Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1751551AbdFAJcc (ORCPT ); Thu, 1 Jun 2017 05:32:32 -0400 Received: from mail-qk0-f196.google.com ([209.85.220.196]:35164 "EHLO mail-qk0-f196.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751279AbdFAJca (ORCPT ); Thu, 1 Jun 2017 05:32:30 -0400 MIME-Version: 1.0 In-Reply-To: <1496305735-21741-1-git-send-email-baijiaju1990@163.com> References: <1496305735-21741-1-git-send-email-baijiaju1990@163.com> From: Moni Shoua Date: Thu, 1 Jun 2017 12:32:29 +0300 X-Google-Sender-Auth: XCnUCxiKrWL8-X6dlNkk9wSBeS0 Message-ID: Subject: Re: [PATCH V3] rxe: Fix a sleep-in-atomic bug in post_one_send To: Jia-Ju Bai Cc: Sean Hefty , Doug Ledford , Hal Rosenstock , Leon Romanovsky , linux-rdma , Linux Kernel Mailinglist Content-Type: text/plain; charset="UTF-8" Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 2219 Lines: 51 On Thu, Jun 1, 2017 at 11:28 AM, Jia-Ju Bai wrote: > The driver may sleep under a spin lock, and the function call path is: > post_one_send (acquire the lock by spin_lock_irqsave) > init_send_wqe > copy_from_user --> may sleep > > To fix it, the lock is released before copy_from_user, and the lock is > acquired again after this function. The parameter "flags" is used to > restore and save the irq status. > Thank Leon for good advice. > > This patch corrects the mistakes in V2. (Thank Ram for pointing it out) > > Signed-off-by: Jia-Ju Bai > --- > drivers/infiniband/sw/rxe/rxe_verbs.c | 13 ++++++++----- > 1 file changed, 8 insertions(+), 5 deletions(-) > > diff --git a/drivers/infiniband/sw/rxe/rxe_verbs.c b/drivers/infiniband/sw/rxe/rxe_verbs.c > index 83d709e..5293d15 100644 > --- a/drivers/infiniband/sw/rxe/rxe_verbs.c > +++ b/drivers/infiniband/sw/rxe/rxe_verbs.c > @@ -721,11 +721,11 @@ static void init_send_wr(struct rxe_qp *qp, struct rxe_send_wr *wr, > > static int init_send_wqe(struct rxe_qp *qp, struct ib_send_wr *ibwr, > unsigned int mask, unsigned int length, > - struct rxe_send_wqe *wqe) > + struct rxe_send_wqe *wqe, unsigned long *flags) > { > int num_sge = ibwr->num_sge; > struct ib_sge *sge; > - int i; > + int i, err; > u8 *p; > > init_send_wr(qp, &wqe->wr, ibwr); > @@ -740,8 +740,11 @@ static int init_send_wqe(struct rxe_qp *qp, struct ib_send_wr *ibwr, > > sge = ibwr->sg_list; > for (i = 0; i < num_sge; i++, sge++) { > - if (qp->is_user && copy_from_user(p, (__user void *) > - (uintptr_t)sge->addr, sge->length)) > + spin_unlock_irqrestore(&qp->sq.sq_lock, *flags); Before the patch, copy_from_user() was called only if qp->is_user was true. After the patch it is always called. Second, I think that there is no flow that leads to this function when qp->is user is true so maybe the correct action is to remove this line completely if (qp->is_user && copy_from_user(p, (__user void *)