MFC r276959:

cxgb: replace r273280 with a more comprehensive fix.

Poll for link state when the link is down, even for interrupt capable
PHYs.

Allow PHYs to report a dubious "partial" link.  If this state is seen 3
consecutive times (each check is ~1s apart) then reset the PHY.  This is
a workaround for a situation where repeatedly toggling the link from the
peer gets the AEL2005 PHY into a state where it never establishes a PCS
block lock even when everything is in order.

8 files changed, 59 insertions, 33 deletions

diff --git a/sys/dev/cxgb/common/cxgb_ael1002.c b/sys/dev/cxgb/common/cxgb_ael1002.c
index bcd2728..e1a3fbe 100644
--- a/sys/dev/cxgb/common/cxgb_ael1002.c
+++ b/sys/dev/cxgb/common/cxgb_ael1002.c
@@ -283,10 +283,10 @@ static int ael1002_intr_noop(struct cphy *phy)
 /*
  * Get link status for a 10GBASE-R device.
  */
-static int get_link_status_r(struct cphy *phy, int *link_ok, int *speed,
+static int get_link_status_r(struct cphy *phy, int *link_state, int *speed,
 			     int *duplex, int *fc)
 {
-	if (link_ok) {
+	if (link_state) {
 		unsigned int stat0, stat1, stat2;
 		int err = mdio_read(phy, MDIO_DEV_PMA_PMD, PMD_RSD, &stat0);
 
@@ -296,10 +296,16 @@ static int get_link_status_r(struct cphy *phy, int *link_ok, int *speed,
 			err = mdio_read(phy, MDIO_DEV_XGXS, XS_LN_STAT, &stat2);
 		if (err)
 			return err;
-		*link_ok = (stat0 & stat1 & (stat2 >> 12)) & 1;
 
-		if (*link_ok == 0)
-			return (0);
+		stat0 &= 1;
+		stat1 &= 1;
+		stat2 = (stat2 >> 12) & 1;
+		if (stat0 & stat1 & stat2)
+			*link_state = PHY_LINK_UP;
+		else if (stat0 == 1 && stat1 == 0 && stat2 == 1)
+			*link_state = PHY_LINK_PARTIAL;
+		else
+			*link_state = PHY_LINK_DOWN;
 	}
 	if (speed)
 		*speed = SPEED_10000;
@@ -1345,10 +1351,8 @@ static int ael2005_intr_handler(struct cphy *phy)
 		return ret;
 
 	ret |= cause;
-	if (!ret) {
-		(void) ael2005_reset(phy, 0);
+	if (!ret)
 		ret |= cphy_cause_link_change;
-	}
 	return ret;
 }
 
@@ -2156,10 +2160,10 @@ int t3_ael2020_phy_prep(pinfo_t *pinfo, int phy_addr,
 /*
  * Get link status for a 10GBASE-X device.
  */
-static int get_link_status_x(struct cphy *phy, int *link_ok, int *speed,
+static int get_link_status_x(struct cphy *phy, int *link_state, int *speed,
 			     int *duplex, int *fc)
 {
-	if (link_ok) {
+	if (link_state) {
 		unsigned int stat0, stat1, stat2;
 		int err = mdio_read(phy, MDIO_DEV_PMA_PMD, PMD_RSD, &stat0);
 
@@ -2169,7 +2173,10 @@ static int get_link_status_x(struct cphy *phy, int *link_ok, int *speed,
 			err = mdio_read(phy, MDIO_DEV_XGXS, XS_LN_STAT, &stat2);
 		if (err)
 			return err;
-		*link_ok = (stat0 & (stat1 >> 12) & (stat2 >> 12)) & 1;
+		if ((stat0 & (stat1 >> 12) & (stat2 >> 12)) & 1)
+			*link_state = PHY_LINK_UP;
+		else
+			*link_state = PHY_LINK_DOWN;
 	}
 	if (speed)
 		*speed = SPEED_10000;
@@ -2230,10 +2237,10 @@ static int xaui_direct_reset(struct cphy *phy, int wait)
 	return 0;
 }
 
