From mboxrd@z Thu Jan 1 00:00:00 1970 From: Robert Jarzmik Subject: Re: [PATCH] mmc: pxamci: fix card detect threaded interrupt Date: Wed, 16 Sep 2015 11:22:58 +0200 Message-ID: <87io7asop9.fsf@belgarion.home> References: <1442048611-23314-1-git-send-email-robert.jarzmik@free.fr> Mime-Version: 1.0 Content-Type: text/plain Return-path: Received: from smtp08.smtpout.orange.fr ([80.12.242.130]:46080 "EHLO smtp.smtpout.orange.fr" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752529AbbIPJ17 (ORCPT ); Wed, 16 Sep 2015 05:27:59 -0400 In-Reply-To: (Ulf Hansson's message of "Wed, 16 Sep 2015 09:24:19 +0200") Sender: linux-mmc-owner@vger.kernel.org List-Id: linux-mmc@vger.kernel.org To: Ulf Hansson Cc: linux-mmc , "linux-kernel@vger.kernel.org" , Petr Cvek Ulf Hansson writes: > On 12 September 2015 at 11:03, Robert Jarzmik wrote: >> Change the interrupt flavor of the card detection, from a hard interrupt >> to a threaded interrupt. There is no strong requirement for a hard >> interrupt. >> >> It fixes the case where the card detection is on a gpio expander, on I2C >> for example on zylonite board. In this case, the card detect netsted >> interrupt is called from a threaded interrupt. The request_irq() fails, >> because a hard irq cannot be a nested interrupt from a threaded >> interrupt (set __setup_irq()). >> >> This was tested on zylonite and mioa701 boards. > > This may indeed fix that separate issue, but looking into pxamci > driver I realize that the card detect configuration suffers from other > issues as well. > > 1. The irq shouldn't be requested until after the host has been added. > Else an IRQ will call mmc_detect_change() with a non-added mmc host, > which is really a bad idea. :-) > 2. The host_ops->get_cd() callback isn't assigned. Thus not allowing > the mmc core to find out whether the card is inserted or not. > > So instead of this quick-fix, I suggest that pxamci convert to use the > mmc slot-gpio API, which means calling mmc_gpio[d]_request_cd() from > ->probe() and assign the ->get_cd() callback, which use > mmc_gpio_get_cd(). > > What do you think? I think you're right on both points, and that you convinced me. I'll revamp the patch to use the slot-gpio API, which I wasn't aware of until now, which as a side effect will pull a bit of maintainance out of pxamci into slot-gpio, which is good for my health. Cheers. -- Robert