From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1752768AbaATEQO (ORCPT ); Sun, 19 Jan 2014 23:16:14 -0500 Received: from mail.linux-iscsi.org ([67.23.28.174]:35088 "EHLO linux-iscsi.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752559AbaATEQK (ORCPT ); Sun, 19 Jan 2014 23:16:10 -0500 Message-ID: <1390191463.5567.760.camel@haakon3.risingtidesystems.com> Subject: Re: [PATCH 1/5] percpu_ida: Make percpu_ida_alloc accept task state bitmask From: "Nicholas A. Bellinger" To: Linus Torvalds Cc: target-devel , linux-kernel , Ingo Molnar , Peter Zijlstra , Kent Overstreet , Jens Axboe Date: Sun, 19 Jan 2014 20:17:43 -0800 In-Reply-To: References: <1390126616-29560-1-git-send-email-nab@linux-iscsi.org> <1390126616-29560-2-git-send-email-nab@linux-iscsi.org> Content-Type: text/plain; charset="UTF-8" X-Mailer: Evolution 3.4.4-1 Mime-Version: 1.0 Content-Transfer-Encoding: 7bit Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Sun, 2014-01-19 at 18:38 -0800, Linus Torvalds wrote: > On Sun, Jan 19, 2014 at 2:16 AM, Nicholas A. Bellinger > wrote: > > > > This patch changes percpu_ida_alloc() to accept task state bitmask > > for prepare_to_wait() to support interruptible sleep for callers > > that require it. > > This patch-series is not bisectable. Afaik, the first patch will break > the build (or at least cause the end result to not actually work). > > This kind of "split up one large patch into many small patches THAT > DON'T ACTUALLY WORK INDIVIDUALLY" model is pure and utter garbage. > > So a big NAK on this series as being completely broken. > So the late night reasoning was to allow the patches to apply cleanly to stable. That was, indeed, a bad decision. > To fix it, I would suggest: > > - make the first patch change all *existing* users (that only have > the atomic vs uninterruptible semantics) pass in either > TASK_UNINTERRUPTIBLE or TASK_RUNNING depending on whether they had > __GFP_WAIT or not. > > So the first patch would not change *any* semantics or behavior, it > would only change the calling convention. > > - do the cleanup patches to block/blk-mq-tag.c to not have those > "gfp" calling convention, and instead passing in the state natively > > - add the TASK_INTERRUPTIBLE case last (which includes the new > "signal_pending_state()" logic in percpu_ida_alloc()) > > that way, all patches compile cleanly and should each work > individually, and they all do clearly just one thing. And the biggest > patch in the series (the first one) doesn't actually make any semantic > changes. -v2 sent out. Please review. Jens, please review the blk-mq specific changes in patch #1, and let me know if you'd like to pick-up #2 via the block tree, or have it included in target-pending/for-next. Thanks, --nab