Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
MIME-Version: 1.0
References: <000000000000a589d005fc52ee2d@google.com> <13528f21-0f36-4fa2-d34f-eecee6720bc1@linux.dev>
 <CAD=hENeCo=-Pk9TWnqxOWP9Pg-JXWk6n6J19gvPo9_h7drROGg@mail.gmail.com>
 <CAD=hENdoyBZaRz7aTy4mX5Kq1OYmWabx2vx8vPH0gQfHO1grzw@mail.gmail.com>
 <0d515e17-5386-61ba-8278-500620969497@linux.dev> <CAD=hENcqa0jQvLjuXw9bMtivCkKpQ9=1e0-y-1oxL23OLjutuw@mail.gmail.com>
 <CAD=hENdXdqfcxjNrNnP8CoaDy6sUJ4g5uxcWE0mj3HtNohDUzw@mail.gmail.com>
 <CAD=hENda4MxgEsgT-GUhYHH66m79wi8yxBQS8CYnxc_DsQKGwg@mail.gmail.com>
 <5b6b8431-92c7-62df-299b-28f3a5f61d5f@linux.dev> <CAD=hENc72B+gLLd_Xn7w8bd_qDw=mFd5sC0RKEsHpNA=85a9KA@mail.gmail.com>
 <e2a3a27e-9c12-f180-4bb6-1906aa1a1844@linux.dev>
In-Reply-To: <e2a3a27e-9c12-f180-4bb6-1906aa1a1844@linux.dev>
From:   Zhu Yanjun <zyjzyj2000@gmail.com>
Date:   Tue, 23 May 2023 14:04:49 +0800
Message-ID: <CAD=hENcMLEfLgjoRqxwwR1jdAKTHfB+iEBtjpfkfhzwsJoLVow@mail.gmail.com>
Subject: Re: [syzbot] [rdma?] INFO: trying to register non-static key in
 skb_dequeue (2)
To:     Guoqing Jiang <guoqing.jiang@linux.dev>
Cc:     syzbot <syzbot+eba589d8f49c73d356da@syzkaller.appspotmail.com>,
        jgg@ziepe.ca, leon@kernel.org, linux-kernel@vger.kernel.org,
        linux-rdma@vger.kernel.org, netdev@vger.kernel.org,
        syzkaller-bugs@googlegroups.com
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

