From: Eric Farman <farman@linux.ibm.com>
To: Damien Le Moal <Damien.LeMoal@wdc.com>,
"linux-block@vger.kernel.org" <linux-block@vger.kernel.org>,
Jens Axboe <axboe@kernel.dk>,
"linux-scsi@vger.kernel.org" <linux-scsi@vger.kernel.org>,
"Martin K . Petersen" <martin.petersen@oracle.com>,
"dm-devel@redhat.com" <dm-devel@redhat.com>,
Mike Snitzer <snitzer@redhat.com>
Cc: Ming Lei <ming.lei@redhat.com>,
Christian Borntraeger <borntraeger@de.ibm.com>
Subject: Re: [PATCH v5 5/7] block: Delay default elevator initialization
Date: Thu, 17 Oct 2019 09:37:06 -0400 [thread overview]
Message-ID: <f8b8a326-49f1-b92a-6263-de945663f27b@linux.ibm.com> (raw)
In-Reply-To: <BYAPR04MB5816A2CFCDA1D7ED10455D93E76D0@BYAPR04MB5816.namprd04.prod.outlook.com>
On 10/16/19 10:39 PM, Damien Le Moal wrote:
> On 2019/10/17 3:51, Eric Farman wrote:
>>
>>
>> On 10/1/19 4:46 PM, Eric Farman wrote:
>>>
>>>
>>> On 9/5/19 5:51 AM, Damien Le Moal wrote:
>>>> When elevator_init_mq() is called from blk_mq_init_allocated_queue(),
>>>> the only information known about the device is the number of hardware
>>>> queues as the block device scan by the device driver is not completed
>>>> yet for most drivers. The device type and elevator required features
>>>> are not set yet, preventing to correctly select the default elevator
>>>> most suitable for the device.
>>>>
>>>> This currently affects all multi-queue zoned block devices which default
>>>> to the "none" elevator instead of the required "mq-deadline" elevator.
>>>> These drives currently include host-managed SMR disks connected to a
>>>> smartpqi HBA and null_blk block devices with zoned mode enabled.
>>>> Upcoming NVMe Zoned Namespace devices will also be affected.
>>>>
>>>> Fix this by adding the boolean elevator_init argument to
>>>> blk_mq_init_allocated_queue() to control the execution of
>>>> elevator_init_mq(). Two cases exist:
>>>> 1) elevator_init = false is used for calls to
>>>> blk_mq_init_allocated_queue() within blk_mq_init_queue(). In this
>>>> case, a call to elevator_init_mq() is added to __device_add_disk(),
>>>> resulting in the delayed initialization of the queue elevator
>>>> after the device driver finished probing the device information. This
>>>> effectively allows elevator_init_mq() access to more information
>>>> about the device.
>>>> 2) elevator_init = true preserves the current behavior of initializing
>>>> the elevator directly from blk_mq_init_allocated_queue(). This case
>>>> is used for the special request based DM devices where the device
>>>> gendisk is created before the queue initialization and device
>>>> information (e.g. queue limits) is already known when the queue
>>>> initialization is executed.
>>>>
>>>> Additionally, to make sure that the elevator initialization is never
>>>> done while requests are in-flight (there should be none when the device
>>>> driver calls device_add_disk()), freeze and quiesce the device request
>>>> queue before calling blk_mq_init_sched() in elevator_init_mq().
>>>>
>>>> Signed-off-by: Damien Le Moal <damien.lemoal@wdc.com>
>>>
>>> Coincidentally, I had been looking into a problem that is fixed in
>>> 5.4-rc1 by this patch. Thanks for that!
>>>
>>> The problem was a delay during boot of a KVM guest with virtio-scsi
>>> devices (or hotplug of such a device to a guest) in recent releases,
>>> especially when virtio-scsi is configured as a module. The symptoms
>>> look like:
>>>
>>> [ 0.975315] virtio_blk virtio2: [vda] 1803060 4096-byte logical
>>> blocks (7.39 GB/6.88 GiB)
>>> [ 0.977859] scsi host0: Virtio SCSI HBA
>>> [ 0.980339] scsi 0:0:0:0: Direct-Access QEMU QEMU HARDDISK
>>> 2.5+ PQ: 0 ANSI: 5
>>> [ 0.981685] vda:VOL1/ 0XA906: vda1
>>> [ 0.988253] alg: No test for crc32be (crc32be-vx)
>>> ...stall...
>>> [ 24.544920] sd 0:0:0:0: Power-on or device reset occurred
>>> [ 24.545176] sd 0:0:0:0: Attached scsi generic sg0 type 0
>>> [ 24.545292] sd 0:0:0:0: [sda] 385 512-byte logical blocks: (197
>>> kB/193 KiB)
>>> [ 24.545368] sd 0:0:0:0: [sda] Write Protect is off
>>> [ 24.545416] sd 0:0:0:0: [sda] Mode Sense: 63 00 00 08
>>> [ 24.545456] sd 0:0:0:0: [sda] Write cache: enabled, read cache:
>>> enabled, doesn't support DPO or FUA
>>> [ 24.547033] sd 0:0:0:0: [sda] Attached SCSI disk
>>>
>>> I debugged this down to the same behavior described/fixed back in 3.18
>>> by commit 17497acbdce9 ("blk-mq, percpu_ref: start q->mq_usage_counter
>>> in atomic mode"), and for the same reason. The delay starts occurring
>>> as soon as q->q_usage_counter is converted to percpu for the one LUN tha
>>> twas found, while scsi_scan_channel() is still working on its loop of
>>> mostly non-existent devices. Exactly when this problem started
>>> re-occuring is not certain to me, though I did see this problem with 5.2
>>> on linux-stable.
