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
In-Reply-To: <1662404120-24338-1-git-send-email-quic_deesin@quicinc.com>
References: <1662404120-24338-1-git-send-email-quic_deesin@quicinc.com>
From:   Stephen Boyd <swboyd@chromium.org>
User-Agent: alot/0.10
Date:   Wed, 7 Sep 2022 19:52:59 -0500
Message-ID: <CAE-0n51oCcb8_1No22gNy2QS=ZL=DxQLZWT9NNGNZGev+4cy8g@mail.gmail.com>
Subject: Re: [PATCH V2 1/2] rpmsg: glink: Add lock to avoid race when rpmsg
 device is released
To:     Deepak Kumar Singh <quic_deesin@quicinc.com>,
        bjorn.andersson@linaro.org, mathieu.poirier@linaro.org,
        quic_clew@quicinc.com
Cc:     linux-kernel@vger.kernel.org, linux-arm-msm@vger.kernel.org,
        linux-remoteproc@vger.kernel.org
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

Quoting Deepak Kumar Singh (2022-09-05 11:55:19)
> When remote host goes down glink char device channel is freed,
> At the same time user space apps can still try to open/poll rpmsg
> char device which will result in calling rpmsg_create_ept. This may
> cause reference to already freed context of glink chardev channel and
> result in below crash signatures -
>
> 1)
> rpmsg_create_ept+0x40/0xa0
> rpmsg_eptdev_open+0x88/0x138
> chrdev_open+0xc4/0x1c8
> do_dentry_open+0x230/0x378
>
> 2)
> rpmsg_poll+0x5c/0x80
> rpmsg_eptdev_poll+0x84/0xa4
> do_sys_poll+0x22c/0x5c8
>
> This patch adds proper lock and check condition to avoid such crash.

Usually it's nice to have a two CPU diagram that shows the problem
scenario with time flowing down and interleaving the calls with
indentation, instead of some callstacks. Can you add that detail here? I
guess it would look like this:

  CPU0                      CPU1
  ----                      ----
  do_sys_poll()             chrdev_open()
   rpmsg_eptdev_poll()       rpmsg_eptdev_open()
                              rpmsg_create_ept()
    rpmsg_poll()
     <access old rpdev pointer?>

but I don't immediately understand the problem. Probably showing the
free path and wherever eptdev is set to NULL will help inform better
than a parallel poll and chrdev open.

>
> Signed-off-by: Deepak Kumar Singh <quic_deesin@quicinc.com>
> ---

Any Fixes tag?