On 1/26/16, 11:26 PM, "Nicholas A. Bellinger" <nab@xxxxxxxxxxxxxxx> wrote: >On Tue, 2016-01-26 at 18:33 +0000, Quinn Tran wrote: >> Roland, >> >> The BUG_ON in this code area is sign of early free and/or double free >> that is triggered in the LUN Reset path. The cmd_kref is being >> decrement too fast. In this case it looks like the cmd is on the way >> to be free (i.e. 2nd free), while a new incarnation of the command is >> being schedule to do new work. The cmd_in_wq flag tells us that the >> work element was scheduled or in the work queue by the new >> incarnation. The early free of the command was triggered in TCM. >> >> Please take a look at the following patches that Nicholas has >> submitted and Qlogic’s attempted to submit in the past. The "qla2xxx: >> Fix interaction issue between qla2xxx and Target Core Module” patch >> was asked to drop because it did not fit Bart’s implementation. >> >> >> QLogic’s patch >> [PATCH v2 08/16] qla2xxx: Replace QLA_TGT_STATE_ABORTED with a bit. >> >> http://marc.info/?l=linux-scsi&m=145038473432645&w=2 >> >> [PATCH 10/20] qla2xxx: Fix interaction issue between qla2xxx and Target Core Module >> >> http://marc.info/?l=linux-scsi&m=144953818431730&w=2 >> >> >> >> Nicholas’s Patch series or just patch 1 in the series. >> >> [PATCH 0/2] target: Fix LUN_RESET active I/O + TMR handling >> >> http://marc.info/?l=linux-scsi&m=145259439425172&w=2 >> >> [PATCH 1/2] target: Fix LUN_RESET active I/O handling for ACK_KREF >> >> http://marc.info/?l=linux-scsi&m=145259452525227&w=2 >> >> >> >> ---- >> >> Nicholas, >> >> Please consider "[PATCH 10/20] qla2xxx: Fix interaction issue between >> qla2xxx and Target Core Module” patch for the older branch or >> Pre-Bart. Thanks. >> > >Thanks Quinn. > >AFAICT, the QUEUE_FULL + tcm_qla2xxx_aborted_task() parts of this patch >are needed as separate driver bug-fixes, beyond the TCM multi-port >LUN_RESET fixes mentioned above. QT - We’re should be ok in the mentioned case. There is a backup mechanism in QLA that will copy the abort request and hold it on the side until HW is caught up. In the mean time, the original command can continue. The following commit is the backup mechanism. commit 33e7997755936ba92516c6ad69cd012c2e7d4dbb Author: Quinn Tran <quinn.tran@xxxxxxxxxx> Date: Thu Sep 25 06:14:55 2014 -0400 qla2xxx: Add support for QFull throttling and Term Exchange retry Through the qla target code, the qlt_send_term_exchange() routine is used in various different places to cleanup an exchange. For the case of IOCB request queue is full, the exchange is left unhandled/ dangling. Existing code does not have re-try logic to cleanup the exchange. This patch add retry logic to cleanup the exchange before letting new commands through. For the case of FW running out of exchanges, driver need to reply SAM_STAT_BUSY to the initiators. This patch add a pending queue for the busy reply in case IOCB queue is unable to handle the cmd. Cc: <stable@xxxxxxxxxxxxxxx> Signed-off-by: Quinn Tran <quinn.tran@xxxxxxxxxx> Signed-off-by: Saurav Kashyap <saurav.kashyap@xxxxxxxxxx> Signed-off-by: Christoph Hellwig <hch@xxxxxx> > >That said, I've applied this patch to target-pending/master as-is, minus >the extra kref debug as originally mentioned by HCH: > >https://git.kernel.org/cgit/linux/kernel/git/nab/target-pending.git/commit/?id=76dce933117e150ea9ddbce7c279808ec316a20a > >Please review. > QT - Thank you. It looks good. > ��.n��������+%������w��{.n����j� ����{ay� ʇڙ���f���h������_�(�階�ݢj"��������G����?���&��
