1 files changed, 104 insertions, 69 deletions

diff --git a/drivers/scsi/qla2xxx/qla_os.c b/drivers/scsi/qla2xxx/qla_os.c
index cb4155f8fd44..12959e3874cb 100644
--- a/drivers/scsi/qla2xxx/qla_os.c
+++ b/drivers/scsi/qla2xxx/qla_os.c
@@ -971,6 +971,13 @@ qla2xxx_mqueuecommand(struct Scsi_Host *host, struct scsi_cmnd *cmd,
 		goto qc24_fail_command;
 	}
 
+	if (!qpair->online) {
+		ql_dbg(ql_dbg_io, vha, 0x3077,
+		       "qpair not online. eeh_busy=%d.\n", ha->flags.eeh_busy);
+		cmd->result = DID_NO_CONNECT << 16;
+		goto qc24_fail_command;
+	}
+
 	if (!fcport || fcport->deleted) {
 		cmd->result = DID_IMM_RETRY << 16;
 		goto qc24_fail_command;
@@ -1200,35 +1207,6 @@ qla2x00_wait_for_chip_reset(scsi_qla_host_t *vha)
 	return return_status;
 }
 
-#define ISP_REG_DISCONNECT 0xffffffffU
-/**************************************************************************
-* qla2x00_isp_reg_stat
-*
-* Description:
-*	Read the host status register of ISP before aborting the command.
-*
-* Input:
-*	ha = pointer to host adapter structure.
-*
-*
-* Returns:
-*	Either true or false.
-*
-* Note:	Return true if there is register disconnect.
-**************************************************************************/
-static inline
-uint32_t qla2x00_isp_reg_stat(struct qla_hw_data *ha)
-{
-	struct device_reg_24xx __iomem *reg = &ha->iobase->isp24;
-	struct device_reg_82xx __iomem *reg82 = &ha->iobase->isp82;
-
-	if (IS_P3P_TYPE(ha))
-		return ((rd_reg_dword(&reg82->host_int)) == ISP_REG_DISCONNECT);
-	else
-		return ((rd_reg_dword(&reg->host_status)) ==
-			ISP_REG_DISCONNECT);
-}
-
 /**************************************************************************
 * qla2xxx_eh_abort
 *
@@ -1262,6 +1240,7 @@ qla2xxx_eh_abort(struct scsi_cmnd *cmd)
 	if (qla2x00_isp_reg_stat(ha)) {
 		ql_log(ql_log_info, vha, 0x8042,
 		    "PCI/Register disconnect, exiting.\n");
+		qla_pci_set_eeh_busy(vha);
 		return FAILED;
 	}
 
@@ -1455,6 +1434,7 @@ qla2xxx_eh_device_reset(struct scsi_cmnd *cmd)
 	if (qla2x00_isp_reg_stat(ha)) {
 		ql_log(ql_log_info, vha, 0x803e,
 		    "PCI/Register disconnect, exiting.\n");
+		qla_pci_set_eeh_busy(vha);
 		return FAILED;
 	}
 
@@ -1471,6 +1451,7 @@ qla2xxx_eh_target_reset(struct scsi_cmnd *cmd)
 	if (qla2x00_isp_reg_stat(ha)) {
 		ql_log(ql_log_info, vha, 0x803f,
 		    "PCI/Register disconnect, exiting.\n");
+		qla_pci_set_eeh_busy(vha);
 		return FAILED;
 	}
 
@@ -1506,6 +1487,7 @@ qla2xxx_eh_bus_reset(struct scsi_cmnd *cmd)
 	if (qla2x00_isp_reg_stat(ha)) {
 		ql_log(ql_log_info, vha, 0x8040,
 		    "PCI/Register disconnect, exiting.\n");
+		qla_pci_set_eeh_busy(vha);
 		return FAILED;
 	}
 
@@ -1583,7 +1565,7 @@ qla2xxx_eh_host_reset(struct scsi_cmnd *cmd)
 	if (qla2x00_isp_reg_stat(ha)) {
 		ql_log(ql_log_info, vha, 0x8041,
 		    "PCI/Register disconnect, exiting.\n");
-		schedule_work(&ha->board_disable);
+		qla_pci_set_eeh_busy(vha);
 		return SUCCESS;
 	}
 
@@ -6668,6 +6650,9 @@ qla2x00_do_dpc(void *data)
 
 		schedule();
 
+		if (test_and_clear_bit(DO_EEH_RECOVERY, &base_vha->dpc_flags))
+			qla_pci_set_eeh_busy(base_vha);
+
 		if (!base_vha->flags.init_done || ha->flags.mbox_busy)
 			goto end_loop;
 
@@ -7384,6 +7369,8 @@ static void qla_pci_error_cleanup(scsi_qla_host_t *vha)
 	int i;
 	unsigned long flags;
 
