From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1751591AbeFEGtk (ORCPT ); Tue, 5 Jun 2018 02:49:40 -0400 Received: from aserp2120.oracle.com ([141.146.126.78]:46102 "EHLO aserp2120.oracle.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751273AbeFEGti (ORCPT ); Tue, 5 Jun 2018 02:49:38 -0400 Subject: Re: scsi/qla2xxx: BUG_ON(blk_queued_rq(req) is triggered in blk_finish_request To: "Madhani, Himanshu" Cc: Dept-Eng QLA2xxx Upstream , "jthumshirn@suse.de" , "linux-scsi@vger.kernel.org" , "linux-kernel@vger.kernel.org" , "jejb@linux.vnet.ibm.com" , "Martin K. Petersen" , Junxiao Bi References: <66FCA66D-62D6-4B3F-A41D-BF50082E7572@cavium.com> <104bc709-fd06-0465-2eaa-44154f9df437@oracle.com> <6edc16d5-da2c-9c94-9352-36ab70649c68@oracle.com> <2E3BD00B-A5FD-46AB-B085-5399EC02FC2A@cavium.com> <49bb2808-bd63-bb89-1c61-6cb1d53178d9@oracle.com> <27031631-a93f-e053-da84-3a11f78b3e62@oracle.com> From: "jianchao.wang" Message-ID: <0d076845-84c4-2f51-6fa5-4e80a3574ad1@oracle.com> Date: Tue, 5 Jun 2018 14:49:30 +0800 User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:52.0) Gecko/20100101 Thunderbird/52.7.0 MIME-Version: 1.0 In-Reply-To: <27031631-a93f-e053-da84-3a11f78b3e62@oracle.com> Content-Type: text/plain; charset=utf-8 Content-Language: en-US Content-Transfer-Encoding: 8bit X-Proofpoint-Virus-Version: vendor=nai engine=5900 definitions=8914 signatures=668702 X-Proofpoint-Spam-Details: rule=notspam policy=default score=0 suspectscore=2 malwarescore=0 phishscore=0 bulkscore=0 spamscore=0 mlxscore=0 mlxlogscore=999 adultscore=0 classifier=spam adjust=0 reason=mlx scancount=1 engine=8.0.1-1805220000 definitions=main-1806050083 Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Hi Himanshu I think the core issue here is: We should invoke scsi_done for the aborted request in qla2x00_sp_compl. The requeue or the other operation on the aborted request should be done by the SCSI mid-layer. Would you please give some comment here ? Thanks in advance Jianchao On 06/05/2018 09:11 AM, jianchao.wang wrote: > Hi Himanshu > > Would you please give some comment here? > > Thanks > Jianchao > > On 05/30/2018 11:28 AM, jianchao.wang wrote: >> Hi Himanshu >> >> Thanks for your kindly response. >> >> On 05/30/2018 01:44 AM, Madhani, Himanshu wrote: >>> Thanks for the information. I was out for couple days. Still working through my emails. >>> >>> Without core dump shared with us, things become hard to debug. We'll take a look at this data. >>> >>> Also, if you have any reproducer test case, Can you share that with us so that we can try to trigger It internally and see if we have better luck debugging that. >> >> This issue occurred in customer's environment twice till now, there is no reproduce steps. >> >> If you have time, would you please look at our analysis to see whether it is reasonable ? >> The kernel is not newest, but there should be the same issue in mainline with scsi + block legacy >> The key point here is: >> when the aborted command returns in irq context, it will invoke scsi_done, then blk_complete_request. >> however, the scsi recovery context could clear the ATOM_COMPLETE and requeue the request before irq >> context get it. >> >> Thanks >> Jianchao >> >>> >>> On 5/28/18, 6:11 PM, "jianchao.wang" wrote: >>> >>> Hi Himanshu >>> >>> do you need any other information ? >>> >>> Thanks >>> Jianchao >>> >>> On 05/25/2018 02:48 PM, jianchao.wang wrote: >>> > Hi Himanshu >>> > >>> > I'm afraid I cannot provide you the vmcore file, it is from our customer. >>> > If any information needed in the vmcore, I could provide with you. >>> > >>> > In this scene, there are two contexts: >>> > 1. the irq context, scsi_done() from qla2x00_sp_compl() >>> > 2. the kworker context, scmd_eh_abort_handler which invokes the >>> > qla2xxx_eh_abort() >>> > >>> > Let's see the irq context. >>> > scsi_done() will invoke the blk_complete_request(). >>> > In common case, the REQ_ATOM_COMPLETE　should have been set by >>> > blk_rq_check_expired() when the request was confirmed timeout. So >>> > __blk_complete_request() will not be invoked. >>> > >>> > On the other hand, in the kworker context, scmd_eh_abort_handler() will do some >>> > other things on this aborted request. >>> > >>> > rtn = scsi_try_to_abort_cmd(sdev->host->hostt, scmd); >>> > if (rtn == SUCCESS) { >>> > set_host_byte(scmd, DID_TIME_OUT) >>> > if (scsi_host_eh_past_deadline(sdev->host)) { >>> > >>>> >>> > } else if (!scsi_noretry_cmd(scmd) && >>> > (++scmd->retries <= scmd->allowed)) { >>> > scsi_queue_insert(scmd, SCSI_MLQUEUE_EH_RETRY) >>> > }>>>> >>> >>>>> >>> > } >>> > The host type of the result will set to DID_TIME_OUT. scsi_noretry_cmd() will >>> > return false and scsi_queue_insert() will be invoked, finally the >>> > blk_requeue_request(). >>> > >>> > blk_requeue_request() will clear the REQ_ATOM_COMPLETE and queue the request on >>> > request_queue->queue_head. >>> > >>> > There could be a race between these two contexts: >>> > what if the blk_clear_rq_complete() in kworker context is invoked before the >>> > blk_mark_rq_complete() in irq context ? >>> > >>> > BLK_SOFTIRQ will be raised by __blk_complete_request() in irq context. Then the >>> > scsi_softirq_done() will be invoked. At the moment, the request has been >>> > requeued by kworker context and scsi_cmnd->result is set 0. On the cpu of the >>> > irq context, the blk_finish_request() will be invoked finally, and panic comes >>> > up due to BUG_ON(blk_queued_rq(req) >>> > >>> > Here is the dmesg log. >>> > [4937745.180213] qla2xxx [0000:92:00.1]-801c:4: Abort command issued nexus=4:5:5 -- 1 2002. >>> > [4937745.180655] qla2xxx [0000:92:00.1]-801c:4: Abort command issued nexus=4:4:73 -- 0 2002. >>> > [4937745.181059] qla2xxx [0000:92:00.1]-801c:4: Abort command issued nexus=4:5:5 -- 1 2002. >>> > [4937745.181514] qla2xxx [0000:92:00.1]-801c:4: Abort command issued nexus=4:4:73 -- 0 2002. >>> > [4937745.181636] ------------[ cut here ]------------ >>> > [4937745.181900] kernel BUG at block/blk-core.c:2601! >>> > [4937745.182095] qla2xxx [0000:92:00.1]-801c:4: Abort command issued nexus=4:5:5 -- 1 2002. >>> > [4937745.182338] qla2xxx [0000:92:00.1]-801c:4: Abort command issued nexus=4:4:73 -- 1 2002. >>> > [4937745.183046] qla2xxx [0000:92:00.1]-801c:4: Abort command issued nexus=4:5:5 -- 1 2002. >>> > [4937745.183213] qla2xxx [0000:92:00.1]-801c:4: Abort command issued nexus=4:4:73 -- 1 2002. >>> > [4937745.183844] qla2xxx [0000:92:00.1]-801c:4: Abort command issued nexus=4:5:5 -- 1 2002. >>> > [4937745.184038] qla2xxx [0000:92:00.1]-801c:4: Abort command issued nexus=4:4:73 -- 1 2002. >>> > [4937745.184919] invalid opcode: 0000 [#1] SMP >>> > [4937745.185173] Modules linked in: ... >>> > [4937745.188237] CPU: 