From mboxrd@z Thu Jan 1 00:00:00 1970 From: Jens Axboe Subject: Re: [PATCH 00/11] removing the on-stack struct request Date: Wed, 23 Apr 2008 10:27:47 +0200 Message-ID: <20080423082747.GH12774@kernel.dk> References: <1208824002-3596-1-git-send-email-fujita.tomonori@lab.ntt.co.jp> <58cb370e0804220214va6b7379tfa5b18acf36fd1a7@mail.gmail.com> <20080423074010.GA8221@gollum.tnic> <20080423172500J.tomof@acm.org> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Return-path: Received: from brick.kernel.dk ([87.55.233.238]:10925 "EHLO kernel.dk" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753242AbYDWI1v (ORCPT ); Wed, 23 Apr 2008 04:27:51 -0400 Content-Disposition: inline In-Reply-To: <20080423172500J.tomof@acm.org> Sender: linux-ide-owner@vger.kernel.org List-Id: linux-ide@vger.kernel.org To: FUJITA Tomonori Cc: petkovbb@gmail.com, bzolnier@gmail.com, linux-ide@vger.kernel.org On Wed, Apr 23 2008, FUJITA Tomonori wrote: > On Wed, 23 Apr 2008 09:40:11 +0200 > Borislav Petkov wrote: > > > On Tue, Apr 22, 2008 at 11:14:28AM +0200, Bartlomiej Zolnierkiewicz wrote: > > > > [..] > > > > > Looks promising but probably I'll not be able to review it properly > > > this week (I'm busy with my real job) and the next week (I'll be busy > > > with pushing overdue IDE updates). OTOH I'm hoping that Borislav > > > would be able to take a look in the meantime. > > > > Yep, they're quite straight-forward and look fine except some nitpicking i sent > > in separate mails. > > Thanks for reviewing, > > Yeah, these patches are pretty straightforward mainly because they are > against the paths where we can use __GFP_WAIT. > > > > FWIW, this is what we wanted to do initially but Jens didn't > > agree to that at the time. I guess, it's because i probably didn't express > > myself clearly enough then. See http://kerneltrap.org/mailarchive/linux-kernel/2008/3/1/1037974 > > for the whole discussion. > > I think that IDE uses request struct from interrupt context at some > places. In such places, I don't think that converting requests on the > stack and a pre-allocated fixed number of requests to > kmalloc/blk_get_request with GFP_ATOMIC is an good idea. It's better > to remove such code (allocating requests from interrupt context). Fully agree, that is what I originally opposed. Then IDE should pre-alloc a request at a safe time for usage like this. -- Jens Axboe