From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Return-Path: From: =?utf-8?Q?Javier_Gonz=C3=A1lez?= Message-Id: Content-Type: multipart/signed; boundary="Apple-Mail=_6B030D3D-F25D-4E13-A4B9-D68F9030134A"; protocol="application/pgp-signature"; micalg=pgp-sha512 Mime-Version: 1.0 (Mac OS X Mail 10.3 \(3273\)) Subject: Re: Large latency on blk_queue_enter Date: Mon, 8 May 2017 17:22:47 +0200 In-Reply-To: <991bbc1d-1849-94d9-5787-69a630e7e10d@kernel.dk> Cc: Ming Lei , Christoph Hellwig , Dan Williams , linux-block@vger.kernel.org, Linux Kernel Mailing List , =?utf-8?Q?Matias_Bj=C3=B8rling?= To: Jens Axboe References: <1656B440-3ECA-4F2B-B95C-418CF0F347E9@lightnvm.io> <20170508122738.GC5696@ming.t460p> <76E35BA3-FEC9-46D6-B36F-554F464FA9ED@lightnvm.io> <661d4b67-cf0c-a703-331b-ce24d75e782d@fb.com> <375D00C3-8B76-40FA-BB81-69829270BF5A@lightnvm.io> <576f9601-b0de-c636-8195-07e12fe99734@fb.com> <991bbc1d-1849-94d9-5787-69a630e7e10d@kernel.dk> List-ID: --Apple-Mail=_6B030D3D-F25D-4E13-A4B9-D68F9030134A Content-Transfer-Encoding: quoted-printable Content-Type: text/plain; charset=utf-8 Javier > On 8 May 2017, at 17.14, Jens Axboe wrote: >=20 > On 05/08/2017 09:08 AM, Jens Axboe wrote: >> On 05/08/2017 09:02 AM, Javier Gonz=C3=A1lez wrote: >>>> On 8 May 2017, at 16.52, Jens Axboe wrote: >>>>=20 >>>> On 05/08/2017 08:46 AM, Javier Gonz=C3=A1lez wrote: >>>>>> On 8 May 2017, at 16.23, Jens Axboe wrote: >>>>>>=20 >>>>>> On 05/08/2017 08:20 AM, Javier Gonz=C3=A1lez wrote: >>>>>>>> On 8 May 2017, at 16.13, Jens Axboe wrote: >>>>>>>>=20 >>>>>>>> On 05/08/2017 07:44 AM, Javier Gonz=C3=A1lez wrote: >>>>>>>>>> On 8 May 2017, at 14.27, Ming Lei = wrote: >>>>>>>>>>=20 >>>>>>>>>> On Mon, May 08, 2017 at 01:54:58PM +0200, Javier Gonz=C3=A1lez = wrote: >>>>>>>>>>> Hi, >>>>>>>>>>>=20 >>>>>>>>>>> I find an unusual added latency(~20-30ms) on blk_queue_enter = when >>>>>>>>>>> allocating a request directly from the NVMe driver through >>>>>>>>>>> nvme_alloc_request. I could use some help confirming that = this is a bug >>>>>>>>>>> and not an expected side effect due to something else. >>>>>>>>>>>=20 >>>>>>>>>>> I can reproduce this latency consistently on LightNVM when = mixing I/O >>>>>>>>>>> from pblk and I/O sent through an ioctl using liblightnvm, = but I don't >>>>>>>>>>> see anything on the LightNVM side that could impact the = request >>>>>>>>>>> allocation. >>>>>>>>>>>=20 >>>>>>>>>>> When I have a 100% read workload sent from pblk, the max. = latency is >>>>>>>>>>> constant throughout several runs at ~80us (which is normal = for the media >>>>>>>>>>> we are using at bs=3D4k, qd=3D1). All pblk I/Os reach the = nvme_nvm_submit_io >>>>>>>>>>> function on lightnvm.c., which uses nvme_alloc_request. When = we send a >>>>>>>>>>> command from user space through an ioctl, then the max = latency goes up >>>>>>>>>>> to ~20-30ms. This happens independently from the actual = command >>>>>>>>>>> (IN/OUT). I tracked down the added latency down to the call >>>>>>>>>>> percpu_ref_tryget_live in blk_queue_enter. Seems that the = queue >>>>>>>>>>> reference counter is not released as it should through = blk_queue_exit in >>>>>>>>>>> blk_mq_alloc_request. For reference, all ioctl I/Os reach = the >>>>>>>>>>> nvme_nvm_submit_user_cmd on lightnvm.c >>>>>>>>>>>=20 >>>>>>>>>>> Do you have any idea about why this might happen? I can dig = more into >>>>>>>>>>> it, but first I wanted to make sure that I am not missing = any obvious >>>>>>>>>>> assumption, which would explain the reference counter to be = held for a >>>>>>>>>>> longer time. >>>>>>>>>>=20 >>>>>>>>>> You need to check if the .q_usage_counter is working at = atomic mode. >>>>>>>>>> This counter is initialized as atomic mode, and finally = switchs to >>>>>>>>>> percpu mode via percpu_ref_switch_to_percpu() in = blk_register_queue(). >>>>>>>>>=20 >>>>>>>>> Thanks for commenting Ming. >>>>>>>>>=20 >>>>>>>>> The .q_usage_counter is not working on atomic mode. The queue = is >>>>>>>>> initialized normally through blk_register_queue() and the = counter is >>>>>>>>> switched to percpu mode, as you mentioned. As I understand it, = this is >>>>>>>>> how it should be, right? >>>>>>>>=20 >>>>>>>> That is how it should be, yes. You're not running with any = heavy >>>>>>>> debugging options, like lockdep or anything like that? >>>>>>>=20 >>>>>>> No lockdep, KASAN, kmemleak or any of the other usual suspects. >>>>>>>=20 >>>>>>> What's interesting is that it only happens when one of the I/Os = comes >>>>>>> from user space through the ioctl. If I have several pblk = instances on >>>>>>> the same device (which would end up allocating a new request in >>>>>>> parallel, potentially on the same core), the latency spike does = not >>>>>>> trigger. >>>>>>>=20 >>>>>>> I also tried to bind the read thread and the liblightnvm thread = issuing >>>>>>> the ioctl to different cores, but it does not help... >>>>>>=20 >>>>>> How do I reproduce this? Off the top of my head, and looking at = the code, >>>>>> I have no idea what is going on here. >>>>>=20 >>>>> Using LightNVM and liblightnvm [1] you can reproduce it by: >>>>>=20 >>>>> 1. Instantiate a pblk instance on the first channel (luns 0 - 7): >>>>> sudo nvme lnvm create -d nvme0n1 -n test0 -t pblk -b 0 -e 7 = -f >>>>> 2. Write 5GB to the test0 block device with a normal fio script >>>>> 3. Read 5GB to verify that latencies are good (max. ~80-90us at = bs=3D4k, qd=3D1) >>>>> 4. Re-run 3. and in parallel send a command through liblightnvm to = a >>>>> different channel. A simple command is an erase (erase block 900 = on >>>>> channel 2, lun 0): >>>>> sudo nvm_vblk line_erase /dev/nvme0n1 2 2 0 0 900 >>>>>=20 >>>>> After 4. you should see a ~25-30ms latency on the read workload. >>>>>=20 >>>>> I tried to reproduce the ioctl in a more generic way to reach >>>>> __nvme_submit_user_cmd(), but SPDK steals the whole device. Also, = qemu >>>>> is not reliable for this kind of performance testing. >>>>>=20 >>>>> If you have a suggestion