From mboxrd@z Thu Jan 1 00:00:00 1970 From: Dmitry Torokhov Subject: Re: [PATCH] uinput: Fix module crash on driver exit with force feedback effects still loaded Date: Sat, 25 Jun 2016 08:44:04 -0700 Message-ID: <20160625154404.GC17812@dtor-ws> References: Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Return-path: Received: from mail-pf0-f193.google.com ([209.85.192.193]:36449 "EHLO mail-pf0-f193.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751450AbcFYPoJ (ORCPT ); Sat, 25 Jun 2016 11:44:09 -0400 Received: by mail-pf0-f193.google.com with SMTP id i123so11866413pfg.3 for ; Sat, 25 Jun 2016 08:44:08 -0700 (PDT) Content-Disposition: inline In-Reply-To: Sender: linux-input-owner@vger.kernel.org List-Id: linux-input@vger.kernel.org To: Manuel Reimer Cc: linux-input On Sun, May 22, 2016 at 08:22:23PM +0200, Manuel Reimer wrote: > Hello, > > I had a deeper look into the uinput kernel module crash problem, > which happens when a game loads effects and the driver exits while > the game is still running. > > The actual problem is, that "input_unregister_device", which is > called in "uinput_destroy_device", seems to include some "cleanup > routine" which includes erasing loaded force feedback effects. This > cleanup routine causes the "uinput_dev_erase_effect" routine to be > called, which will wait for the, already exited, daemon to answer > this request. There are 2 cases when we destroy the device: - when daemon is exiting. In this case you are absolutely right, we have no hope of talking to the device and should not be trying to send requests to it; - when userspace explicitly requested the device to be destroyed via UI_DEV_DESTROY ioctl. In this case I believe the kernel should attempt orderly device shutdown which includes stopping all FF effects. I guess we'll need to introduce another bit of state and adjust behavior accordingly. > > The existing code contains a device state test, but this test is > placed wrong. My patch moves the device state test, which was in > "uinput_request_send", to the "uinput_request_submit" function. This The device state is supposed to be protected by udev->mutex, your change violates this constraint. > > way the erase requests, triggered by "input_unregister_device", are > rejected immediately, so they no longer cause any trouble. > > Should I have added a one-line comment above the device state test, > mentioning the cleanup routine in "input_unregister_device"? > > Signed-off-by: Manuel Reimer > > --- a/drivers/input/misc/uinput.c 2016-05-13 16:06:31.919351656 +0200 > +++ b/drivers/input/misc/uinput.c 2016-05-22 19:33:30.814403482 +0200 > @@ -117,11 +117,6 @@ static int uinput_request_send(struct ui > if (retval) > return retval; > > - if (udev->state != UIST_CREATED) { > - retval = -ENODEV; > - goto out; > - } > - > init_completion(&request->done); > > /* > @@ -130,7 +125,6 @@ static int uinput_request_send(struct ui > */ > uinput_dev_event(udev->dev, EV_UINPUT, request->code, request->id); > > - out: > mutex_unlock(&udev->mutex); > return retval; > } > @@ -140,6 +134,9 @@ static int uinput_request_submit(struct > { > int error; > > + if (udev->state != UIST_CREATED) > + return -ENODEV; > + > error = uinput_request_reserve_slot(udev, request); > if (error) > return error; Thanks. -- Dmitry