>>
>> This problem started occurring reliably with kernel 4.16 because of
>> commit b5b6e8c8d3b4 ("scsi: virtio_scsi: fix IO hang caused by automatic
>> irq vector affinity") which forced blk-mq on for virtio-scsi devices.
>>
>> I'm able to reproduce the behavior (fixed by this commit) on 4.15.0, as
>> well as 4.14.0 and 4.9.0, if I enable SCSI_MQ_DEFAULT in the kernel
>> config. I cannot reproduce it with 4.4.0, but didn't chase further to
>> see why that was the case.
>
> This is really strange since in 4.9 there is no elevator initialization being
> done for a mq device. So with SCSI_MQ_DEFAULT set, you should end up with no
> elevator for your disk and the presence or not of this patch or variations of it
> should not change anything.
>
> This patch may be hiding your problem root cause rather than fixing it...
Neat. Back to debugging 5.3 then.
Thanks,
Eric
>
>>
>> That force_blk_mq commit went into linux-stable, so SCSI_MQ_DEFAULT need
>> not be enabled on 4.14.y or 4.9.y stable branches to see the behavior
>> fixed by this commit.
>>
>>>
>>> When I run with a 5.3 kernel, the problem is easily reproducible. So I
>>> bisected between 5.3 and 5.4-rc1, and got here. Cherry-picking this
>>> patch on top of 5.3 cleans up the boot/hotplug process and removes any
>>> stall. Any chance this could be cc'd to stable?
>>
>> Please? Anything I can do to help with that effort?
>>
>> Thanks,
>> Eric
>>
>>> Any data someone wants
>>> to see behavioral changes?
>>>
>>> Thanks,
>>> Eric
>>>
>>>> ---
>>>> block/blk-mq.c | 12 +++++++++---
>>>> block/elevator.c | 7 +++++++
>>>> block/genhd.c | 9 +++++++++
>>>> drivers/md/dm-rq.c | 2 +-
>>>> include/linux/blk-mq.h | 3 ++-
>>>> 5 files changed, 28 insertions(+), 5 deletions(-)
>>>>
>>>> diff --git a/block/blk-mq.c b/block/blk-mq.c
>>>> index ee4caf0c0807..240416057f28 100644
>>>> --- a/block/blk-mq.c
>>>> +++ b/block/blk-mq.c
>>>> @@ -2689,7 +2689,11 @@ struct request_queue *blk_mq_init_queue(struct blk_mq_tag_set *set)
>>>> if (!uninit_q)
>>>> return ERR_PTR(-ENOMEM);
>>>>
>>>> - q = blk_mq_init_allocated_queue(set, uninit_q);
>>>> + /*
>>>> + * Initialize the queue without an elevator. device_add_disk() will do
>>>> + * the initialization.
>>>> + */
>>>> + q = blk_mq_init_allocated_queue(set, uninit_q, false);
>>>> if (IS_ERR(q))
>>>> blk_cleanup_queue(uninit_q);
>>>>
>>>> @@ -2840,7 +2844,8 @@ static unsigned int nr_hw_queues(struct blk_mq_tag_set *set)
>>>> }
>>>>
>>>> struct request_queue *blk_mq_init_allocated_queue(struct blk_mq_tag_set *set,
>>>> - struct request_queue *q)
>>>> + struct request_queue *q,
>>>> + bool elevator_init)
>>>> {
>>>> /* mark the queue as mq asap */
>>>> q->mq_ops = set->ops;
>>>> @@ -2902,7 +2907,8 @@ struct request_queue *blk_mq_init_allocated_queue(struct blk_mq_tag_set *set,
>>>> blk_mq_add_queue_tag_set(set, q);
>>>> blk_mq_map_swqueue(q);
>>>>
>>>> - elevator_init_mq(q);
>>>> + if (elevator_init)
>>>> + elevator_init_mq(q);
>>>>
>>>> return q;
>>>>
>>>> diff --git a/block/elevator.c b/block/elevator.c
>>>> index 520d6b224b74..096a670d22d7 100644
>>>> --- a/block/elevator.c
>>>> +++ b/block/elevator.c
>>>> @@ -712,7 +712,14 @@ void elevator_init_mq(struct request_queue *q)
>>>> if (!e)
>>>> return;
>>>>
>>>> + blk_mq_freeze_queue(q);
>>>> + blk_mq_quiesce_queue(q);
>>>> +
>>>> err = blk_mq_init_sched(q, e);
>>>> +
>>>> + blk_mq_unquiesce_queue(q);
>>>> + blk_mq_unfreeze_queue(q);
>>>> +
>>>> if (err) {
>>>> pr_warn("\"%s\" elevator initialization failed, "
>>>> "falling back to \"none\"\n", e->elevator_name);
>>>> diff --git a/block/genhd.c b/block/genhd.c
>>>> index 54f1f0d381f4..26b31fcae217 100644
>>>> --- a/block/genhd.c
>>>> +++ b/block/genhd.c
>>>> @@ -695,6 +695,15 @@ static void __device_add_disk(struct device *parent, struct gendisk *disk,
>>>> dev_t devt;
>>>> int retval;
>>>>
>>>> + /*
>>>> + * The disk queue should now be all set with enough information about
>>>> + * the device for the elevator code to pick an adequate default
>>>> + * elevator if one is needed, that is, for devices requesting queue
>>>> + * registration.
