Re: [EXTERNAL] Re: [Patch v7 5/5] RDMA/mana_ib: Send event to qp

[Jason Gunthorpe <jgg@ziepe.ca>]

On Fri, Oct 27, 2023 at 09:35:05PM +0000, Long Li wrote:
> > Subject: RE: [EXTERNAL] Re: [Patch v7 5/5] RDMA/mana_ib: Send event to qp
> > 
> > 
> > > -----Original Message-----
> > > From: Jason Gunthorpe <jgg@ziepe.ca>
> > > Sent: Monday, October 23, 2023 11:24 AM
> > > To: sharmaajay@linuxonhyperv.com
> > > Cc: Long Li <longli@microsoft.com>; Leon Romanovsky <leon@kernel.org>;
> > > Dexuan Cui <decui@microsoft.com>; Wei Liu <wei.liu@kernel.org>; David S.
> > > Miller <davem@davemloft.net>; Eric Dumazet <edumazet@google.com>;
> > > Jakub Kicinski <kuba@kernel.org>; Paolo Abeni <pabeni@redhat.com>;
> > > linux- rdma@vger.kernel.org; linux-hyperv@vger.kernel.org;
> > > netdev@vger.kernel.org; linux-kernel@vger.kernel.org; Ajay Sharma
> > > <sharmaajay@microsoft.com>
> > > Subject: [EXTERNAL] Re: [Patch v7 5/5] RDMA/mana_ib: Send event to qp
> > >
> > > On Mon, Oct 16, 2023 at 03:12:02PM -0700,
> > sharmaajay@linuxonhyperv.com
> > > wrote:
> > >
> > > > diff --git a/drivers/infiniband/hw/mana/qp.c
> > > > b/drivers/infiniband/hw/mana/qp.c index ef3275ac92a0..19fae28985c3
> > > > 100644
> > > > --- a/drivers/infiniband/hw/mana/qp.c
> > > > +++ b/drivers/infiniband/hw/mana/qp.c
> > > > @@ -210,6 +210,8 @@ static int mana_ib_create_qp_rss(struct ib_qp
> > > *ibqp, struct ib_pd *pd,
> > > >  		wq->id = wq_spec.queue_index;
> > > >  		cq->id = cq_spec.queue_index;
> > > >
> > > > +		xa_store(&mib_dev->rq_to_qp_lookup_table, wq->id, qp,
> > > GFP_KERNEL);
> > > > +
> > >
> > > A store with no erase?
> > >
> > > A load with no locking?
> > >
> > > This can't be right
> > >
> > > Jason
> > 
> > This wq->id is assigned from the HW and is guaranteed to be unique. May be I
> > am not following why do we need a lock here. Can you please explain ?
> > Ajay
> 
> I think we need to check the return value of xa_store(), and call xa_erase() in mana_ib_destroy_qp().
> 
> wq->id is generated by the hardware. If we believe in hardware
> always behaves in good manner, we don't need a lock.

It has nothing to do with how the ID is created, you need to explain
how the missing erase can't race with any loads, in a comment above
the unlocked xa_load.

Jason