On Tue, May 23, 2023 at 1:56=E2=80=AFPM Guoqing Jiang <guoqing.jiang@linux.=
dev> wrote:
>
>
>
> On 5/23/23 13:52, Zhu Yanjun wrote:
> > On Tue, May 23, 2023 at 1:44=E2=80=AFPM Guoqing Jiang <guoqing.jiang@li=
nux.dev> wrote:
> >>
> >>
> >> On 5/23/23 13:18, Zhu Yanjun wrote:
> >>> On Tue, May 23, 2023 at 1:08=E2=80=AFPM Zhu Yanjun <zyjzyj2000@gmail.=
com> wrote:
> >>>> On Tue, May 23, 2023 at 12:29=E2=80=AFPM Zhu Yanjun <zyjzyj2000@gmai=
l.com> wrote:
> >>>>> On Tue, May 23, 2023 at 12:10=E2=80=AFPM Guoqing Jiang <guoqing.jia=
ng@linux.dev> wrote:
> >>>>>>
> >>>>>> On 5/23/23 12:02, Zhu Yanjun wrote:
> >>>>>>> On Tue, May 23, 2023 at 11:47=E2=80=AFAM Zhu Yanjun <zyjzyj2000@g=
mail.com> wrote:
> >>>>>>>> On Tue, May 23, 2023 at 10:26=E2=80=AFAM Guoqing Jiang <guoqing.=
jiang@linux.dev> wrote:
> >>>>>>>>> On 5/23/23 10:13, syzbot wrote:
> >>>>>>>>>> Hello,
> >>>>>>>>>>
> >>>>>>>>>> syzbot tried to test the proposed patch but the build/boot fai=
led:
> >>>>>>>>>>
> >>>>>>>>>> failed to apply patch:
> >>>>>>>>>> checking file drivers/infiniband/sw/rxe/rxe_qp.c
> >>>>>>>>>> patch: **** unexpected end of file in patch
> >>>>>>>> This is not the root cause. The fix is not good.
> >>>>>>> This problem is about "INFO: trying to register non-static key. T=
he
> >>>>>>> code is fine but needs lockdep annotation, or maybe"
> >>>>> This warning is from "lock is not initialized". This is a
> >>>>> use-before-initialized problem.
> >>>>> The correct fix is to initialize the lock that is complained before=
 it is used.
> >>>>>
> >>>>> Zhu Yanjun
> >>>> Based on the call trace, the followings are the order of this call t=
race.
> >>>>
> >>>> 291 /* called by the create qp verb */
> >>>> 292 int rxe_qp_from_init(struct rxe_dev *rxe, struct rxe_qp *qp,
> >>>> struct rxe_pd *pd,
> >>>> 297 {
> >>>>               ...
> >>>> 317         rxe_qp_init_misc(rxe, qp, init);
> >>>>               ...
> >>>> 322
> >>>> 323         err =3D rxe_qp_init_resp(rxe, qp, init, udata, uresp);
> >>>> 324         if (err)
> >>>> 325                 goto err2;   <--- error
> >>>>
> >>>>               ...
> >>>>
> >>>> 334 err2:
> >>>> 335         rxe_queue_cleanup(qp->sq.queue); <--- Goto here
> >>>> 336         qp->sq.queue =3D NULL;
> >>>>
> >>>> In rxe_qp_init_resp, the error occurs before skb_queue_head_init.
> >>>> So this call trace appeared.
> >>> 250 static int rxe_qp_init_resp(struct rxe_dev *rxe, struct rxe_qp *q=
p,
> >>> 254 {
> >>>                           ...
> >>> 264
> >>> 265                 type =3D QUEUE_TYPE_FROM_CLIENT;
> >>> 266                 qp->rq.queue =3D rxe_queue_init(rxe, &qp->rq.max_=
wr,
> >>> 267                                         wqe_size, type);
> >>> 268                 if (!qp->rq.queue)
> >>> 269                         return -ENOMEM;    <---Error here
> >>> 270
> >>>
> >>> ...
> >>>
> >>> 282         skb_queue_head_init(&qp->resp_pkts); <-this is not called=
.
> >>> ...
> >>> This will make spin_lock of resp_pkts is used before initialized.
> >> IMHO, the above is same as
> >>
> >>> Which is caused by  "skb_queue_head_init(&qp->resp_pkts)" is not call=
ed
> >>> given rxe_qp_init_resp returns error, but the cleanup still trigger t=
he
> >>> chain.
> >>>
> >>> rxe_qp_do_cleanup -> rxe_completer -> drain_resp_pkts ->
> >>> skb_dequeue(&qp->resp_pkts)
> >> my previous analysis. If not, could you provide another better way to
> >> fix it?
> > Move the initialization to the beginning. This can fix this problem.
> > See below:
> >
> > "
> > diff --git a/drivers/infiniband/sw/rxe/rxe_qp.c
> > b/drivers/infiniband/sw/rxe/rxe_qp.c
> > index c5451a4488ca..22ef6188d7b1 100644
> > --- a/drivers/infiniband/sw/rxe/rxe_qp.c
> > +++ b/drivers/infiniband/sw/rxe/rxe_qp.c
> > @@ -176,6 +176,9 @@ static void rxe_qp_init_misc(struct rxe_dev *rxe,
> > struct rxe_qp *qp,
> >          spin_lock_init(&qp->rq.producer_lock);
> >          spin_lock_init(&qp->rq.consumer_lock);
> >
> > +       skb_queue_head_init(&qp->req_pkts);
> > +       skb_queue_head_init(&qp->resp_pkts);
> > +
> >          atomic_set(&qp->ssn, 0);
> >          atomic_set(&qp->skb_out, 0);
> >   }
> > @@ -234,8 +237,6 @@ static int rxe_qp_init_req(struct rxe_dev *rxe,
> > struct rxe_qp *qp,
> >          qp->req.opcode          =3D -1;
> >          qp->comp.opcode         =3D -1;
> >
> > -       skb_queue_head_init(&qp->req_pkts);
> > -
> >          rxe_init_task(&qp->req.task, qp, rxe_requester);
> >          rxe_init_task(&qp->comp.task, qp, rxe_completer);
> >
> > @@ -279,8 +280,6 @@ static int rxe_qp_init_resp(struct rxe_dev *rxe,
> > struct rxe_qp *qp,
> >                  }
> >          }
> >
> > -       skb_queue_head_init(&qp->resp_pkts);
> > -
> >          rxe_init_task(&qp->resp.task, qp, rxe_responder);
> >
> >          qp->resp.opcode         =3D OPCODE_NONE;
> > "
>
> It is weird to me that init them in init_misc instead of init_req/resp,
> given they
> are dedicated/used for the different purpose. But just my 0.02$.

There are some initialization problems in qp init. This needs
refactoring the related functions to fix all the problems.
Currently I am working on this. You know, this is not an easy task.
It will take me a lot of time and effort.
Now I use init_misc to initialize the related variables just as I did
in the past.
This is because init_misc is designed to initialize some variables.
And it will not cause the similar use-before-initialization problems.

Zhu Yanjun
>
> Guoqing