>>>> + */
>>>> + if (register_queue)
>>>> + elevator_init_mq(disk->queue);
>>>> +
>>>> /* minors == 0 indicates to use ext devt from part0 and should
>>>> * be accompanied with EXT_DEVT flag. Make sure all
>>>> * parameters make sense.
>>>> diff --git a/drivers/md/dm-rq.c b/drivers/md/dm-rq.c
>>>> index 21d5c1784d0c..3f8577e2c13b 100644
>>>> --- a/drivers/md/dm-rq.c
>>>> +++ b/drivers/md/dm-rq.c
>>>> @@ -563,7 +563,7 @@ int dm_mq_init_request_queue(struct mapped_device *md, struct dm_table *t)
>>>> if (err)
>>>> goto out_kfree_tag_set;
>>>>
>>>> - q = blk_mq_init_allocated_queue(md->tag_set, md->queue);
>>>> + q = blk_mq_init_allocated_queue(md->tag_set, md->queue, true);
>>>> if (IS_ERR(q)) {
>>>> err = PTR_ERR(q);
>>>> goto out_tag_set;
>>>> diff --git a/include/linux/blk-mq.h b/include/linux/blk-mq.h
>>>> index 62a3bb715899..0bf056de5cc3 100644
>>>> --- a/include/linux/blk-mq.h
>>>> +++ b/include/linux/blk-mq.h
>>>> @@ -248,7 +248,8 @@ enum {
>>>>
>>>> struct request_queue *blk_mq_init_queue(struct blk_mq_tag_set *);
>>>> struct request_queue *blk_mq_init_allocated_queue(struct blk_mq_tag_set *set,
>>>> - struct request_queue *q);
>>>> + struct request_queue *q,
>>>> + bool elevator_init);
>>>> struct request_queue *blk_mq_init_sq_queue(struct blk_mq_tag_set *set,
>>>> const struct blk_mq_ops *ops,
>>>> unsigned int queue_depth,
>>>>
>>
>
>
next prev parent reply other threads:[~2019-10-17 13:37 UTC|newest]
Thread overview: 14+ messages / expand[flat|nested] mbox.gz Atom feed top
2019-09-05 9:51 [PATCH v5 0/7] Elevator cleanups and improvements Damien Le Moal
2019-09-05 9:51 ` [PATCH v5 1/7] block: Cleanup elevator_init_mq() use Damien Le Moal
2019-09-05 9:51 ` [PATCH v5 2/7] block: Change elevator_init_mq() to always succeed Damien Le Moal
2019-09-05 9:51 ` [PATCH v5 3/7] block: Introduce elevator features Damien Le Moal
2019-09-05 9:51 ` [PATCH v5 4/7] block: Improve default elevator selection Damien Le Moal
2019-09-05 9:51 ` [PATCH v5 5/7] block: Delay default elevator initialization Damien Le Moal
2019-10-01 20:46 ` Eric Farman
2019-10-16 18:51 ` Eric Farman
2019-10-17 2:39 ` Damien Le Moal
2019-10-17 13:37 ` Eric Farman [this message]
2019-09-05 9:51 ` [PATCH v5 6/7] block: Set ELEVATOR_F_ZBD_SEQ_WRITE for nullblk zoned disks Damien Le Moal
2019-09-05 9:51 ` [PATCH v5 7/7] sd: Set ELEVATOR_F_ZBD_SEQ_WRITE for ZBC disks Damien Le Moal
2019-09-06 0:31 ` [PATCH v5 0/7] Elevator cleanups and improvements Ming Lei
2019-09-06 1:53 ` Jens Axboe
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=f8b8a326-49f1-b92a-6263-de945663f27b@linux.ibm.com \
--to=farman@linux.ibm.com \
--cc=Damien.LeMoal@wdc.com \
--cc=axboe@kernel.dk \
--cc=borntraeger@de.ibm.com \
--cc=dm-devel@redhat.com \
--cc=linux-block@vger.kernel.org \
--cc=linux-scsi@vger.kernel.org \
--cc=martin.petersen@oracle.com \
--cc=ming.lei@redhat.com \
--cc=snitzer@redhat.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).