lkml.org 
[lkml]   [2020]   [Sep]   [17]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
    Patch in this message
    /
    From
    Subject[PATCH AUTOSEL 5.4 089/330] scsi: lpfc: Fix incomplete NVME discovery when target
    Date
    From: James Smart <jsmart2021@gmail.com>

    [ Upstream commit be0709e449ac9d9753a5c17e5b770d6e5e930e4a ]

    NVMe device re-discovery does not complete. Dev_loss_tmo messages seen on
    initiator after recovery from a link disturbance.

    The failing case is the following:

    When the driver (as a NVME target) receives a PLOGI, the driver initiates
    an "unreg rpi" mailbox command. While the mailbox command is in progress,
    the driver requests that an ACC be sent to the initiator. The target's ACC
    is received by the initiator and the initiator then transmits a PLOGI. The
    driver receives the PLOGI prior to receiving the completion for the PLOGI
    response WQE that sent the ACC. (Different delivery sources from the hw so
    the race is very possible). Given the PLOGI is prior to the ACC completion
    (signifying PLOGI exchange complete), the driver LS_RJT's the PRLI. The
    "unreg rpi" mailbox then completes. Since PRLI has been received, the
    driver transmits a PLOGI to restart discovery, which the initiator then
    ACC's. If the driver processes the (re)PLOGI ACC prior to the completing
    the handling for the earlier ACC it sent the intiators original PLOGI,
    there is no state change for completion of the (re)PLOGI. The ndlp remains
    in "PLOGI Sent" and the initiator continues sending PRLI's which are
    rejected by the target until timeout or retry is reached.

    Fix by: When in target mode, defer sending an ACC for the received PLOGI
    until unreg RPI completes.

    Link: https://lore.kernel.org/r/20191218235808.31922-2-jsmart2021@gmail.com
    Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
    Signed-off-by: James Smart <jsmart2021@gmail.com>
    Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
    Signed-off-by: Sasha Levin <sashal@kernel.org>
    ---
    drivers/scsi/lpfc/lpfc_nportdisc.c | 108 ++++++++++++++++++++++++++---
    1 file changed, 99 insertions(+), 9 deletions(-)

    diff --git a/drivers/scsi/lpfc/lpfc_nportdisc.c b/drivers/scsi/lpfc/lpfc_nportdisc.c
    index 2a340624bfc99..590a49e847626 100644
    --- a/drivers/scsi/lpfc/lpfc_nportdisc.c
    +++ b/drivers/scsi/lpfc/lpfc_nportdisc.c
    @@ -308,7 +308,7 @@ lpfc_defer_pt2pt_acc(struct lpfc_hba *phba, LPFC_MBOXQ_t *link_mbox)
    mb->mbxStatus);
    mempool_free(login_mbox, phba->mbox_mem_pool);
    mempool_free(link_mbox, phba->mbox_mem_pool);
    - lpfc_sli_release_iocbq(phba, save_iocb);
    + kfree(save_iocb);
    return;
    }

    @@ -325,7 +325,61 @@ lpfc_defer_pt2pt_acc(struct lpfc_hba *phba, LPFC_MBOXQ_t *link_mbox)
    }

    mempool_free(link_mbox, phba->mbox_mem_pool);
    - lpfc_sli_release_iocbq(phba, save_iocb);
    + kfree(save_iocb);
    +}
    +
    +/**
    + * lpfc_defer_tgt_acc - Progress SLI4 target rcv PLOGI handler
    + * @phba: Pointer to HBA context object.
    + * @pmb: Pointer to mailbox object.
    + *
    + * This function provides the unreg rpi mailbox completion handler for a tgt.
    + * The routine frees the memory resources associated with the completed
    + * mailbox command and transmits the ELS ACC.
    + *
    + * This routine is only called if we are SLI4, acting in target
    + * mode and the remote NPort issues the PLOGI after link up.
    + **/
    +void
    +lpfc_defer_acc_rsp(struct lpfc_hba *phba, LPFC_MBOXQ_t *pmb)
    +{
    + struct lpfc_vport *vport = pmb->vport;
    + struct lpfc_nodelist *ndlp = pmb->ctx_ndlp;
    + LPFC_MBOXQ_t *mbox = pmb->context3;
    + struct lpfc_iocbq *piocb = NULL;
    + int rc;
    +
    + if (mbox) {
    + pmb->context3 = NULL;
    + piocb = mbox->context3;
    + mbox->context3 = NULL;
    + }
    +
    + /*
    + * Complete the unreg rpi mbx request, and update flags.
    + * This will also restart any deferred events.
    + */
    + lpfc_nlp_get(ndlp);
    + lpfc_sli4_unreg_rpi_cmpl_clr(phba, pmb);
    +
    + if (!piocb) {
    + lpfc_printf_vlog(vport, KERN_ERR, LOG_DISCOVERY | LOG_ELS,
    + "4578 PLOGI ACC fail\n");
    + if (mbox)
    + mempool_free(mbox, phba->mbox_mem_pool);
    + goto out;
    + }
    +
    + rc = lpfc_els_rsp_acc(vport, ELS_CMD_PLOGI, piocb, ndlp, mbox);
    + if (rc) {
    + lpfc_printf_vlog(vport, KERN_ERR, LOG_DISCOVERY | LOG_ELS,
    + "4579 PLOGI ACC fail %x\n", rc);
    + if (mbox)
    + mempool_free(mbox, phba->mbox_mem_pool);
    + }
    + kfree(piocb);
    +out:
    + lpfc_nlp_put(ndlp);
    }

