diff options
author | Martin K. Petersen <martin.petersen@oracle.com> | 2023-09-22 04:44:12 +0300 |
---|---|---|
committer | Martin K. Petersen <martin.petersen@oracle.com> | 2023-09-22 04:44:12 +0300 |
commit | fe15880f317348ba9e81c46f270d1c1ee0002e8e (patch) | |
tree | ab44a7af7f67655724ce2f54672626846382027a /drivers/scsi/lpfc/lpfc_nvme.c | |
parent | 17d11949827375d74bdf5df18c3f5813ee7d5933 (diff) | |
parent | 80975adc79dde8985f4ea68fe0b158bc8a913580 (diff) | |
download | linux-fe15880f317348ba9e81c46f270d1c1ee0002e8e.tar.xz |
Merge patch series "scsi: pm8001: Bug fix and cleanup"
Damien Le Moal <dlemoal@kernel.org> says:
The first patch of this series fixes an issue with IRQ setup which
prevents the controller from resuming after a system suspend. The
following patches are code cleanup without any functional changes.
[mkp: The first patch went into v6.6-rc2 and thus this merge
constitutes the remaining patches of the series]
Link: https://lore.kernel.org/r/20230911232745.325149-1-dlemoal@kernel.org
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Diffstat (limited to 'drivers/scsi/lpfc/lpfc_nvme.c')
-rw-r--r-- | drivers/scsi/lpfc/lpfc_nvme.c | 24 |
1 files changed, 17 insertions, 7 deletions
diff --git a/drivers/scsi/lpfc/lpfc_nvme.c b/drivers/scsi/lpfc/lpfc_nvme.c index 39acbcb7ec66..96e11a26c297 100644 --- a/drivers/scsi/lpfc/lpfc_nvme.c +++ b/drivers/scsi/lpfc/lpfc_nvme.c @@ -228,8 +228,7 @@ lpfc_nvme_remoteport_delete(struct nvme_fc_remote_port *remoteport) spin_unlock_irq(&ndlp->lock); /* On a devloss timeout event, one more put is executed provided the - * NVME and SCSI rport unregister requests are complete. If the vport - * is unloading, this extra put is executed by lpfc_drop_node. + * NVME and SCSI rport unregister requests are complete. */ if (!(ndlp->fc4_xpt_flags & fc4_xpt_flags)) lpfc_disc_state_machine(vport, ndlp, NULL, NLP_EVT_DEVICE_RM); @@ -2567,11 +2566,7 @@ lpfc_nvme_rescan_port(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp) * nvme_transport perspective. Loss of an rport just means IO cannot * be sent and recovery is completely up to the initator. * For now, the driver just unbinds the DID and port_role so that - * no further IO can be issued. Changes are planned for later. - * - * Notes - the ndlp reference count is not decremented here since - * since there is no nvme_transport api for devloss. Node ref count - * is only adjusted in driver unload. + * no further IO can be issued. */ void lpfc_nvme_unregister_port(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp) @@ -2646,6 +2641,21 @@ lpfc_nvme_unregister_port(struct lpfc_vport *vport, struct lpfc_nodelist *ndlp) "6167 NVME unregister failed %d " "port_state x%x\n", ret, remoteport->port_state); + + if (vport->load_flag & FC_UNLOADING) { + /* Only 1 thread can drop the initial node + * reference. Check if another thread has set + * NLP_DROPPED. + */ + spin_lock_irq(&ndlp->lock); + if (!(ndlp->nlp_flag & NLP_DROPPED)) { + ndlp->nlp_flag |= NLP_DROPPED; + spin_unlock_irq(&ndlp->lock); + lpfc_nlp_put(ndlp); + return; + } + spin_unlock_irq(&ndlp->lock); + } } } return; |