From patchwork Wed Mar 27 13:51:00 2019 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Martin Wilck X-Patchwork-Id: 10873505 Return-Path: Received: from mail.wl.linuxfoundation.org (pdx-wl-mail.web.codeaurora.org [172.30.200.125]) by pdx-korg-patchwork-2.web.codeaurora.org (Postfix) with ESMTP id DE698139A for ; Wed, 27 Mar 2019 13:51:51 +0000 (UTC) Received: from mail.wl.linuxfoundation.org (localhost [127.0.0.1]) by mail.wl.linuxfoundation.org (Postfix) with ESMTP id C9C9F28C02 for ; Wed, 27 Mar 2019 13:51:51 +0000 (UTC) Received: by mail.wl.linuxfoundation.org (Postfix, from userid 486) id C7E8028D75; Wed, 27 Mar 2019 13:51:51 +0000 (UTC) X-Spam-Checker-Version: SpamAssassin 3.3.1 (2010-03-16) on pdx-wl-mail.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-7.9 required=2.0 tests=BAYES_00,MAILING_LIST_MULTI, RCVD_IN_DNSWL_HI autolearn=ham version=3.3.1 Received: from vger.kernel.org (vger.kernel.org [209.132.180.67]) by mail.wl.linuxfoundation.org (Postfix) with ESMTP id 40D0E28C0C for ; Wed, 27 Mar 2019 13:51:51 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1727993AbfC0Nvu (ORCPT ); Wed, 27 Mar 2019 09:51:50 -0400 Received: from smtp2.provo.novell.com ([137.65.250.81]:59154 "EHLO smtp2.provo.novell.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1725797AbfC0Nvu (ORCPT ); Wed, 27 Mar 2019 09:51:50 -0400 Received: from apollon.suse.de.de (prva10-snat226-2.provo.novell.com [137.65.226.36]) by smtp2.provo.novell.com with ESMTP (TLS encrypted); Wed, 27 Mar 2019 07:51:44 -0600 From: Martin Wilck To: Jens Axboe , Tejun Heo , Hannes Reinecke , "Martin K. Petersen" , Christoph Hellwig Cc: James Bottomley , Bart Van Assche , Martin Wilck , linux-block@vger.kernel.org, linux-scsi@vger.kernel.org, linux-ide@vger.kernel.org Subject: [PATCH v3 0/5] block: skip media change event handling if unsupported Date: Wed, 27 Mar 2019 14:51:00 +0100 Message-Id: <20190327135105.30893-1-mwilck@suse.com> X-Mailer: git-send-email 2.21.0 MIME-Version: 1.0 Sender: linux-block-owner@vger.kernel.org Precedence: bulk List-ID: X-Mailing-List: linux-block@vger.kernel.org X-Virus-Scanned: ClamAV using ClamSMTP The block layer currently can't distinguish between gendisk devices that don't support any media change events, and devices that do support them, but only for internal purposes. Therefore the check_events() function is called e.g. for ordinary non-removable SCSI disks. While these devices are not polled under normal conditions, the check_events function is called on certain synchronization points, in particular while the device is opened, closed, or probed. Under unfavorable conditions this can lead to processes being stalled on a blocked queue: a close() schedules a work item for check_events() which gets blocked in the work queue, and subsequent open() tries to flush the workqueue. The flush then stalls too, as long as the the blocked work item can't finish. In principle, the gendisk->events field would make it very easy for the block layer to check only for events actually supported by the device. Currently this is impossible, because there are lots of drivers which don't set gendisk->events although they implement the check_events() method. This was introduced in commit 7c88a168da80 ("block: don't propagate unlisted DISK_EVENTs to userland") and follow-up patches because uevent generation by these drivers was found to possibly generate infinite event loops between kernel and user space. A side effect of these patches was that the distinction between such devices and devices supporting no events at all was lost. This series implements a slightly different approach to event handling and uevent suppression. The drivers are changed to set the events field again. Whether or not uevents should be generated is controlled by a separate flag bit, which is only set by the drivers that are known to generate proper uevents (sd and sr). Once this is done, devices that don't support any media change events can be clearly identified, and the whole events checking code path can be skipped. This simplifies handling of non-removable SCSI disks. I have tested this with removable and non-removable SCSI disks, SCSI cdrom, and ide-cd. This patch set targets the same problem as Hannes' late submission "sd: skip non-removable devices in sd_check_events()". Changes in v3: - Improved formatting of commit messages (Christoph) - 2/5: Rather than adding the event flag bits to the "events" field, add a new field "event_flags" to struct gendisk (Christoph) - 2/5: Add a pair of parentheses around flag test (Christoph) Changes in v2: Removed the unused async_events field from struct gendisk. This simplifies the event handling logic a bit. Martin Wilck (5): block: genhd: remove async_events field block: disk_events: introduce event flags Revert "ide: unexport DISK_EVENT_MEDIA_CHANGE for ide-gd and ide-cd" Revert "block: unexport DISK_EVENT_MEDIA_CHANGE for legacy/fringe drivers" block: check_events: don't bother with events if unsupported block/genhd.c | 48 ++++++++++++++++++++++---------------- drivers/block/amiflop.c | 1 + drivers/block/ataflop.c | 1 + drivers/block/floppy.c | 1 + drivers/block/paride/pcd.c | 1 + drivers/block/paride/pd.c | 1 + drivers/block/paride/pf.c | 1 + drivers/block/pktcdvd.c | 1 - drivers/block/swim.c | 1 + drivers/block/swim3.c | 1 + drivers/block/xsysace.c | 1 + drivers/cdrom/gdrom.c | 1 + drivers/ide/ide-cd.c | 1 + drivers/ide/ide-cd_ioctl.c | 5 ++-- drivers/ide/ide-gd.c | 6 +++-- drivers/scsi/sd.c | 1 + drivers/scsi/sr.c | 1 + include/linux/genhd.h | 11 +++++++-- 18 files changed, 57 insertions(+), 27 deletions(-)