-static int xaui_direct_get_link_status(struct cphy *phy, int *link_ok,
+static int xaui_direct_get_link_status(struct cphy *phy, int *link_state,
 				       int *speed, int *duplex, int *fc)
 {
-	if (link_ok) {
+	if (link_state) {
 		unsigned int status;
 		adapter_t *adapter = phy->adapter;
 
@@ -2245,7 +2252,7 @@ static int xaui_direct_get_link_status(struct cphy *phy, int *link_ok,
 				     XGM_REG(A_XGM_SERDES_STAT2, phy->addr)) |
 			 t3_read_reg(adapter,
 				     XGM_REG(A_XGM_SERDES_STAT3, phy->addr));
-		*link_ok = !(status & F_LOWSIG0);
+		*link_state = status & F_LOWSIG0 ? PHY_LINK_DOWN : PHY_LINK_UP;
 	}
 	if (speed)
 		*speed = SPEED_10000;
diff --git a/sys/dev/cxgb/common/cxgb_aq100x.c b/sys/dev/cxgb/common/cxgb_aq100x.c
index d36d07a..b253e67 100644
--- a/sys/dev/cxgb/common/cxgb_aq100x.c
+++ b/sys/dev/cxgb/common/cxgb_aq100x.c
@@ -350,7 +350,7 @@ aq100x_set_speed_duplex(struct cphy *phy, int speed, int duplex)
 }
 
 static int
-aq100x_get_link_status(struct cphy *phy, int *link_ok, int *speed, int *duplex,
+aq100x_get_link_status(struct cphy *phy, int *link_state, int *speed, int *duplex,
 		       int *fc)
 {
 	int err;
@@ -440,8 +440,8 @@ aq100x_get_link_status(struct cphy *phy, int *link_ok, int *speed, int *duplex,
 
 	link = 1;
 done:
-	if (link_ok)
-		*link_ok = link;
+	if (link_state)
+		*link_state = link ? PHY_LINK_UP : PHY_LINK_DOWN;
 	return (0);
 }
 
diff --git a/sys/dev/cxgb/common/cxgb_common.h b/sys/dev/cxgb/common/cxgb_common.h
index df4ddda..32f0a93 100644
--- a/sys/dev/cxgb/common/cxgb_common.h
+++ b/sys/dev/cxgb/common/cxgb_common.h
@@ -543,6 +543,12 @@ enum {
 	phy_modtype_unknown
 };
 
+enum {
+	PHY_LINK_DOWN = 0,
+	PHY_LINK_UP,
+	PHY_LINK_PARTIAL
+};
+
 /* PHY operations */
 struct cphy_ops {
 	int (*reset)(struct cphy *phy, int wait);
@@ -558,7 +564,7 @@ struct cphy_ops {
 	int (*advertise)(struct cphy *phy, unsigned int advertise_map);
 	int (*set_loopback)(struct cphy *phy, int mmd, int dir, int enable);
 	int (*set_speed_duplex)(struct cphy *phy, int speed, int duplex);
-	int (*get_link_status)(struct cphy *phy, int *link_ok, int *speed,
+	int (*get_link_status)(struct cphy *phy, int *link_state, int *speed,
 			       int *duplex, int *fc);
 	int (*power_down)(struct cphy *phy, int enable);
 };
@@ -567,6 +573,7 @@ struct cphy_ops {
 struct cphy {
 	u8 addr;                             /* PHY address */
 	u8 modtype;                          /* PHY module type */
+	u8 rst;
 	unsigned int priv;                   /* scratch pad */
 	unsigned int caps;                   /* PHY capabilities */
 	adapter_t *adapter;                  /* associated adapter */
diff --git a/sys/dev/cxgb/common/cxgb_mv88e1xxx.c b/sys/dev/cxgb/common/cxgb_mv88e1xxx.c
index 6281ac8..4093aa7 100644
--- a/sys/dev/cxgb/common/cxgb_mv88e1xxx.c
+++ b/sys/dev/cxgb/common/cxgb_mv88e1xxx.c
@@ -185,7 +185,7 @@ static int mv88e1xxx_set_loopback(struct cphy *cphy, int mmd, int dir, int on)
 			 	   on ? BMCR_LOOPBACK : 0);
 }
 
