[SCSI] qla2xxx: Add critical temperature handling for ISPFX00.

Signed-off-by: Armen Baloyan <armen.baloyan@qlogic.com>
Acked-by: Srinivasa Rao <srinivasa.rao@qlogic.com>
Signed-off-by: Saurav Kashyap <saurav.kashyap@qlogic.com>
Signed-off-by: James Bottomley <JBottomley@Parallels.com>

drivers/scsi/qla2xxx/qla_dbg.c

@@ -11,7 +11,7 @@
  * ----------------------------------------------------------------------
  * |             Level            |   Last Value Used  |     Holes	|
  * ----------------------------------------------------------------------
- * | Module Init and Probe        |       0x0151       | 0x4b,0xba,0xfa |
+ * | Module Init and Probe        |       0x0152       | 0x4b,0xba,0xfa |
  * | Mailbox commands             |       0x1181       | 0x111a-0x111b  |
  * |                              |                    | 0x1155-0x1158  |
  * |                              |                    | 0x1018-0x1019  |
@@ -26,11 +26,11 @@
  * |                              |                    | 0x3036,0x3038  |
  * |                              |                    | 0x303a		|
  * | DPC Thread                   |       0x4022       | 0x4002,0x4013  |
- * | Async Events                 |       0x5081       | 0x502b-0x502f  |
+ * | Async Events                 |       0x5083       | 0x502b-0x502f  |
  * |                              |                    | 0x5047,0x5052  |
  * |                              |                    | 0x5040,0x5075  |
  * |                              |                    | 0x503d,0x5044  |
- * | Timer Routines               |       0x6011       |                |
+ * | Timer Routines               |       0x6012       |                |
  * | User Space Interactions      |       0x70dd       | 0x7018,0x702e, |
  * |                              |                    | 0x7020,0x7024, |
  * |                              |                    | 0x7039,0x7045, |

drivers/scsi/qla2xxx/qla_def.h

@@ -3369,6 +3369,7 @@ typedef struct scsi_qla_host {
 #define PORT_UPDATE_NEEDED	24
 #define FX00_RESET_RECOVERY	25
 #define FX00_TARGET_SCAN	26
+#define FX00_CRITEMP_RECOVERY	27
 
 	uint32_t	device_flags;
 #define SWITCH_FOUND		BIT_0

drivers/scsi/qla2xxx/qla_mr.c

@@ -1372,21 +1372,22 @@ qlafx00_configure_devices(scsi_qla_host_t *vha)
 }
 
 static void
-qlafx00_abort_isp_cleanup(scsi_qla_host_t *vha)
+qlafx00_abort_isp_cleanup(scsi_qla_host_t *vha, bool critemp)
 {
 	struct qla_hw_data *ha = vha->hw;
 	fc_port_t *fcport;
 
 	vha->flags.online = 0;
-	ha->flags.chip_reset_done = 0;
 	ha->mr.fw_hbt_en = 0;
+
+	if (!critemp) {
+		ha->flags.chip_reset_done = 0;
 		clear_bit(ISP_ABORT_NEEDED, &vha->dpc_flags);
 		vha->qla_stats.total_isp_aborts++;
-
 		ql_log(ql_log_info, vha, 0x013f,
 		    "Performing ISP error recovery - ha = %p.\n", ha);
-
 		ha->isp_ops->reset_chip(vha);
+	}
 
 	if (atomic_read(&vha->loop_state) != LOOP_DOWN) {
 		atomic_set(&vha->loop_state, LOOP_DOWN);
@@ -1406,11 +1407,18 @@ qlafx00_abort_isp_cleanup(scsi_qla_host_t *vha)
 	}
 
 	if (!ha->flags.eeh_busy) {
+		if (critemp) {
+			qla2x00_abort_all_cmds(vha, DID_NO_CONNECT << 16);
+		} else {
 			/* Requeue all commands in outstanding command list. */
 			qla2x00_abort_all_cmds(vha, DID_RESET << 16);
 		}
+	}
 
 	qla2x00_free_irqs(vha);
+	if (critemp)
+		set_bit(FX00_CRITEMP_RECOVERY, &vha->dpc_flags);
+	else
 		set_bit(FX00_RESET_RECOVERY, &vha->dpc_flags);
 
 	/* Clear the Interrupts */
@@ -1498,6 +1506,7 @@ qlafx00_timer_routine(scsi_qla_host_t *vha)
 	uint32_t fw_heart_beat;
 	uint32_t aenmbx0;
 	struct device_reg_fx00 __iomem *reg = &ha->iobase->ispfx00;
+	uint32_t tempc;
 