18 PID: 0 Comm: swapper/18 Not tainted 4.1.12-61.1.9.el6uek.x86_64 #2 >>> > [4937745.188691] Hardware name: Oracle Corporation SUN BLADE X4-2B /BD,ASSY,PCB,X4-2B , BIOS 28010200 01/16/2014 >>> > [4937745.189157] task: ffff880227fa9c00 ti: ffff880227fc0000 task.ti: ffff880227fc0000 >>> > [4937745.189607] RIP: e030:[] [] blk_finish_request+0xd0/0xe0 >>> > [4937745.190104] RSP: e02b:ffff88022d483d78 EFLAGS: 00010083 >>> > [4937745.190339] RAX: 0000000000000280 RBX: ffff8801c4508730 RCX: 0000000000000280 >>> > [4937745.190791] RDX: 0000000000008b5a RSI: 0000000000000000 RDI: ffff8801c4508730 >>> > [4937745.191246] RBP: ffff88022d483d88 R08: 0000000000000000 R09: 0000000000004000 >>> > [4937745.191694] R10: 000000000000000b R11: 0000000000007ffe R12: 0000000000000000 >>> > [4937745.192153] R13: ffff88022374d000 R14: ffff880184cb88c0 R15: ffff8800536ccc90 >>> > [4937745.197930] FS: 0000000000000000(0000) GS:ffff88022d480000(0000) knlGS:ffff88022d480000 >>> > [4937745.198389] CS: e033 DS: 002b ES: 002b CR0: 0000000080050033 >>> > [4937745.198627] CR2: 00000000016e1f18 CR3: 000000010caed000 CR4: 0000000000042660 >>> > [4937745.199080] Stack: >>> > [4937745.199309] ffff8801c4508730 0000000000000000 ffff88022d483dd8 ffffffff814ac82f >>> > [4937745.199790] ffff88022d483da8 0000000000000280 ffff88022d483dd8 ffff880184cb88c0 >>> > [4937745.200290] ffff8801c4508730 0000000000000000 0000000000000000 0000000000000000 >>> > [4937745.200745] Call Trace: >>> > [4937745.200968] >>> > [4937745.200986] [] scsi_end_request+0x11f/0x1e0 >>> > [4937745.201422] [] scsi_io_completion+0xac/0x500 >>> > [4937745.201651] [] scsi_finish_command+0xf0/0x150 >>> > [4937745.201875] [] scsi_softirq_done+0x14a/0x160 >>> > [4937745.202103] [] blk_done_softirq+0x86/0xa0 >>> > [4937745.202333] [] __do_softirq+0x10a/0x350 >>> > [4937745.202558] [] irq_exit+0x125/0x130 >>> > [4937745.202787] [] xen_evtchn_do_upcall+0x39/0x50 >>> > [4937745.203020] [] xen_do_hypervisor_callback+0x1e/0x40 >>> > [4937745.203247] >>> > [4937745.203263] [] ? xen_hypercall_sched_op+0xa/0x20 >>> > [4937745.203703] [] ? xen_hypercall_sched_op+0xa/0x20 >>> > [4937745.203930] [] ? xen_safe_halt+0x10/0x20 >>> > [4937745.204160] [] ? default_idle+0x1e/0x100 >>> > [4937745.204387] [] ? arch_cpu_idle+0xf/0x20 >>> > [4937745.204614] [] ? cpuidle_idle_call+0x135/0x1e0 >>> > [4937745.204841] [] ? cpu_idle_loop+0x1f5/0x220 >>> > [4937745.205070] [] ? cpu_startup_entry+0x1b/0x70 >>> > [4937745.205294] [] ? cpu_startup_entry+0x5f/0x70 >>> > [4937745.205524] [] ? cpu_bringup_and_idle+0x2a/0x40 >>> > [4937745.205749] Code: 89 de e8 a4 14 00 00 e9 6b ff ff ff 83 7b 48 01 0f 85 >>> > 78 ff ff ff 48 8b 7b 30 48 81 c7 98 01 00 00 e8 75 6b ea ff e9 63 ff ff ff >>> > <0f> 0b eb fe 66 66 66 2e 0f 1f 84 00 00 00 00 00 55 48 89 e5 48 >>> > [4937745.206811] RIP [] blk_finish_request+0xd0/0xe0 >>> > [4937745.207046] RSP >>> > [4937745.207687] ---[ end trace 0b43a57bdab9464f ]--- >>> > [4937745.207693] ------------[ cut here ]------------ >>> > >>> > On 05/25/2018 02:30 PM, Madhani, Himanshu wrote: >>> >> Hi Jianchao, >>> >> >>> >> Do you have any debug logs collected for us to look at this in details. >>> >> >>> >> Can you provide me crash/vmlinux/modules for details analysis. >>> >> >>> >> Thanks, >>> >> himanshu >>> >> >>> >> On 5/24/18, 6:49 AM, "Madhani, Himanshu" wrote: >>> >> >>> >> >>> >> > On