-static int mv88e1xxx_get_link_status(struct cphy *cphy, int *link_ok,
+static int mv88e1xxx_get_link_status(struct cphy *cphy, int *link_state,
 				     int *speed, int *duplex, int *fc)
 {
 	u32 status;
@@ -206,8 +206,9 @@ static int mv88e1xxx_get_link_status(struct cphy *cphy, int *link_ok,
 		else
 			sp = SPEED_1000;
 	}
-	if (link_ok)
-		*link_ok = (status & V_PSSR_LINK) != 0;
+	if (link_state)
+		*link_state = status & V_PSSR_LINK ? PHY_LINK_UP :
+		    PHY_LINK_DOWN;
 	if (speed)
 		*speed = sp;
 	if (duplex)
diff --git a/sys/dev/cxgb/common/cxgb_t3_hw.c b/sys/dev/cxgb/common/cxgb_t3_hw.c
index 32aab35..942571a 100644
--- a/sys/dev/cxgb/common/cxgb_t3_hw.c
+++ b/sys/dev/cxgb/common/cxgb_t3_hw.c
@@ -1520,7 +1520,7 @@ static void t3_clear_faults(adapter_t *adapter, int port_id)
  */
 void t3_link_changed(adapter_t *adapter, int port_id)
 {
-	int link_ok, speed, duplex, fc, link_fault;
+	int link_ok, speed, duplex, fc, link_fault, link_state;
 	struct port_info *pi = adap2pinfo(adapter, port_id);
 	struct cphy *phy = &pi->phy;
 	struct cmac *mac = &pi->mac;
@@ -1532,7 +1532,14 @@ void t3_link_changed(adapter_t *adapter, int port_id)
 	fc = lc->fc;
 	link_fault = 0;
 
-	phy->ops->get_link_status(phy, &link_ok, &speed, &duplex, &fc);
+	phy->ops->get_link_status(phy, &link_state, &speed, &duplex, &fc);
+	link_ok = (link_state == PHY_LINK_UP);
+	if (link_state != PHY_LINK_PARTIAL)
+		phy->rst = 0;
+	else if (++phy->rst == 3) {
+		phy->ops->reset(phy, 0);
+		phy->rst = 0;
+	}
 
 	if (link_ok == 0)
 		pi->link_fault = LF_NO;
diff --git a/sys/dev/cxgb/common/cxgb_tn1010.c b/sys/dev/cxgb/common/cxgb_tn1010.c
index 4575828..1e8bf0b 100644
--- a/sys/dev/cxgb/common/cxgb_tn1010.c
+++ b/sys/dev/cxgb/common/cxgb_tn1010.c
@@ -129,7 +129,7 @@ static int tn1010_advertise(struct cphy *phy, unsigned int advert)
 			  ADVERTISE_LOOP_TIMING);
 }
 
-static int tn1010_get_link_status(struct cphy *phy, int *link_ok,
+static int tn1010_get_link_status(struct cphy *phy, int *link_state,
 				  int *speed, int *duplex, int *fc)
 {
 	unsigned int status, lpa, adv;
@@ -139,8 +139,9 @@ static int tn1010_get_link_status(struct cphy *phy, int *link_ok,
 	if (err)
 		return err;
 
