[SCSI] lpfc 8.3.19: Added support for ELS RRQ command

Added support for ELS RRQ command

- Add new routine lpfc_set_rrq_active() to track XRI qualifier state.
- Add new module parameter lpfc_enable_rrq to control RRQ operation.
- Add logic to ELS RRQ completion handler and xri qualifier timeout
  to clear XRI qualifier state.
- Use OX_ID from XRI_ABORTED_CQE for RRQ payload.
- Tie abort and XRI_ABORTED_CQE andler to RRQ generation.
Signed-off-by: Alex Iannicelli <alex.iannicelli@emulex.com>
Signed-off-by: James Smart <james.smart@emulex.com>
Signed-off-by: James Bottomley <James.Bottomley@suse.de>

drivers/scsi/lpfc/lpfc.h

@@ -486,7 +486,7 @@ struct lpfc_hba {
 	int (*lpfc_new_scsi_buf)
 		(struct lpfc_vport *, int);
 	struct lpfc_scsi_buf * (*lpfc_get_scsi_buf)
-		(struct lpfc_hba *);
+		(struct lpfc_hba *, struct lpfc_nodelist *);
 	int (*lpfc_scsi_prep_dma_buf)
 		(struct lpfc_hba *, struct lpfc_scsi_buf *);
 	void (*lpfc_scsi_unprep_dma_buf)
@@ -574,6 +574,7 @@ struct lpfc_hba {
 #define HBA_FIP_SUPPORT		0x800 /* FIP support in HBA */
 #define HBA_AER_ENABLED		0x1000 /* AER enabled with HBA */
 #define HBA_DEVLOSS_TMO         0x2000 /* HBA in devloss timeout */
+#define HBA_RRQ_ACTIVE		0x4000 /* process the rrq active list */
 	uint32_t fcp_ring_in_use; /* When polling test if intr-hndlr active*/
 	struct lpfc_dmabuf slim2p;
 
@@ -623,6 +624,7 @@ struct lpfc_hba {
 	/* HBA Config Parameters */
 	uint32_t cfg_ack0;
 	uint32_t cfg_enable_npiv;
+	uint32_t cfg_enable_rrq;
 	uint32_t cfg_topology;
 	uint32_t cfg_link_speed;
 	uint32_t cfg_cr_delay;
@@ -733,6 +735,7 @@ struct lpfc_hba {
 	uint32_t total_scsi_bufs;
 	struct list_head lpfc_iocb_list;
 	uint32_t total_iocbq_bufs;
+	struct list_head active_rrq_list;
 	spinlock_t hbalock;
 
 	/* pci_mem_pools */
@@ -745,6 +748,7 @@ struct lpfc_hba {
 
 	mempool_t *mbox_mem_pool;
 	mempool_t *nlp_mem_pool;
+	mempool_t *rrq_pool;
 
 	struct fc_host_statistics link_stats;
 	enum intr_type_t intr_type;
@@ -801,6 +805,7 @@ struct lpfc_hba {
 	unsigned long skipped_hb;
 	struct timer_list hb_tmofunc;
 	uint8_t hb_outstanding;
+	struct timer_list rrq_tmr;
 	enum hba_temp_state over_temp_state;
 	/* ndlp reference management */
 	spinlock_t ndlp_lock;

drivers/scsi/lpfc/lpfc_attr.c

@@ -1977,6 +1977,13 @@ lpfc_param_show(enable_npiv);
 lpfc_param_init(enable_npiv, 1, 0, 1);
 static DEVICE_ATTR(lpfc_enable_npiv, S_IRUGO, lpfc_enable_npiv_show, NULL);
 