+	ql_dbg(ql_dbg_aer, vha, 0x9000,
+	       "%s\n", __func__);
 	ha->chip_reset++;
 
 	ha->base_qpair->chip_reset = ha->chip_reset;
@@ -7393,28 +7380,16 @@ static void qla_pci_error_cleanup(scsi_qla_host_t *vha)
 			    ha->base_qpair->chip_reset;
 	}
 
-	/* purge MBox commands */
-	if (atomic_read(&ha->num_pend_mbx_stage3)) {
-		clear_bit(MBX_INTR_WAIT, &ha->mbx_cmd_flags);
-		complete(&ha->mbx_intr_comp);
-	}
-
-	i = 0;
-
-	while (atomic_read(&ha->num_pend_mbx_stage3) ||
-	    atomic_read(&ha->num_pend_mbx_stage2) ||
-	    atomic_read(&ha->num_pend_mbx_stage1)) {
-		msleep(20);
-		i++;
-		if (i > 50)
-			break;
-	}
-
-	ha->flags.purge_mbox = 0;
+	/*
+	 * purge mailbox might take a while. Slot Reset/chip reset
+	 * will take care of the purge
+	 */
 
 	mutex_lock(&ha->mq_lock);
+	ha->base_qpair->online = 0;
 	list_for_each_entry(qpair, &base_vha->qp_list, qp_list_elem)
 		qpair->online = 0;
+	wmb();
 	mutex_unlock(&ha->mq_lock);
 
 	qla2x00_mark_all_devices_lost(vha);
@@ -7451,14 +7426,17 @@ qla2xxx_pci_error_detected(struct pci_dev *pdev, pci_channel_state_t state)
 {
 	scsi_qla_host_t *vha = pci_get_drvdata(pdev);
 	struct qla_hw_data *ha = vha->hw;
+	pci_ers_result_t ret = PCI_ERS_RESULT_NEED_RESET;
 
-	ql_dbg(ql_dbg_aer, vha, 0x9000,
-	    "PCI error detected, state %x.\n", state);
+	ql_log(ql_log_warn, vha, 0x9000,
+	       "PCI error detected, state %x.\n", state);
+	ha->pci_error_state = QLA_PCI_ERR_DETECTED;
 
 	if (!atomic_read(&pdev->enable_cnt)) {
 		ql_log(ql_log_info, vha, 0xffff,
 			"PCI device is disabled,state %x\n", state);
-		return PCI_ERS_RESULT_NEED_RESET;
+		ret = PCI_ERS_RESULT_NEED_RESET;
+		goto out;
 	}
 
 	switch (state) {
@@ -7468,11 +7446,12 @@ qla2xxx_pci_error_detected(struct pci_dev *pdev, pci_channel_state_t state)
 			set_bit(QPAIR_ONLINE_CHECK_NEEDED, &vha->dpc_flags);
 			qla2xxx_wake_dpc(vha);
 		}
-		return PCI_ERS_RESULT_CAN_RECOVER;
+		ret = PCI_ERS_RESULT_CAN_RECOVER;
+		break;
 	case pci_channel_io_frozen:
-		ha->flags.eeh_busy = 1;
-		qla_pci_error_cleanup(vha);
-		return PCI_ERS_RESULT_NEED_RESET;
+		qla_pci_set_eeh_busy(vha);
+		ret = PCI_ERS_RESULT_NEED_RESET;
+		break;
 	case pci_channel_io_perm_failure:
 		ha->flags.pci_channel_io_perm_failure = 1;
 		qla2x00_abort_all_cmds(vha, DID_NO_CONNECT << 16);
@@ -7480,9 +7459,12 @@ qla2xxx_pci_error_detected(struct pci_dev *pdev, pci_channel_state_t state)
 			set_bit(QPAIR_ONLINE_CHECK_NEEDED, &vha->dpc_flags);
 			qla2xxx_wake_dpc(vha);
 		}
-		return PCI_ERS_RESULT_DISCONNECT;
+		ret = PCI_ERS_RESULT_DISCONNECT;
 	}
-	return PCI_ERS_RESULT_NEED_RESET;
+out:
+	ql_dbg(ql_dbg_aer, vha, 0x600d,
+	       "PCI error detected returning [%x].\n", ret);
+	return ret;
 }
 
 static pci_ers_result_t
@@ -7496,6 +7478,10 @@ qla2xxx_pci_mmio_enabled(struct pci_dev *pdev)
 	struct device_reg_2xxx __iomem *reg = &ha->iobase->isp;
 	struct device_reg_24xx __iomem *reg24 = &ha->iobase->isp24;
 
+	ql_log(ql_log_warn, base_vha, 0x9000,
+	       "mmio enabled\n");
+
+	ha->pci_error_state = QLA_PCI_MMIO_ENABLED;
 	if (IS_QLA82XX(ha))
 		return PCI_ERS_RESULT_RECOVERED;
 