 	/* Check firmware health */
 	if (ha->mr.fw_hbt_cnt)
@@ -1569,6 +1578,29 @@ qlafx00_timer_routine(scsi_qla_host_t *vha)
 		ha->mr.old_aenmbx0_state = aenmbx0;
 		ha->mr.fw_reset_timer_tick--;
 	}
+	if (test_bit(FX00_CRITEMP_RECOVERY, &vha->dpc_flags)) {
+		/*
+		 * Critical temperature recovery to be
+		 * performed in timer routine
+		 */
+		if (ha->mr.fw_critemp_timer_tick == 0) {
+			tempc = QLAFX00_GET_TEMPERATURE(ha);
+			ql_log(ql_dbg_timer, vha, 0x6012,
+			    "ISPFx00(%s): Critical temp timer, "
+			    "current SOC temperature: %d\n",
+			    __func__, tempc);
+			if (tempc < ha->mr.critical_temperature) {
+				set_bit(ISP_ABORT_NEEDED, &vha->dpc_flags);
+				clear_bit(FX00_CRITEMP_RECOVERY,
+				    &vha->dpc_flags);
+				qla2xxx_wake_dpc(vha);
+			}
+			ha->mr.fw_critemp_timer_tick =
+			    QLAFX00_CRITEMP_INTERVAL;
+		} else {
+			ha->mr.fw_critemp_timer_tick--;
+		}
+	}
 }
 
 /*
@@ -1596,7 +1628,7 @@ qlafx00_reset_initialize(scsi_qla_host_t *vha)
 
 	if (vha->flags.online) {
 		scsi_block_requests(vha->host);
-		qlafx00_abort_isp_cleanup(vha);
+		qlafx00_abort_isp_cleanup(vha, false);
 	}
 
 	ql_log(ql_log_info, vha, 0x0143,
@@ -1628,7 +1660,7 @@ qlafx00_abort_isp(scsi_qla_host_t *vha)
 		}
 
 		scsi_block_requests(vha->host);
-		qlafx00_abort_isp_cleanup(vha);
+		qlafx00_abort_isp_cleanup(vha, false);
 	} else {
 		scsi_block_requests(vha->host);
 		clear_bit(ISP_ABORT_NEEDED, &vha->dpc_flags);
@@ -1722,6 +1754,16 @@ qlafx00_process_aen(struct scsi_qla_host *vha, struct qla_work_evt *evt)
 		aen_code = FCH_EVT_LINKDOWN;
 		aen_data = 0;
 		break;
+	case QLAFX00_MBA_TEMP_OVER:
+	case QLAFX00_MBA_TEMP_CRIT:	/* Critical temperature event */
+		ql_log(ql_log_info, vha, 0x5082,
+		    "Process critical temperature event "
+		    "aenmb[0]: %x\n",
+		    evt->u.aenfx.evtcode);
+		scsi_block_requests(vha->host);
+		qlafx00_abort_isp_cleanup(vha, true);
+		scsi_unblock_requests(vha->host);
+		break;
 	}
 
 	fc_host_post_event(vha->host, fc_get_event_number(),
@@ -1913,6 +1955,7 @@ qlafx00_fx_disc(scsi_qla_host_t *vha, fc_port_t *fcport, uint16_t fx_type)
 		    sizeof(vha->hw->mr.uboot_version));
 		memcpy(&vha->hw->mr.fru_serial_num, pinfo->fru_serial_num,
 		    sizeof(vha->hw->mr.fru_serial_num));
+		vha->hw->mr.critical_temperature = pinfo->nominal_temp_value;
 	} else if (fx_type == FXDISC_GET_PORT_INFO) {
 		struct port_info_data *pinfo =
 		    (struct port_info_data *) fdisc->u.fxiocb.rsp_addr;
@@ -2055,6 +2098,7 @@ qlafx00_initialize_adapter(scsi_qla_host_t *vha)
 {
 	int	rval;
 	struct qla_hw_data *ha = vha->hw;
+	uint32_t tempc;
 
 	/* Clear adapter flags. */
 	vha->flags.online = 0;
@@ -2105,6 +2149,11 @@ qlafx00_initialize_adapter(scsi_qla_host_t *vha)
 	rval = qla2x00_init_rings(vha);
 	ha->flags.chip_reset_done = 1;
 
+	tempc = QLAFX00_GET_TEMPERATURE(ha);
+	ql_dbg(ql_dbg_init, vha, 0x0152,
+	    "ISPFx00(%s): Critical temp timer, current SOC temperature: 0x%x\n",
+	    __func__, tempc);
+
 	return rval;
 }
 
