hal: fixes in the calibration logic

- load the cal file at the right moment (prior to startup of the FSMs) - increase phase tollerance for the first time calibration (some bins are missing on some transceivers, therefore aiming for just one is risky) - added timeout & retry on rt_ipc communication (likely masks a bug in mini-rpc)

hal: fixes in the calibration logic
- load the cal file at the right moment (prior to startup of the FSMs) - increase phase tollerance for the first time calibration (some bins are missing on some transceivers, therefore aiming for just one is risky) - added timeout & retry on rt_ipc communication (likely masks a bug in mini-rpc)
166367ea · Tomasz Wlostowski · 7f4a59f7 · 166367ea
Commit 166367ea authored Aug 21, 2019 by Tomasz Wlostowski
Hide whitespace changes
Inline Side-by-side

Showing with 31 additions and 16 deletions

hal_ports.c userspace/wrsw_hal/hal_ports.c +31 -16

No files found.
--- a/userspace/wrsw_hal/hal_ports.c
+++ b/userspace/wrsw_hal/hal_ports.c
@@ -43,6 +43,13 @@
 #define UPDATE_SFP_DOM_PERIOD 1000 /* ms */
 #define LINK_LED_BLINK_PERIOD 200 /* ms */

+#define TX_CAL_PHASE_MEAS_TIMEOUT 1000 /* ms */
+
+#define TX_CAL_TOLLERANCE 300 /* ps */
+#define TX_CAL_FIRST_CAL_TOLLERANCE 750 /* ps */
+#define TX_CAL_FIRST_CAL_EXPECTED_PHASE 750 /* ps */
+
+
 typedef struct {
 	struct pp_instance * ppi;              /* pointer to the ppi instance */
 	struct pp_servo      servo_snapshot;  /* image of a the ppsi servo */
@@ -64,6 +71,7 @@ struct hal_port_tx_setup_state
 	int tollerance;
 	int update_cnt;
 	int expected_phase_valid;
+	timeout_t timeout;
 };

 /* Low Phase Drift Calibration for rx */
@@ -350,7 +358,6 @@ int hal_port_init_shmem(char *logfilename)
 		return -1;
 	}

-	load_tx_calibration_file();

 	hal_shmem->ports = ports;

@@ -358,6 +365,8 @@ int hal_port_init_shmem(char *logfilename)
 		if (hal_port_init(index) < 0)
 			break;

+	load_tx_calibration_file();
+
 	hal_port_nports = index;

 	pr_info("Number of physical ports supported in HW: %d\n",
@@ -1104,7 +1113,7 @@ static void tx_fsm_init(struct hal_port_state *p)
 	fsm->state = TX_SETUP_STATE_START;
 	fsm->expected_phase = 0;
 	fsm->expected_phase_valid = 0;
-	fsm->tollerance = 300;
+	fsm->tollerance = TX_CAL_TOLLERANCE;
 	fsm->update_cnt = 0;
 	fsm->cal_saved_phase_valid = 0;
 	fsm->cal_saved_phase = 0;
@@ -1120,7 +1129,7 @@ static int within_range(int x, int minval, int maxval, int wrap)
 {
 	int rv;

-	printf("min %d max %d x %d ", minval, maxval, x);
+	//printf("min %d max %d x %d ", minval, maxval, x);

 	while (maxval >= wrap)
 		maxval -= wrap; 
@@ -1172,7 +1181,7 @@ static void update_tx_calibration_file(void)

 	if (file_exists(calibration_file_name))
 	{
-		//pr_info("TX phase calibration data already exists, no need to update\n");
+		pr_info("TX phase calibration data already exists, no need to update\n");
 		return;
 	}

@@ -1293,12 +1302,21 @@ static int tx_fsm_update(struct hal_port_state *p)

 			rts_enable_ptracker(p->hw_index, 1);
 			tx_fsm_pll_state.channels[p->hw_index].flags = 0;
+			libwr_tmo_init(&fsm->timeout, TX_CAL_PHASE_MEAS_TIMEOUT, 1);
+	
 		}
 		break;
 	}

 	case TX_SETUP_STATE_MEASURE_PHASE:
 	{
+		if (libwr_tmo_expired(&fsm->timeout) )
+		{
+			pr_info("Port %d: tx phase measurement timeout expired, retrying\n", p->hw_index+1);
+			fsm->state = TX_SETUP_STATE_RESET_PCS;
+			return 0;
+		}
+
 		if (!(tx_fsm_pll_state.channels[p->hw_index].flags & CHAN_PMEAS_READY))
 			return 0; // keep waiting

@@ -1309,18 +1327,18 @@ static int tx_fsm_update(struct hal_port_state *p)
 		{
 			if (fsm->cal_saved_phase_valid)
 			{
+				// got calibration file already? Aim EXACTLY for the phase
+				// bin we used in the first calibration
 				pr_info("Using phase from file :%d\n", fsm->cal_saved_phase);
 				fsm->expected_phase = fsm->cal_saved_phase;
 			}
 			else
 			{
-				int phi = phase;
+				// First time calibrating? Give ourselves some freedom, let's say
+				// the first 1.5 ns of the 16 ns ref clock cycle, so that we have enough setup time

-				do // find the phase bin right after the rising parallel clock edge
-				{
-					fsm->expected_phase = phi;
-					phi -= 800;
-				} while (phi > 0);
+				fsm->tollerance = TX_CAL_FIRST_CAL_TOLLERANCE;
+				fsm->expected_phase = TX_CAL_FIRST_CAL_EXPECTED_PHASE;
 			}
 			fsm->expected_phase_valid = 1;
 		}
@@ -1501,14 +1519,11 @@ static int rx_fsm_update(struct hal_port_state *p)
 				      MDIO_LPC_CTRL);
 			pcs_writel(p, BMCR_ANENABLE | BMCR_ANRESTART, MII_BMCR);
 			//TODO-ML: change to port name
-			pr_info("Port %d: RX calibration complete at phase %d "
+			pr_info("wri%d: RX calibration complete at phase %d "
 				"ps (after %d attempts).\n", p->hw_index + 1,
 				phase, fsm->attempts);
-			printf("MDIO_MCR %04x\n", pcs_readl(p, 0));
 			rts_enable_ptracker(p->hw_index, 0);
-
-			sleep(1);
-			printf("MDIO_MSR %04x\n", pcs_readl(p, 1));
+			sleep(1); // fixme: really needed?
 		}

 		break;
@@ -1520,7 +1535,7 @@ static int rx_fsm_update(struct hal_port_state *p)

 		if (!(dbg0 & MDIO_LPC_STAT_LINK_UP))
 		{
-			pr_info("rxcal: port %d went down\n", p->hw_index + 1);
+			pr_info("rxcal: early link flag lost on port wri%d\n", p->hw_index + 1);
 			fsm->state = RX_SETUP_STATE_INIT;

 			return 0;