@@ -7519,10 +7505,11 @@ qla2xxx_pci_mmio_enabled(struct pci_dev *pdev)
 		ql_log(ql_log_info, base_vha, 0x9003,
 		    "RISC paused -- mmio_enabled, Dumping firmware.\n");
 		qla2xxx_dump_fw(base_vha);
-
-		return PCI_ERS_RESULT_NEED_RESET;
-	} else
-		return PCI_ERS_RESULT_RECOVERED;
+	}
+	/* set PCI_ERS_RESULT_NEED_RESET to trigger call to qla2xxx_pci_slot_reset */
+	ql_dbg(ql_dbg_aer, base_vha, 0x600d,
+	       "mmio enabled returning.\n");
+	return PCI_ERS_RESULT_NEED_RESET;
 }
 
 static pci_ers_result_t
@@ -7534,9 +7521,10 @@ qla2xxx_pci_slot_reset(struct pci_dev *pdev)
 	int rc;
 	struct qla_qpair *qpair = NULL;
 
-	ql_dbg(ql_dbg_aer, base_vha, 0x9004,
-	    "Slot Reset.\n");
+	ql_log(ql_log_warn, base_vha, 0x9004,
+	       "Slot Reset.\n");
 
+	ha->pci_error_state = QLA_PCI_SLOT_RESET;
 	/* Workaround: qla2xxx driver which access hardware earlier
 	 * needs error state to be pci_channel_io_online.
 	 * Otherwise mailbox command timesout.
@@ -7570,16 +7558,24 @@ qla2xxx_pci_slot_reset(struct pci_dev *pdev)
 		qpair->online = 1;
 	mutex_unlock(&ha->mq_lock);
 
+	ha->flags.eeh_busy = 0;
 	base_vha->flags.online = 1;
 	set_bit(ABORT_ISP_ACTIVE, &base_vha->dpc_flags);
-	if (ha->isp_ops->abort_isp(base_vha) == QLA_SUCCESS)
-		ret =  PCI_ERS_RESULT_RECOVERED;
+	ha->isp_ops->abort_isp(base_vha);
 	clear_bit(ABORT_ISP_ACTIVE, &base_vha->dpc_flags);
 
+	if (qla2x00_isp_reg_stat(ha)) {
+		ha->flags.eeh_busy = 1;
+		qla_pci_error_cleanup(base_vha);
+		ql_log(ql_log_warn, base_vha, 0x9005,
+		       "Device unable to recover from PCI error.\n");
+	} else {
+		ret =  PCI_ERS_RESULT_RECOVERED;
+	}
 
 exit_slot_reset:
 	ql_dbg(ql_dbg_aer, base_vha, 0x900e,
-	    "slot_reset return %x.\n", ret);
+	    "Slot Reset returning %x.\n", ret);
 
 	return ret;
 }
@@ -7591,16 +7587,55 @@ qla2xxx_pci_resume(struct pci_dev *pdev)
 	struct qla_hw_data *ha = base_vha->hw;
 	int ret;
 
-	ql_dbg(ql_dbg_aer, base_vha, 0x900f,
-	    "pci_resume.\n");
+	ql_log(ql_log_warn, base_vha, 0x900f,
+	       "Pci Resume.\n");
 
-	ha->flags.eeh_busy = 0;
 
 	ret = qla2x00_wait_for_hba_online(base_vha);
 	if (ret != QLA_SUCCESS) {
 		ql_log(ql_log_fatal, base_vha, 0x9002,
 		    "The device failed to resume I/O from slot/link_reset.\n");
 	}
+	ha->pci_error_state = QLA_PCI_RESUME;
+	ql_dbg(ql_dbg_aer, base_vha, 0x600d,
+	       "Pci Resume returning.\n");
+}
+
+void qla_pci_set_eeh_busy(struct scsi_qla_host *vha)
+{
+	struct qla_hw_data *ha = vha->hw;
+	struct scsi_qla_host *base_vha = pci_get_drvdata(ha->pdev);
+	bool do_cleanup = false;
+	unsigned long flags;
+
+	if (ha->flags.eeh_busy)
+		return;
+
+	spin_lock_irqsave(&base_vha->work_lock, flags);
+	if (!ha->flags.eeh_busy) {
+		ha->flags.eeh_busy = 1;
+		do_cleanup = true;
+	}
+	spin_unlock_irqrestore(&base_vha->work_lock, flags);
+
+	if (do_cleanup)
+		qla_pci_error_cleanup(base_vha);
+}
+
+/*
+ * this routine will schedule a task to pause IO from interrupt context
+ * if caller sees a PCIE error event (register read = 0xf's)
+ */
+void qla_schedule_eeh_work(struct scsi_qla_host *vha)
+{
+	struct qla_hw_data *ha = vha->hw;
+	struct scsi_qla_host *base_vha = pci_get_drvdata(ha->pdev);
+
+	if (ha->flags.eeh_busy)
+		return;
+
+	set_bit(DO_EEH_RECOVERY, &base_vha->dpc_flags);
+	qla2xxx_wake_dpc(base_vha);
 }
 
 static void