On Wed, Nov 09, 2022 at 03:08:33PM -0400, Jason Gunthorpe wrote:
> On Mon, Nov 07, 2022 at 10:51:34AM +0200, Leon Romanovsky wrote:
> > From: Mark Zhang <markzhang@xxxxxxxxxx>
> >
> > The MR restrack also needs to be released when delete it, otherwise it
> > cause memory leak as the task struct won't be released.
> >
> > Fixes: 13ef5539def7 ("RDMA/restrack: Count references to the verbs objects")
> > Signed-off-by: Mark Zhang <markzhang@xxxxxxxxxx>
> > Reviewed-by: Michael Guralnik <michaelgur@xxxxxxxxxx>
> > Signed-off-by: Leon Romanovsky <leonro@xxxxxxxxxx>
> > ---
> > drivers/infiniband/core/restrack.c | 2 --
> > 1 file changed, 2 deletions(-)
> >
> > diff --git a/drivers/infiniband/core/restrack.c b/drivers/infiniband/core/restrack.c
> > index 1f935d9f6178..01a499a8b88d 100644
> > --- a/drivers/infiniband/core/restrack.c
> > +++ b/drivers/infiniband/core/restrack.c
> > @@ -343,8 +343,6 @@ void rdma_restrack_del(struct rdma_restrack_entry *res)
> > rt = &dev->res[res->type];
> >
> > old = xa_erase(&rt->xa, res->id);
> > - if (res->type == RDMA_RESTRACK_MR)
> > - return;
>
> This needs more explanation, there was some good reason we needed to
> avoid the wait_for_completion() for the driver allocated objects, but I
> can't remember it anymore.
>
> You added this code in the v2 of the original series, maybe it had
> something to do with mlx4?
I failed to remember either, but if you want even more magic in your life,
see this hilarious thread:
https://lore.kernel.org/linux-rdma/9ba5a611ceac86774d3d0fda12704cecc30606f9.1618753038.git.leonro@xxxxxxxxxx/
I kept this patch in my queue ~1 month and didn't get any complains.
Thanks
>
> Jason