+int lpfc_enable_rrq;
+module_param(lpfc_enable_rrq, int, 0);
+MODULE_PARM_DESC(lpfc_enable_rrq, "Enable RRQ functionality");
+lpfc_param_show(enable_rrq);
+lpfc_param_init(enable_rrq, 0, 0, 1);
+static DEVICE_ATTR(lpfc_enable_rrq, S_IRUGO, lpfc_enable_rrq_show, NULL);
+
 /*
 # lpfc_suppress_link_up:  Bring link up at initialization
 #            0x0  = bring link up (issue MBX_INIT_LINK)
@@ -3394,6 +3401,7 @@ struct device_attribute *lpfc_hba_attrs[] = {
 	&dev_attr_lpfc_fdmi_on,
 	&dev_attr_lpfc_max_luns,
 	&dev_attr_lpfc_enable_npiv,
+	&dev_attr_lpfc_enable_rrq,
 	&dev_attr_nport_evt_cnt,
 	&dev_attr_board_mode,
 	&dev_attr_max_vpi,
@@ -4610,6 +4618,7 @@ lpfc_get_cfgparam(struct lpfc_hba *phba)
 	lpfc_link_speed_init(phba, lpfc_link_speed);
 	lpfc_poll_tmo_init(phba, lpfc_poll_tmo);
 	lpfc_enable_npiv_init(phba, lpfc_enable_npiv);
+	lpfc_enable_rrq_init(phba, lpfc_enable_rrq);
 	lpfc_use_msi_init(phba, lpfc_use_msi);
 	lpfc_fcp_imax_init(phba, lpfc_fcp_imax);
 	lpfc_fcp_wq_count_init(phba, lpfc_fcp_wq_count);

drivers/scsi/lpfc/lpfc_crtn.h

@@ -416,5 +416,13 @@ struct lpfc_iocbq *lpfc_sli_ringtx_get(struct lpfc_hba *,
 int __lpfc_sli_issue_iocb(struct lpfc_hba *, uint32_t,
 	struct lpfc_iocbq *, uint32_t);
 uint32_t lpfc_drain_txq(struct lpfc_hba *);
-
-
+void lpfc_clr_rrq_active(struct lpfc_hba *, uint16_t, struct lpfc_node_rrq *);
+int lpfc_test_rrq_active(struct lpfc_hba *, struct lpfc_nodelist *, uint16_t);
+void lpfc_handle_rrq_active(struct lpfc_hba *);
+int lpfc_send_rrq(struct lpfc_hba *, struct lpfc_node_rrq *);
+int lpfc_set_rrq_active(struct lpfc_hba *, struct lpfc_nodelist *,
+	uint16_t, uint16_t, uint16_t);
+void lpfc_cleanup_wt_rrqs(struct lpfc_hba *);
+void lpfc_cleanup_vports_rrqs(struct lpfc_vport *);
+struct lpfc_node_rrq *lpfc_get_active_rrq(struct lpfc_vport *, uint16_t,
+	uint32_t);

drivers/scsi/lpfc/lpfc_disc.h

@@ -68,6 +68,12 @@ struct lpfc_fast_path_event {
 	} un;
 };
 
+#define LPFC_SLI4_MAX_XRI	1024	/* Used to make the ndlp's xri_bitmap */
+#define XRI_BITMAP_ULONGS (LPFC_SLI4_MAX_XRI / BITS_PER_LONG)
+struct lpfc_node_rrqs {
+	unsigned long xri_bitmap[XRI_BITMAP_ULONGS];
+};
+
 struct lpfc_nodelist {
 	struct list_head nlp_listp;
 	struct lpfc_name nlp_portname;
@@ -110,8 +116,19 @@ struct lpfc_nodelist {
 	atomic_t cmd_pending;
 	uint32_t cmd_qdepth;
 	unsigned long last_change_time;
+	struct lpfc_node_rrqs active_rrqs;
 	struct lpfc_scsicmd_bkt *lat_data;	/* Latency data */
 };
+struct lpfc_node_rrq {
+	struct list_head list;
+	uint16_t xritag;
+	uint16_t send_rrq;
+	uint16_t rxid;
+	uint32_t         nlp_DID;		/* FC D_ID of entry */
+	struct lpfc_vport *vport;
+	struct lpfc_nodelist *ndlp;
+	unsigned long rrq_stop_time;
+};
 
 /* Defines for nlp_flag (uint32) */
 #define NLP_IGNR_REG_CMPL  0x00000001 /* Rcvd rscn before we cmpl reg login */

drivers/scsi/lpfc/lpfc_hbadisc.c

@@ -607,6 +607,8 @@ lpfc_work_done(struct lpfc_hba *phba)
 