@@ -2854,6 +2903,17 @@ qlafx00_async_event(scsi_qla_host_t *vha)
 		    ha->aenmb[0], ha->aenmb[1], ha->aenmb[2], ha->aenmb[3]);
 		data_size = 4;
 		break;
+
+	case QLAFX00_MBA_TEMP_OVER:	/* Over temperature event */
+	case QLAFX00_MBA_TEMP_CRIT:	/* Critical temperature event */
+		ql_log(ql_log_info, vha, 0x5083,
+		    "Asynchronous critical temperature event received "
+		    "aenmb[0]: %x\n",
+		ha->aenmb[0]);
+		qlafx00_post_aenfx_work(vha, ha->aenmb[0],
+		    (uint32_t *)ha->aenmb, 1);
+		break;
+
 	default:
 		ha->aenmb[1] = RD_REG_WORD(&reg->aenmailbox1);
 		ha->aenmb[2] = RD_REG_WORD(&reg->aenmailbox2);

drivers/scsi/qla2xxx/qla_mr.h

@@ -329,11 +329,13 @@ struct config_info_data {
 	uint64_t	adapter_id;
 
 	uint32_t	cluster_key_len;
-	uint8_t		cluster_key[10];
+	uint8_t		cluster_key[16];
 
 	uint64_t	cluster_master_id;
 	uint64_t	cluster_slave_id;
 	uint8_t		cluster_flags;
+	uint32_t	enabled_capabilities;
+	uint32_t	nominal_temp_value;
 } __packed;
 
 #define FXDISC_GET_CONFIG_INFO		0x01
@@ -346,6 +348,7 @@ struct config_info_data {
 #define QLAFX00_ICR_ENB_MASK            0x80000000
 #define QLAFX00_ICR_DIS_MASK            0x7fffffff
 #define QLAFX00_HST_RST_REG		0x18264
+#define QLAFX00_SOC_TEMP_REG		0x184C4
 #define QLAFX00_HST_TO_HBA_REG		0x20A04
 #define QLAFX00_HBA_TO_HOST_REG		0x21B70
 #define QLAFX00_HST_INT_STS_BITS	0x7
@@ -361,6 +364,9 @@ struct config_info_data {
 #define QLAFX00_INTR_ALL_CMPLT		0x7
 
 #define QLAFX00_MBA_SYSTEM_ERR		0x8002
+#define QLAFX00_MBA_TEMP_OVER		0x8005
+#define QLAFX00_MBA_TEMP_NORM		0x8006
+#define	QLAFX00_MBA_TEMP_CRIT		0x8007
 #define QLAFX00_MBA_LINK_UP		0x8011
 #define QLAFX00_MBA_LINK_DOWN		0x8012
 #define QLAFX00_MBA_PORT_UPDATE		0x8014
@@ -501,12 +507,31 @@ struct mr_data_fx00 {
 	uint32_t old_fw_hbt_cnt;
 	uint16_t fw_reset_timer_tick;
 	uint8_t fw_reset_timer_exp;
+	uint16_t fw_critemp_timer_tick;
 	uint32_t old_aenmbx0_state;
+	uint32_t critical_temperature;
 };
 
+/*
+ * SoC Junction Temperature is stored in
+ * bits 9:1 of SoC Junction Temperature Register
+ * in a firmware specific format format.
+ * To get the temperature in Celsius degrees
+ * the value from this bitfiled should be converted
+ * using this formula:
+ * Temperature (degrees C) = ((3,153,000 - (10,000 * X)) / 13,825)
+ * where X is the bit field value
+ * this macro reads the register, extracts the bitfield value,
+ * performs the calcualtions and returns temperature in Celsius
+ */
+#define QLAFX00_GET_TEMPERATURE(ha) ((3153000 - (10000 * \
+	((QLAFX00_RD_REG(ha, QLAFX00_SOC_TEMP_REG) & 0x3FE) >> 1))) / 13825)
+
+
 #define QLAFX00_LOOP_DOWN_TIME		615     /* 600 */
 #define QLAFX00_HEARTBEAT_INTERVAL	6	/* number of seconds */
 #define QLAFX00_HEARTBEAT_MISS_CNT	3	/* number of miss */
 #define QLAFX00_RESET_INTERVAL		120	/* number of seconds */
 #define QLAFX00_MAX_RESET_INTERVAL	600	/* number of seconds */
+#define QLAFX00_CRITEMP_INTERVAL	60	/* number of seconds */
 #endif

drivers/scsi/qla2xxx/qla_os.c

@@ -2575,6 +2575,7 @@ qla2x00_probe_one(struct pci_dev *pdev, const struct pci_device_id *id)
 		ha->port_down_retry_count = 30; /* default value */
 		ha->mr.fw_hbt_cnt = QLAFX00_HEARTBEAT_INTERVAL;
 		ha->mr.fw_reset_timer_tick = QLAFX00_RESET_INTERVAL;
+		ha->mr.fw_critemp_timer_tick = QLAFX00_CRITEMP_INTERVAL;
 		ha->mr.fw_hbt_en = 1;
 	}