11:15:14AM -0400, Doug Ledford wrote:
> > > On Sun, 2017-08-27 at 14:10 +0300, Leon Romanovsky wrote:
> > > > On Thu, Aug 24, 2017 at 04:40:42PM -0400, Doug Ledford wrote:
> > > > > On Thu, 2017-08-17 at 15:50 +0300, Leon Romanovsky wrote:
> > > > > > From: Parav Pandit <parav-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
> > > > > >
> > > > > > This patch introduces two helper functions to copy ah
> > > > > > attributes
> > > > > > from uverbs to internal ib_ah_attr structure and the other
> > > > > > way
> > > > > > during modify qp and query qp respectively.
> > > > > >
> > > > > > Signed-off-by: Parav Pandit <parav-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
> > > > > > Reviewed-by: Daniel Jurgens <danielj-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>
> > > > > > Signed-off-by: Leon Romanovsky <leon-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org>
> > > > > > Reviewed-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org
> > > > > > >
> > > > > > ---
> > > > > >  drivers/infiniband/core/uverbs_cmd.c | 124 ++++++++++++++---
> > > > > > ----
> > > > > > ----
> > > > > > ----------
> > > > > >  1 file changed, 49 insertions(+), 75 deletions(-)
> > > > >
> > > > > I had to fix this patch up considerably...
> > > > >
> > > > > > diff --git a/drivers/infiniband/core/uverbs_cmd.c
> > > > > > b/drivers/infiniband/core/uverbs_cmd.c
> > > > > > index 670176b670a0..515425a50059 100644
> > > > > > --- a/drivers/infiniband/core/uverbs_cmd.c
> > > > > > +++ b/drivers/infiniband/core/uverbs_cmd.c
> > > > > >
> > > > > > @@ -1936,6 +1926,28 @@ static int modify_qp_mask(enum
> > > > > > ib_qp_type
> > > > > > qp_type, int mask)
> > > > > >  	}
> > > > > >  }
> > > > > >
> > > > > > +static void copy_ah_attr_from_uverbs(struct ib_device *dev,
> > >
> > >                                              ^^^^^^^^^^^^^^^^^^^^
> > > From your patch...
> > >
> > > > > > +				     struct rdma_ah_attr
> > > > > > *rdma_attr,
> > > > > > +				     struct
> > > > > > ib_uverbs_qp_dest
> > > > > >
> > > > >
> > > > > And we had a fix here where we only copy the ah attributes if
> > > > > the
> > > > > user
> > > > > set the flag for it, so what I ended up with looks like this:
> > > > >
> > > > >         if (cmd->base.attr_mask & IB_QP_AV)
> > > > >                 copy_ah_attr_from_uverbs(qp->device, &attr-
> > > > > > ah_attr,
> > > > >
> > > > >                                          &cmd->base.dest);
> > > > >
> > > > >         if (cmd->base.attr_mask & IB_QP_ALT_PATH)
> > > > >                 copy_ah_attr_from_uverbs(qp->device, &attr-
> > > > > > alt_ah_attr,
> > > > >
> > > > >                                          &cmd->base.alt_dest);
> > > >
> > > > Doug,
> > > >
> > > > It should be "qp" and not "qp->device" in
> > > > copy_ah_attr_from_uverbs().
> > > >
> > > > 1998         if (cmd->base.attr_mask & IB_QP_AV)
> > > > 1999                 copy_ah_attr_from_uverbs(qp, &attr->ah_attr,
> > > > &cmd->base.dest);
> > > > 2000         if (cmd->base.attr_mask & IB_QP_ALT_PATH)
> > > > 2001                 copy_ah_attr_from_uverbs(qp, &attr-
> > > > >alt_ah_attr,
> > > > 2002                                          &cmd-
> > > > >base.alt_dest);
> > > > 2003         ret = ib_modify_qp_with_udata(qp, attr,
> > > > 2004                                       modify_qp_mask(qp-
> > > > > qp_type,
> > > >
> > > > 2005                                                      cmd-
> > > > > base.attr_mask),
> > > >
> > > > 2006                                       udata);
> > >
> > > I'm pretty sure you need to re-read your own patch Leon ;-).  The
> > > helper you added uses struct ib_device, so a pointer to the qp
> > > would
> > > not work.  And it wouldn't have even compiled if I had that wrong.
> >
> > OK, I reread it and it helped me to find the differences.
> >
> > In rdma-rc, there is the patch from Noa "IB/core: Avoid accessing
> > non-allocated memory when inferring port type", which plays exactly
> > at
> > the same place as this one.
> >
> > Because, I'm creating my rdma-next for testing, I'm merging -rcX with
> > -next and my merge conflict resolution was to move
> > +               attr->ah_attr.type = rdma_ah_find_type(qp->device,
> > +                                                      cmd-
> > >base.dest.port_num);
> > to the copy_ah_attr_from_uverbs and to change function signature to
> > be
> > "static void copy_ah_attr_from_uverbs(struct ib_qp *qp,"
>
> It turns out this is kind of a moot point.  In uverbs_marshall.c there
> already exists void ib_copy_ah_attr_to_user().  We should have used
> that instead of creating new versions of this in this file.

I don't see clean way to reuse those functions.

The ib_copy_qp_attr_to_user has second parameter of "struct
ib_uverbs_qp_attr", while ib_uverbs_query_qp returns results in "struct
ib_uverbs_query_qp_resp" and operates on it.

Thanks

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]