May 24, 2018, at 2:09 AM, jianchao.wang wrote: >>> >> > >>> >> > Hi Himanshu >>> >> > >>> >> > Thanks god get response from your side finally :) >>> >> > I get qla2xxx-upstream@qlogic.com from the maintainer information of driver/scsi/qla2xxx >>> >> > Looks like it should be modified. >>> >> > >>> >> Sorry for delay. I do try to respond quickly to any emails received on that ID as well. >>> >> >>> >> We’ll keep it for now and I’ll look into updating it, if needed. >>> >> >>> >> > Thanks >>> >> > Jianchao >>> >> > >>> >> > On 05/24/2018 12:21 PM, Madhani, Himanshu wrote: >>> >> >> Hi, >>> >> >> >>> >> >> Yes. We'll look at this issue. >>> >> >> >>> >> >> Thanks, >>> >> >> Himanshu >>> >> >> >>> >> >>> -----Original Message----- >>> >> >>> From: jianchao.wang [mailto:jianchao.w.wang@oracle.com] >>> >> >>> Sent: Wednesday, May 23, 2018 6:51 PM >>> >> >>> To: Dept-Eng QLA2xxx Upstream ; Madhani, >>> >> >>> Himanshu ; jthumshirn@suse.de >>> >> >>> Cc: linux-scsi@vger.kernel.org; linux-kernel@vger.kernel.org; >>> >> >>> jejb@linux.vnet.ibm.com; Martin K. Petersen ; >>> >> >>> Junxiao Bi >>> >> >>> Subject: Re: BUG: scsi/qla2xxx: BUG_ON(blk_queued_rq(req) is triggered in >>> >> >>> blk_finish_request >>> >> >>> >>> >> >>> Would anyone please take a look at this ? >>> >> >>> >>> >> >>> Thanks in advance >>> >> >>> Jianchao >>> >> >>> >>> >> >>> On 05/23/2018 11:55 AM, jianchao.wang wrote: >>> >> >>>> >>> >> >>>> >>> >> >>>> Hi all >>> >> >>>> >>> >> >>>> Our customer met a panic triggered by BUG_ON in blk_finish_request. >>> >> >>>>> From the dmesg log, the BUG_ON was triggered after command abort >>> >> >>> occurred many times. >>> >> >>>> There is a race condition in the following scenario. >>> >> >>>> >>> >> >>>> cpu A cpu B >>> >> >>>> kworker interrupt >>> >> >>>> >>> >> >>>> scmd_eh_abort_handler() >>> >> >>>> -> scsi_try_to_abort_cmd() >>> >> >>>> -> qla2xxx_eh_abort() >>> >> >>>> -> qla2x00_eh_wait_on_command() qla2x00_status_entry() >>> >> >>>> -> qla2x00_sp_compl() >>> >> >>>> -> qla2x00_sp_free_dma() >>> >> >>>> -> scsi_queue_insert() >>> >> >>>> -> __scsi_queue_insert() >>> >> >>>> -> blk_requeue_request() >>> >> >>>> -> blk_clear_rq_complete() >>> >> >>>> -> scsi_done >>> >> >>>> -> blk_complete_request >>> >> >>>> -> blk_mark_rq_complete >>> >> >>>> -> elv_requeue_request() -> __blk_complete_request() >>> >> >>>> -> __elv_add_request() >>> >> >>>> // req will be queued here >>> >> >>>> BLK_SOFTIRQ >>> >> >>>> scsi_softirq_done() >>> >> >>>> -> scsi_finish_command() >>> >> >>>> -> scsi_io_completion() >>> >> >>>> -> scsi_end_request() >>> >> >>>> -> blk_finish_request() // >>> >> >>> BUG_ON(blk_queued_rq(req)) !!! >>> >> >>>> >>> >> >>>> The issue will not be triggered most of time, because the request is marked as >>> >> >>> complete by timeout path. >>> >> >>>> So the scsi_done from qla2x00_sp_compl does nothing. >>> >> >>>> But as the scenario above, if the complete state has been cleaned by >>> >> >>> blk_requeue_request, we will get >>> >> >>>> the request both requeued and completed, and then >>> >> >>> BUG_ON(blk_queued_rq(req)) in blk_finish_request comes up. >>> >> >>>> >>> >> >>>> Is there any solution for this in qla2xxx driver side ? >>> >> >>>> >>> >> >>>> Thanks >>> >> >>>> Jianchao >>> >> >>>> >>> >> >>>> >>> >> >>> >> Thanks, >>> >> - Himanshu >>> >> >>> >> >>> >> >>> > >>> >>> >