 	/* Process SLI4 events */
 	if (phba->pci_dev_grp == LPFC_PCI_DEV_OC) {
+		if (phba->hba_flag & HBA_RRQ_ACTIVE)
+			lpfc_handle_rrq_active(phba);
 		if (phba->hba_flag & FCP_XRI_ABORT_EVENT)
 			lpfc_sli4_fcp_xri_abort_event_proc(phba);
 		if (phba->hba_flag & ELS_XRI_ABORT_EVENT)
@@ -966,6 +968,7 @@ lpfc_linkup(struct lpfc_hba *phba)
 	struct lpfc_vport **vports;
 	int i;
 
+	lpfc_cleanup_wt_rrqs(phba);
 	phba->link_state = LPFC_LINK_UP;
 
 	/* Unblock fabric iocbs if they are blocked */
@@ -3161,6 +3164,7 @@ lpfc_mbx_cmpl_unreg_vpi(struct lpfc_hba *phba, LPFC_MBOXQ_t *pmb)
 	spin_unlock_irq(shost->host_lock);
 	vport->unreg_vpi_cmpl = VPORT_OK;
 	mempool_free(pmb, phba->mbox_mem_pool);
+	lpfc_cleanup_vports_rrqs(vport);
 	/*
 	 * This shost reference might have been taken at the beginning of
 	 * lpfc_vport_delete()

drivers/scsi/lpfc/lpfc_hw.h

@@ -880,6 +880,24 @@ struct  RLS_RSP {		/* Structure is in Big Endian format */
 	uint32_t crcCnt;
 };
 
+struct RRQ {			/* Structure is in Big Endian format */
+	uint32_t rrq;
+#define rrq_rsvd_SHIFT		24
+#define rrq_rsvd_MASK		0x000000ff
+#define rrq_rsvd_WORD		rrq
+#define rrq_did_SHIFT		0
+#define rrq_did_MASK		0x00ffffff
+#define rrq_did_WORD		rrq
+	uint32_t rrq_exchg;
+#define rrq_oxid_SHIFT		16
+#define rrq_oxid_MASK		0xffff
+#define rrq_oxid_WORD		rrq_exchg
+#define rrq_rxid_SHIFT		0
+#define rrq_rxid_MASK		0xffff
+#define rrq_rxid_WORD		rrq_exchg
+};
+
+
 struct RTV_RSP {		/* Structure is in Big Endian format */
 	uint32_t ratov;
 	uint32_t edtov;

drivers/scsi/lpfc/lpfc_init.c

@@ -929,6 +929,35 @@ lpfc_hb_timeout(unsigned long ptr)
 	return;
 }
 
+/**
+ * lpfc_rrq_timeout - The RRQ-timer timeout handler
+ * @ptr: unsigned long holds the pointer to lpfc hba data structure.
+ *
+ * This is the RRQ-timer timeout handler registered to the lpfc driver. When
+ * this timer fires, a RRQ timeout event shall be posted to the lpfc driver
+ * work-port-events bitmap and the worker thread is notified. This timeout
+ * event will be used by the worker thread to invoke the actual timeout
+ * handler routine, lpfc_rrq_handler. Any periodical operations will
+ * be performed in the timeout handler and the RRQ timeout event bit shall
+ * be cleared by the worker thread after it has taken the event bitmap out.
+ **/
+static void
+lpfc_rrq_timeout(unsigned long ptr)
+{
+	struct lpfc_hba *phba;
+	uint32_t tmo_posted;
+	unsigned long iflag;
+
+	phba = (struct lpfc_hba *)ptr;
+	spin_lock_irqsave(&phba->pport->work_port_lock, iflag);
+	tmo_posted = phba->hba_flag & HBA_RRQ_ACTIVE;
+	if (!tmo_posted)
+		phba->hba_flag |= HBA_RRQ_ACTIVE;
+	spin_unlock_irqrestore(&phba->pport->work_port_lock, iflag);
+	if (!tmo_posted)
+		lpfc_worker_wake_up(phba);
+}
+
 /**
  * lpfc_hb_mbox_cmpl - The lpfc heart-beat mailbox command callback function
  * @phba: pointer to lpfc hba data structure.
@@ -3990,6 +4019,9 @@ lpfc_sli4_driver_resource_setup(struct lpfc_hba *phba)
 	init_timer(&phba->hb_tmofunc);
 	phba->hb_tmofunc.function = lpfc_hb_timeout;
 	phba->hb_tmofunc.data = (unsigned long)phba;
+	init_timer(&phba->rrq_tmr);
+	phba->rrq_tmr.function = lpfc_rrq_timeout;
+	phba->rrq_tmr.data = (unsigned long)phba;
 
 	psli = &phba->sli;
 	/* MBOX heartbeat timer */
@@ -8192,6 +8224,8 @@ lpfc_pci_probe_one_s4(struct pci_dev *pdev, const struct pci_device_id *pid)
 		goto out_unset_driver_resource_s4;
 	}
 
