From mboxrd@z Thu Jan 1 00:00:00 1970 From: Mauro Carvalho Chehab Subject: Re: [PATCH 3/9] IR: replace spinlock with mutex. Date: Wed, 28 Jul 2010 15:37:28 -0300 Message-ID: <4C5078E8.5010409@redhat.com> References: <1280330051-27732-1-git-send-email-maximlevitsky@gmail.com> <1280330051-27732-4-git-send-email-maximlevitsky@gmail.com> <4C5054B6.5050207@redhat.com> <1280334778.28785.6.camel@localhost.localdomain> <20100728174338.GD26480@redhat.com> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <20100728174338.GD26480@redhat.com> Sender: linux-media-owner@vger.kernel.org To: Jarod Wilson Cc: Maxim Levitsky , lirc-list@lists.sourceforge.net, Jarod Wilson , linux-input@vger.kernel.org, linux-media@vger.kernel.org List-Id: linux-input@vger.kernel.org Em 28-07-2010 14:43, Jarod Wilson escreveu: > On Wed, Jul 28, 2010 at 07:32:58PM +0300, Maxim Levitsky wrote: >> On Wed, 2010-07-28 at 13:03 -0300, Mauro Carvalho Chehab wrote: >>> Em 28-07-2010 12:14, Maxim Levitsky escreveu: >>>> Some handlers (lirc for example) allocates memory on initialization, >>>> doing so in atomic context is cumbersome. >>>> Fixes warning about sleeping function in atomic context. >>> >>> You should not replace it by a mutex, as the decoding code may happen during >>> IRQ time on several drivers. >> I though decoding code is run by a work queue? > > Yeah, it is. (INIT_WORK(&ir->raw->rx_work, ir_raw_event_work); in > ir_raw_event_register). > >> I don't see any atomic codepath here... > > I think the ir_raw_event_store variants are the only things that are run > from an interrupt context, and none of them touch ir_raw_handler_lock. > That lock is advertised as being for the protection of ir_raw_handler_list > and ir_raw_client_list, which are primarily manipulated by > register/unregister functions, and we just lock them when doing actual IR > decode work (via said work queue) so we don't feed raw IR somewhere that > we shouldn't. I think Maxim is correct here, we should be okay with > changing this to a mutex, unless I'm missing something else. You're probably right. The previous code used to do this at IRQ time, but a latter patch changed it to happen via a workqueue. So, I'm OK with this patch. Cheers, Mauro.