From mboxrd@z Thu Jan 1 00:00:00 1970 From: Christoph Hellwig Subject: Re: [PATCH 1/2] libata: only call ->done once all per-tag ressources are released Date: Mon, 5 Oct 2015 08:01:17 +0200 Message-ID: <20151005060117.GA2294@lst.de> References: <1443892871-10413-1-git-send-email-hch@lst.de> <1443892871-10413-2-git-send-email-hch@lst.de> <20151004173429.GD19652@htj.duckdns.org> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Return-path: Received: from verein.lst.de ([213.95.11.211]:52546 "EHLO newverein.lst.de" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751961AbbJEGBT (ORCPT ); Mon, 5 Oct 2015 02:01:19 -0400 Content-Disposition: inline In-Reply-To: <20151004173429.GD19652@htj.duckdns.org> Sender: linux-ide-owner@vger.kernel.org List-Id: linux-ide@vger.kernel.org To: Tejun Heo Cc: Christoph Hellwig , linux-ide@vger.kernel.org On Sun, Oct 04, 2015 at 01:34:29PM -0400, Tejun Heo wrote: > Hmmm... has this actually been observed? All these are run under ata > port lock and so is the command issue path, so even if the tag gets > reissued, it will have to wait till the completion path is done. No - I stumbled over this while trying to debug https://lkml.org/lkml/2015/6/25/620 (which could use some libata experience, btw), but it didn't help. I still would like to see it fixes as similar patterns in SCSI drivers have caused crashes.