+	INIT_LIST_HEAD(&phba->active_rrq_list);
+
 	/* Set up common device driver resources */
 	error = lpfc_setup_driver_resource_phase2(phba);
 	if (error) {

drivers/scsi/lpfc/lpfc_mem.c

@@ -113,11 +113,16 @@ lpfc_mem_alloc(struct lpfc_hba *phba, int align)
 		goto fail_free_mbox_pool;
 
 	if (phba->sli_rev == LPFC_SLI_REV4) {
+		phba->rrq_pool =
+			mempool_create_kmalloc_pool(LPFC_MEM_POOL_SIZE,
+						sizeof(struct lpfc_node_rrq));
+		if (!phba->rrq_pool)
+			goto fail_free_nlp_mem_pool;
 		phba->lpfc_hrb_pool = pci_pool_create("lpfc_hrb_pool",
 					      phba->pcidev,
 					      LPFC_HDR_BUF_SIZE, align, 0);
 		if (!phba->lpfc_hrb_pool)
-			goto fail_free_nlp_mem_pool;
+			goto fail_free_rrq_mem_pool;
 
 		phba->lpfc_drb_pool = pci_pool_create("lpfc_drb_pool",
 					      phba->pcidev,
@@ -147,6 +152,9 @@ lpfc_mem_alloc(struct lpfc_hba *phba, int align)
  fail_free_hrb_pool:
 	pci_pool_destroy(phba->lpfc_hrb_pool);
 	phba->lpfc_hrb_pool = NULL;
+ fail_free_rrq_mem_pool:
+	mempool_destroy(phba->rrq_pool);
+	phba->rrq_pool = NULL;
  fail_free_nlp_mem_pool:
 	mempool_destroy(phba->nlp_mem_pool);
 	phba->nlp_mem_pool = NULL;

drivers/scsi/lpfc/lpfc_scsi.c

@@ -621,10 +621,13 @@ lpfc_sli4_fcp_xri_aborted(struct lpfc_hba *phba,
 			  struct sli4_wcqe_xri_aborted *axri)
 {
 	uint16_t xri = bf_get(lpfc_wcqe_xa_xri, axri);
+	uint16_t rxid = bf_get(lpfc_wcqe_xa_remote_xid, axri);
 	struct lpfc_scsi_buf *psb, *next_psb;
 	unsigned long iflag = 0;
 	struct lpfc_iocbq *iocbq;
 	int i;
+	struct lpfc_nodelist *ndlp;
+	int rrq_empty = 0;
 	struct lpfc_sli_ring *pring = &phba->sli.ring[LPFC_ELS_RING];
 
 	spin_lock_irqsave(&phba->hbalock, iflag);
@@ -637,8 +640,14 @@ lpfc_sli4_fcp_xri_aborted(struct lpfc_hba *phba,
 			psb->status = IOSTAT_SUCCESS;
 			spin_unlock(
 				&phba->sli4_hba.abts_scsi_buf_list_lock);
+			ndlp = psb->rdata->pnode;
+			rrq_empty = list_empty(&phba->active_rrq_list);
 			spin_unlock_irqrestore(&phba->hbalock, iflag);
+			if (ndlp)
+				lpfc_set_rrq_active(phba, ndlp, xri, rxid, 1);
 			lpfc_release_scsi_buf_s4(phba, psb);
+			if (rrq_empty)
+				lpfc_worker_wake_up(phba);
 			return;
 		}
 	}
@@ -914,7 +923,7 @@ lpfc_new_scsi_buf(struct lpfc_vport *vport, int num_to_alloc)
 }
 