on how I can mix an ioctl with normal = block I/O >>>>> read on a standard NVMe device, I'm happy to try it and see if I = can >>>>> reproduce the issue. >>>>=20 >>>> Just to rule out this being any hardware related delays in = processing >>>> IO: >>>>=20 >>>> 1) Does it reproduce with a simpler command, anything close to a = no-op >>>> that you can test? >>>=20 >>> Yes. I tried with a 4KB read and with a fake command I drop right = after >>> allocation. >>>=20 >>>> 2) What did you use to time the stall being blk_queue_enter()? >>>=20 >>> I have some debug code measuring time with ktime_get() in different >>> places in the stack, and among other places, around = blk_queue_enter(). I >>> use them then to measure max latency and expose it through sysfs. I = can >>> see that the latency peak is recorded in the probe before >>> blk_queue_enter() and not in the one after. >>>=20 >>> I also did an experiment, where the normal I/O path allocates the >>> request with BLK_MQ_REQ_NOWAIT. When running the experiment above, = the >>> read test fails since we reach: >>> if (nowait) >>> return -EBUSY; >>>=20 >>> in blk_queue_enter. >>=20 >> OK, that's starting to make more sense, that indicates that there is = indeed >> something wrong with the refs. Does the below help? >=20 > No, that can't be right, it does look balanced to begin with. > blk_mq_alloc_request() always grabs a queue ref, and always drops it. = If > we return with a request succesfully allocated, then we have an extra > ref on it, which is dropped when it is later freed. I agree, it seems more like a reference is put too late. I looked into into the places where the reference is put, but it all seems normal. In any case, I run it (just to see), and it did not help. > Something smells fishy, I'll dig a bit. Thanks! I continue looking into it myself; let me know if I can help with something more specific. Javier --Apple-Mail=_6B030D3D-F25D-4E13-A4B9-D68F9030134A Content-Transfer-Encoding: 7bit Content-Disposition: attachment; filename=signature.asc Content-Type: application/pgp-signature; name=signature.asc Content-Description: Message signed with OpenPGP -----BEGIN PGP SIGNATURE----- iQIcBAEBCgAGBQJZEI1IAAoJEGMfBTt1mRjKVcMP/i3mo47yq6DEZllIbMBei5LX wkwCMfjGIil3FBmjvaBsQp7g63y5IovpOTH40gjkobfzq+7q2wadlwCC2T3i8E3z 6Lx2YzXwpRFUeR600Ja+KWrq1nBtraW63Z4HqOb30HxFQdINYVwuSanzod92EShB lOD/kTw+ZLXfPB/ZqQoxkM4rpQ8ShH0JPRlizhofFvae/Q1KVMZsa5nMEFRG9LlM aNtsbetHVPWKj2S7vOUAKz5jXJ5qkXH+QcHQGSVABiaftPDfP7FpFwNOztC1nbO6 zaj24M69glOD95IB2477JcEds7lN3gZFl1EzwjYygFDmT0P3XilxHkjiOvaQcbAF p3eNQ786Xaf+v5jXv8j3DNwEoH15SJKsJgn5pdexyLiuRLZwbVUuI8Npd3x8a+Q1 B1PKPsWlMpU8zoj6MQnpp30/F9mcKvplvn0sPO5F/1SzCKQepUZxQVKez0jxZk0A hwlh22oXfMnrsZMliHk0+8BXm8XrXzXMG+8FVeyfQfl4T/gRaVp6pl13ZbmrUtDl 1Dw8fkyhGZUDFg7ftIEx/3bsvsBHeqG2ysNrAXzwOxLq5RgGdyD7X4qnF8exE+bS EBnPIjejacIbx9VVMUwhCqzBVFsvW70Xj78kUN0tcnj3OdKxOMZbhcrAgruCMbLA o/nf9DcsEhUlzz71EFSd =RRya -----END PGP SIGNATURE----- --Apple-Mail=_6B030D3D-F25D-4E13-A4B9-D68F9030134A--