From mboxrd@z Thu Jan 1 00:00:00 1970 Return-path: Received: from mx1.redhat.com ([209.132.183.28]:5969 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S932302Ab0EDAHd convert rfc822-to-8bit (ORCPT ); Mon, 3 May 2010 20:07:33 -0400 Date: Mon, 3 May 2010 21:07:26 -0300 From: Mauro Carvalho Chehab To: matthieu castet Cc: linux-media@vger.kernel.org Subject: Re: [PATCH] fix dvb frontend lockup Message-ID: <20100503210726.66a41c01@pedra> In-Reply-To: <4BAE810A.6030405@free.fr> References: <4BA603AB.6070809@free.fr> <4BAE810A.6030405@free.fr> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 8BIT Sender: linux-media-owner@vger.kernel.org List-ID: Hi Matthieu, Em Sat, 27 Mar 2010 23:04:58 +0100 matthieu castet escreveu: > Etiquetas: NonJunk, unknown-1 > From: matthieu castet > To: linux-media@vger.kernel.org > Subject: [PATCH] fix dvb frontend lockup > Date: Sat, 27 Mar 2010 23:04:58 +0100 > Sender: linux-media-owner@vger.kernel.org > User-Agent: Mozilla/5.0 (X11; U; Linux i686; fr; rv:1.8.1.23) Gecko/20090823 SeaMonkey/1.1.18 > > matthieu castet a écrit : > > Hi, > > > > With my current kernel (2.6.32), if my dvb device is removed while in > > use, I got [1]. > > > > After checking the source code, the problem seems to happen also in > > master : > > > > If there are users (for example users == -2) : > > - dvb_unregister_frontend : > > - stop kernel thread with dvb_frontend_stop : > > - fepriv->exit = 1; > > - thread loop catch stop event and break while loop > > - fepriv->thread = NULL; and fepriv->exit = 0; > > - dvb_unregister_frontend wait on "fepriv->dvbdev->wait_queue" that > > fepriv->dvbdev->users==-1. > > The user finish : > > - dvb_frontend_release - set users to -1 > > - don't wait wait_queue because fepriv->exit != 1 > > > > => dvb_unregister_frontend never exit the wait queue. > > > > > > Matthieu > > > > > > [ 4920.484047] khubd D c2008000 0 198 2 0x00000000 > > [ 4920.484056] f64c8000 00000046 00000000 c2008000 00000004 c13fa000 > > c13fa000 f > > 64c8000 > > [ 4920.484066] f64c81bc c2008000 00000000 d9d9dce6 00000452 00000001 > > f64c8000 c > > 102daad > > [ 4920.484075] 00100100 f64c81bc 00000286 f0a7ccc0 f0913404 f0cba404 > > f644de58 f > > 092b0a8 > > [ 4920.484084] Call Trace: > > [ 4920.484102] [] ? default_wake_function+0x0/0x8 > > [ 4920.484147] [] ? dvb_unregister_frontend+0x95/0xcc [dvb_core] > > [ 4920.484157] [] ? autoremove_wake_function+0x0/0x2d > > [ 4920.484168] [] ? dvb_usb_adapter_frontend_exit+0x12/0x21 > > [dvb_usb] > > [ 4920.484176] [] ? dvb_usb_exit+0x26/0x88 [dvb_usb] > > [ 4920.484184] [] ? dvb_usb_device_exit+0x3a/0x4a [dvb_usb] > > [ 4920.484217] [] ? usb_unbind_interface+0x3f/0xb4 [usbcore] > > [ 4920.484227] [] ? __device_release_driver+0x74/0xb7 > > [ 4920.484233] [] ? device_release_driver+0x15/0x1e > > [ 4920.484243] [] ? bus_remove_device+0x6e/0x87 > > [ 4920.484249] [] ? device_del+0xfa/0x152 > > [ 4920.484264] [] ? usb_disable_device+0x59/0xb9 [usbcore] > > [ 4920.484279] [] ? usb_disconnect+0x70/0xdc [usbcore] > > [ 4920.484294] [] ? hub_thread+0x521/0xe1d [usbcore] > > [ 4920.484301] [] ? autoremove_wake_function+0x0/0x2d > > [ 4920.484316] [] ? hub_thread+0x0/0xe1d [usbcore] > > [ 4920.484321] [] ? kthread+0x61/0x66 > > [ 4920.484327] [] ? kthread+0x0/0x66 > > [ 4920.484336] [] ? kernel_thread_helper+0x7/0x10 > > > Here a patch that fix the issue > > > Signed-off-by: Matthieu CASTET > > --- 1/linux/drivers/media/dvb/dvb-core/dvb_frontend.c 2010-03-27 22:59:50.000000000 +0100 > +++ 2/linux/drivers/media/dvb/dvb-core/dvb_frontend.c 2010-03-27 23:01:34.000000000 +0100 > @@ -686,7 +686,10 @@ > } > > fepriv->thread = NULL; > - fepriv->exit = 0; > + if (kthread_should_stop()) > + fepriv->exit = 2; Your patch makes sense for me. However, using a "magic" 2 value for exit will not be clear enough that this means that the device got removed. The better is to define something like: #define DVB_FE_NORMAL_EXIT 1 #define DVB_FE_DEVICE_REMOVED 2 And use either one of the macro values for fepriv->exit. This will be clear for those that may later need to touch on that code. > + else > + fepriv->exit = 0; > mb(); > > dvb_frontend_wakeup(fe); > @@ -1929,6 +1932,8 @@ > int ret; > > dprintk ("%s\n", __func__); > + if (fepriv->exit == 2) > + return -ENODEV; > > if (adapter->mfe_shared) { > mutex_lock (&adapter->mfe_lock); > @@ -2021,7 +2026,7 @@ > ret = dvb_generic_release (inode, file); > > if (dvbdev->users == -1) { > - if (fepriv->exit == 1) { > + if (fepriv->exit) { > fops_put(file->f_op); > file->f_op = NULL; > wake_up(&dvbdev->wait_queue) -- Cheers, Mauro