 /**
- * lpfc_get_scsi_buf - Get a scsi buffer from lpfc_scsi_buf_list of the HBA
+ * lpfc_get_scsi_buf_s3 - Get a scsi buffer from lpfc_scsi_buf_list of the HBA
  * @phba: The HBA for which this call is being executed.
  *
  * This routine removes a scsi buffer from head of @phba lpfc_scsi_buf_list list
@@ -925,7 +934,7 @@ lpfc_new_scsi_buf(struct lpfc_vport *vport, int num_to_alloc)
  *   Pointer to lpfc_scsi_buf - Success
  **/
 static struct lpfc_scsi_buf*
-lpfc_get_scsi_buf(struct lpfc_hba * phba)
+lpfc_get_scsi_buf_s3(struct lpfc_hba *phba, struct lpfc_nodelist *ndlp)
 {
 	struct  lpfc_scsi_buf * lpfc_cmd = NULL;
 	struct list_head *scsi_buf_list = &phba->lpfc_scsi_buf_list;
@@ -941,6 +950,67 @@ lpfc_get_scsi_buf(struct lpfc_hba * phba)
 	spin_unlock_irqrestore(&phba->scsi_buf_list_lock, iflag);
 	return  lpfc_cmd;
 }
+/**
+ * lpfc_get_scsi_buf_s4 - Get a scsi buffer from lpfc_scsi_buf_list of the HBA
+ * @phba: The HBA for which this call is being executed.
+ *
+ * This routine removes a scsi buffer from head of @phba lpfc_scsi_buf_list list
+ * and returns to caller.
+ *
+ * Return codes:
+ *   NULL - Error
+ *   Pointer to lpfc_scsi_buf - Success
+ **/
+static struct lpfc_scsi_buf*
+lpfc_get_scsi_buf_s4(struct lpfc_hba *phba, struct lpfc_nodelist *ndlp)
+{
+	struct  lpfc_scsi_buf *lpfc_cmd = NULL;
+	struct  lpfc_scsi_buf *start_lpfc_cmd = NULL;
+	struct list_head *scsi_buf_list = &phba->lpfc_scsi_buf_list;
+	unsigned long iflag = 0;
+	int found = 0;
+
+	spin_lock_irqsave(&phba->scsi_buf_list_lock, iflag);
+	list_remove_head(scsi_buf_list, lpfc_cmd, struct lpfc_scsi_buf, list);
+	spin_unlock_irqrestore(&phba->scsi_buf_list_lock, iflag);
+	while (!found && lpfc_cmd) {
+		if (lpfc_test_rrq_active(phba, ndlp,
+					 lpfc_cmd->cur_iocbq.sli4_xritag)) {
+			lpfc_release_scsi_buf_s4(phba, lpfc_cmd);
+			spin_lock_irqsave(&phba->scsi_buf_list_lock, iflag);
+			list_remove_head(scsi_buf_list, lpfc_cmd,
+					 struct lpfc_scsi_buf, list);
+			spin_unlock_irqrestore(&phba->scsi_buf_list_lock,
+						 iflag);
+			if (lpfc_cmd == start_lpfc_cmd) {
+				lpfc_cmd = NULL;
+				break;
+			} else
+				continue;
+		}
+		found = 1;
+		lpfc_cmd->seg_cnt = 0;
+		lpfc_cmd->nonsg_phys = 0;
+		lpfc_cmd->prot_seg_cnt = 0;
+	}
+	return  lpfc_cmd;
+}
+/**
+ * lpfc_get_scsi_buf - Get a scsi buffer from lpfc_scsi_buf_list of the HBA
+ * @phba: The HBA for which this call is being executed.
+ *
+ * This routine removes a scsi buffer from head of @phba lpfc_scsi_buf_list list
+ * and returns to caller.
+ *
+ * Return codes:
+ *   NULL - Error
+ *   Pointer to lpfc_scsi_buf - Success
+ **/
+static struct lpfc_scsi_buf*
+lpfc_get_scsi_buf(struct lpfc_hba *phba, struct lpfc_nodelist *ndlp)
+{
+	return  phba->lpfc_get_scsi_buf(phba, ndlp);
+}
 