    static int
    @@ -345,6 +399,7 @@ lpfc_rcv_plogi(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp,
    struct lpfc_iocbq *save_iocb;
    struct ls_rjt stat;
    uint32_t vid, flag;
    + u16 rpi;
    int rc, defer_acc;

    memset(&stat, 0, sizeof (struct ls_rjt));
    @@ -488,7 +543,7 @@ lpfc_rcv_plogi(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp,
    link_mbox->vport = vport;
    link_mbox->ctx_ndlp = ndlp;

    - save_iocb = lpfc_sli_get_iocbq(phba);
    + save_iocb = kzalloc(sizeof(*save_iocb), GFP_KERNEL);
    if (!save_iocb)
    goto out;
    /* Save info from cmd IOCB used in rsp */
    @@ -513,7 +568,36 @@ lpfc_rcv_plogi(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp,
    goto out;

    /* Registering an existing RPI behaves differently for SLI3 vs SLI4 */
    - if (phba->sli_rev == LPFC_SLI_REV4)
    + if (phba->nvmet_support && !defer_acc) {
    + link_mbox = mempool_alloc(phba->mbox_mem_pool, GFP_KERNEL);
    + if (!link_mbox)
    + goto out;
    +
    + /* As unique identifiers such as iotag would be overwritten
    + * with those from the cmdiocb, allocate separate temporary
    + * storage for the copy.
    + */
    + save_iocb = kzalloc(sizeof(*save_iocb), GFP_KERNEL);
    + if (!save_iocb)
    + goto out;
    +
    + /* Unreg RPI is required for SLI4. */
    + rpi = phba->sli4_hba.rpi_ids[ndlp->nlp_rpi];
    + lpfc_unreg_login(phba, vport->vpi, rpi, link_mbox);
    + link_mbox->vport = vport;
    + link_mbox->ctx_ndlp = ndlp;
    + link_mbox->mbox_cmpl = lpfc_defer_acc_rsp;
    +
    + if (((ndlp->nlp_DID & Fabric_DID_MASK) != Fabric_DID_MASK) &&
    + (!(vport->fc_flag & FC_OFFLINE_MODE)))
    + ndlp->nlp_flag |= NLP_UNREG_INP;
    +
    + /* Save info from cmd IOCB used in rsp */
    + memcpy(save_iocb, cmdiocb, sizeof(*save_iocb));
    +
    + /* Delay sending ACC till unreg RPI completes. */
    + defer_acc = 1;
    + } else if (phba->sli_rev == LPFC_SLI_REV4)
    lpfc_unreg_rpi(vport, ndlp);

    rc = lpfc_reg_rpi(phba, vport->vpi, icmd->un.rcvels.remoteID,
    @@ -553,6 +637,9 @@ lpfc_rcv_plogi(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp,
    if ((vport->port_type == LPFC_NPIV_PORT &&
    vport->cfg_restrict_login)) {

    + /* no deferred ACC */
    + kfree(save_iocb);
    +
    /* In order to preserve RPIs, we want to cleanup
    * the default RPI the firmware created to rcv
    * this ELS request. The only way to do this is
    @@ -571,8 +658,12 @@ lpfc_rcv_plogi(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp,
    }
    if (defer_acc) {
    /* So the order here should be:
    - * Issue CONFIG_LINK mbox
    - * CONFIG_LINK cmpl
    + * SLI3 pt2pt
    + * Issue CONFIG_LINK mbox
    + * CONFIG_LINK cmpl
    + * SLI4 tgt
    + * Issue UNREG RPI mbx
    + * UNREG RPI cmpl
    * Issue PLOGI ACC
    * PLOGI ACC cmpl
    * Issue REG_LOGIN mbox
    @@ -596,10 +687,9 @@ lpfc_rcv_plogi(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp,
    out:
    if (defer_acc)
    lpfc_printf_log(phba, KERN_ERR, LOG_DISCOVERY,
    - "4577 pt2pt discovery failure: %p %p %p\n",
    + "4577 discovery failure: %p %p %p\n",
    save_iocb, link_mbox, login_mbox);
    - if (save_iocb)
    - lpfc_sli_release_iocbq(phba, save_iocb);
    + kfree(save_iocb);
    if (link_mbox)
    mempool_free(link_mbox, phba->mbox_mem_pool);
    if (login_mbox)
    --
    2.25.1
    \
     
     \ /
      Last update: 2020-09-18 05:11    [W:4.908 / U:1.068 seconds]
    ©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site