-	if (link_ok)
-		*link_ok = (status & F_LINK_STAT) != 0;
+	if (link_state)
+		*link_state = status & F_LINK_STAT ? PHY_LINK_UP :
+		    PHY_LINK_DOWN;
 
 	if (G_ANEG_STAT(status) == ANEG_COMPLETE) {
 		sp = (status & F_ANEG_SPEED_1G) ? SPEED_1000 : SPEED_10000;
diff --git a/sys/dev/cxgb/common/cxgb_vsc8211.c b/sys/dev/cxgb/common/cxgb_vsc8211.c
index 9d83859..4928e4a 100644
--- a/sys/dev/cxgb/common/cxgb_vsc8211.c
+++ b/sys/dev/cxgb/common/cxgb_vsc8211.c
@@ -129,7 +129,7 @@ static int vsc8211_autoneg_restart(struct cphy *cphy)
 				   BMCR_ANRESTART);
 }
 
-static int vsc8211_get_link_status(struct cphy *cphy, int *link_ok,
+static int vsc8211_get_link_status(struct cphy *cphy, int *link_state,
 				     int *speed, int *duplex, int *fc)
 {
 	unsigned int bmcr, status, lpa, adv;
@@ -141,7 +141,7 @@ static int vsc8211_get_link_status(struct cphy *cphy, int *link_ok,
 	if (err)
 		return err;
 
-	if (link_ok) {
+	if (link_state) {
 		/*
 		 * BMSR_LSTATUS is latch-low, so if it is 0 we need to read it
 		 * once more to get the current link state.
@@ -150,7 +150,8 @@ static int vsc8211_get_link_status(struct cphy *cphy, int *link_ok,
 			err = mdio_read(cphy, 0, MII_BMSR, &status);
 		if (err)
 			return err;
-		*link_ok = (status & BMSR_LSTATUS) != 0;
+		*link_state = status & BMSR_LSTATUS ? PHY_LINK_UP :
+		    PHY_LINK_DOWN;
 	}
 	if (!(bmcr & BMCR_ANENABLE)) {
 		dplx = (bmcr & BMCR_FULLDPLX) ? DUPLEX_FULL : DUPLEX_HALF;
@@ -201,7 +202,7 @@ static int vsc8211_get_link_status(struct cphy *cphy, int *link_ok,
 	return 0;
 }
 
-static int vsc8211_get_link_status_fiber(struct cphy *cphy, int *link_ok,
+static int vsc8211_get_link_status_fiber(struct cphy *cphy, int *link_state,
 					 int *speed, int *duplex, int *fc)
 {
 	unsigned int bmcr, status, lpa, adv;
@@ -213,7 +214,7 @@ static int vsc8211_get_link_status_fiber(struct cphy *cphy, int *link_ok,
 	if (err)
 		return err;
 
-	if (link_ok) {
+	if (link_state) {
 		/*
 		 * BMSR_LSTATUS is latch-low, so if it is 0 we need to read it
 		 * once more to get the current link state.
@@ -222,7 +223,8 @@ static int vsc8211_get_link_status_fiber(struct cphy *cphy, int *link_ok,
 			err = mdio_read(cphy, 0, MII_BMSR, &status);
 		if (err)
 			return err;
-		*link_ok = (status & BMSR_LSTATUS) != 0;
+		*link_state = status & BMSR_LSTATUS ? PHY_LINK_UP :
+		    PHY_LINK_DOWN;
 	}
 	if (!(bmcr & BMCR_ANENABLE)) {
 		dplx = (bmcr & BMCR_FULLDPLX) ? DUPLEX_FULL : DUPLEX_HALF;
diff --git a/sys/dev/cxgb/cxgb_main.c b/sys/dev/cxgb/cxgb_main.c
index 81e2f5a..293de3d 100644
--- a/sys/dev/cxgb/cxgb_main.c
+++ b/sys/dev/cxgb/cxgb_main.c
@@ -2226,7 +2226,8 @@ check_link_status(void *arg, int pending)
 
 	t3_link_changed(sc, pi->port_id);
 
-	if (pi->link_fault || !(pi->phy.caps & SUPPORTED_LINK_IRQ))
+	if (pi->link_fault || !(pi->phy.caps & SUPPORTED_LINK_IRQ) ||
+	    pi->link_config.link_ok == 0)
 		callout_reset(&pi->link_check_ch, hz, link_check_callout, pi);
 }