 /**
  * lpfc_release_scsi_buf - Return a scsi buffer back to hba scsi buf list
@@ -2744,18 +2814,19 @@ lpfc_scsi_api_table_setup(struct lpfc_hba *phba, uint8_t dev_grp)
 
 	phba->lpfc_scsi_unprep_dma_buf = lpfc_scsi_unprep_dma_buf;
 	phba->lpfc_scsi_prep_cmnd = lpfc_scsi_prep_cmnd;
-	phba->lpfc_get_scsi_buf = lpfc_get_scsi_buf;
 
 	switch (dev_grp) {
 	case LPFC_PCI_DEV_LP:
 		phba->lpfc_new_scsi_buf = lpfc_new_scsi_buf_s3;
 		phba->lpfc_scsi_prep_dma_buf = lpfc_scsi_prep_dma_buf_s3;
 		phba->lpfc_release_scsi_buf = lpfc_release_scsi_buf_s3;
+		phba->lpfc_get_scsi_buf = lpfc_get_scsi_buf_s3;
 		break;
 	case LPFC_PCI_DEV_OC:
 		phba->lpfc_new_scsi_buf = lpfc_new_scsi_buf_s4;
 		phba->lpfc_scsi_prep_dma_buf = lpfc_scsi_prep_dma_buf_s4;
 		phba->lpfc_release_scsi_buf = lpfc_release_scsi_buf_s4;
+		phba->lpfc_get_scsi_buf = lpfc_get_scsi_buf_s4;
 		break;
 	default:
 		lpfc_printf_log(phba, KERN_ERR, LOG_INIT,
@@ -2764,7 +2835,6 @@ lpfc_scsi_api_table_setup(struct lpfc_hba *phba, uint8_t dev_grp)
 		return -ENODEV;
 		break;
 	}
-	phba->lpfc_get_scsi_buf = lpfc_get_scsi_buf;
 	phba->lpfc_rampdown_queue_depth = lpfc_rampdown_queue_depth;
 	phba->lpfc_scsi_cmd_iocb_cmpl = lpfc_scsi_cmd_iocb_cmpl;
 	return 0;
@@ -2940,7 +3010,7 @@ lpfc_queuecommand_lck(struct scsi_cmnd *cmnd, void (*done) (struct scsi_cmnd *))
 	if (atomic_read(&ndlp->cmd_pending) >= ndlp->cmd_qdepth)
 		goto out_host_busy;
 
-	lpfc_cmd = lpfc_get_scsi_buf(phba);
+	lpfc_cmd = lpfc_get_scsi_buf(phba, ndlp);
 	if (lpfc_cmd == NULL) {
 		lpfc_rampdown_queue_depth(phba);
 
@@ -3239,7 +3309,7 @@ lpfc_send_taskmgmt(struct lpfc_vport *vport, struct lpfc_rport_data *rdata,
 	if (!pnode || !NLP_CHK_NODE_ACT(pnode))
 		return FAILED;
 
-	lpfc_cmd = lpfc_get_scsi_buf(phba);
+	lpfc_cmd = lpfc_get_scsi_buf(phba, rdata->pnode);
 	if (lpfc_cmd == NULL)
 		return FAILED;
 	lpfc_cmd->timeout = 60;

drivers/scsi/lpfc/lpfc_sli.h

@@ -82,6 +82,7 @@ struct lpfc_iocbq {
 		struct lpfc_iocbq    *rsp_iocb;
 		struct lpfcMboxq     *mbox;
 		struct lpfc_nodelist *ndlp;
+		struct lpfc_node_rrq *rrq;
 	} context_un;
 
 	void (*fabric_iocb_cmpl) (struct lpfc_hba *, struct lpfc_iocbq *,

drivers/scsi/lpfc/lpfc_sli4.h

@@ -466,6 +466,7 @@ struct lpfc_sglq {
 	struct list_head clist;
 	enum lpfc_sge_type buff_type; /* is this a scsi sgl */
 	enum lpfc_sgl_state state;
+	struct lpfc_nodelist *ndlp; /* ndlp associated with IO */
 	uint16_t iotag;         /* pre-assigned IO tag */
 	uint16_t sli4_xritag;   /* pre-assigned XRI, (OXID) tag. */
 	struct sli4_sge *sgl;	/* pre-assigned SGL */