Skip to contentSkip to conversation

tg3: Fix phylib locking strategy

PATCH Superseded Matt Carlson 2 participants 3 messages Oct 5, 2009 – Oct 6, 2009

Subsystems (2)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 tg3: Fix phylib locking strategy

Felix, can you verify this solves your problem?

---

Felix Radensky noted that chip resets were generating stack trace dumps. This is because the driver is attempting to acquire the mdio bus mutex while holding the tp->lock spinlock. The fix is to change the code such that every phy access takes the tp->lock spinlock instead.

---
 drivers/net/tg3.c |   37 ++++++++++++-------------------------
 drivers/net/tg3.h |    1 -
 2 files changed, 12 insertions(+), 26 deletions(-)
drivers/net/tg3.c +12 -25
index f09bc5d..8afc60e 100644
@@ -902,11 +902,12 @@ static int tg3_mdio_read(struct mii_bus *bp, int mii_id, int reg)
 	struct tg3 *tp = bp->priv;
 	u32 val;
 
-	if (tp->tg3_flags3 & TG3_FLG3_MDIOBUS_PAUSED)
-		return -EAGAIN;
+	spin_lock(&tp->lock);
 
 	if (tg3_readphy(tp, reg, &val))
-		return -EIO;
+		val = -EIO;
+
+	spin_unlock(&tp->lock);
 
 	return val;
 }
@@ -914,14 +915,16 @@ static int tg3_mdio_read(struct mii_bus *bp, int mii_id, int reg)
 static int tg3_mdio_write(struct mii_bus *bp, int mii_id, int reg, u16 val)
 {
 	struct tg3 *tp = bp->priv;
+	u32 ret = 0;
 
-	if (tp->tg3_flags3 & TG3_FLG3_MDIOBUS_PAUSED)
-		return -EAGAIN;
+	spin_lock(&tp->lock);
 
 	if (tg3_writephy(tp, reg, val))
-		return -EIO;
+		ret = -EIO;
 
-	return 0;
+	spin_unlock(&tp->lock);
+
+	return ret;
 }
 
 static int tg3_mdio_reset(struct mii_bus *bp)
@@ -1011,12 +1014,6 @@ static void tg3_mdio_config_5785(struct tg3 *tp)
 
 static void tg3_mdio_start(struct tg3 *tp)
 {
-	if (tp->tg3_flags3 & TG3_FLG3_MDIOBUS_INITED) {
-		mutex_lock(&tp->mdio_bus->mdio_lock);
-		tp->tg3_flags3 &= ~TG3_FLG3_MDIOBUS_PAUSED;
-		mutex_unlock(&tp->mdio_bus->mdio_lock);
-	}
-
 	tp->mi_mode &= ~MAC_MI_MODE_AUTO_POLL;
 	tw32_f(MAC_MI_MODE, tp->mi_mode);
 	udelay(80);
@@ -1041,15 +1038,6 @@ static void tg3_mdio_start(struct tg3 *tp)
 		tg3_mdio_config_5785(tp);
 }
 
-static void tg3_mdio_stop(struct tg3 *tp)
-{
-	if (tp->tg3_flags3 & TG3_FLG3_MDIOBUS_INITED) {
-		mutex_lock(&tp->mdio_bus->mdio_lock);
-		tp->tg3_flags3 |= TG3_FLG3_MDIOBUS_PAUSED;
-		mutex_unlock(&tp->mdio_bus->mdio_lock);
-	}
-}
-
 static int tg3_mdio_init(struct tg3 *tp)
 {
 	int i;
@@ -1141,7 +1129,6 @@ static void tg3_mdio_fini(struct tg3 *tp)
 		tp->tg3_flags3 &= ~TG3_FLG3_MDIOBUS_INITED;
 		mdiobus_unregister(tp->mdio_bus);
 		mdiobus_free(tp->mdio_bus);
-		tp->tg3_flags3 &= ~TG3_FLG3_MDIOBUS_PAUSED;
 	}
 }
 
@@ -6392,8 +6379,6 @@ static int tg3_chip_reset(struct tg3 *tp)
 
 	tg3_nvram_lock(tp);
 
-	tg3_mdio_stop(tp);
-
 	tg3_ape_lock(tp, TG3_APE_LOCK_GRC);
 
 	/* No matching tg3_nvram_unlock() after this because
@@ -8698,6 +8683,8 @@ static int tg3_close(struct net_device *dev)
 
 	del_timer_sync(&tp->timer);
 
+	tg3_phy_stop(tp);
+
 	tg3_full_lock(tp, 1);
 #if 0
 	tg3_dump_state(tp);
drivers/net/tg3.h +0 -1
index 524691c..bab7940 100644
@@ -2748,7 +2748,6 @@ struct tg3 {
 #define TG3_FLG3_5701_DMA_BUG		0x00000008
 #define TG3_FLG3_USE_PHYLIB		0x00000010
 #define TG3_FLG3_MDIOBUS_INITED		0x00000020
-#define TG3_FLG3_MDIOBUS_PAUSED		0x00000040
 #define TG3_FLG3_PHY_CONNECTED		0x00000080
 #define TG3_FLG3_RGMII_STD_IBND_DISABLE	0x00000100
 #define TG3_FLG3_RGMII_EXT_IBND_RX_EN	0x00000200
Signature · 1 line
1.6.4.4
#1 lore raw
Quoted 20 lines · Matt Carlson

Felix, can you verify this solves your problem?

---

Felix Radensky noted that chip resets were generating stack trace dumps. This is because the driver is attempting to acquire the mdio bus mutex while holding the tp->lock spinlock. The fix is to change the code such that every phy access takes the tp->lock spinlock instead.

---
 drivers/net/tg3.c |   37 ++++++++++++-------------------------
 drivers/net/tg3.h |    1 -
 2 files changed, 12 insertions(+), 26 deletions(-)
drivers/net/tg3.c +0 -0
index f09bc5d..8afc60e 100644

drivers/net/tg3.c static int tg3_mdio_read(struct mii_bus *bp, int mii_id, int reg) · line 906

Quoted 7 lines
drivers/net/tg3.c +1 -2
@@ -902,11 +902,12 @@ static int tg3_mdio_read(struct mii_bus *bp, int mii_id, int reg)
 	struct tg3 *tp = bp->priv;
 	u32 val;
 
-	if (tp->tg3_flags3 & TG3_FLG3_MDIOBUS_PAUSED)
-		return -EAGAIN;
+	spin_lock(&tp->lock);
 

Matt, do we need spin_lock_bh() here? The tp->lock can be taken in NAPI poll BH context.

Quoted 96 lines
drivers/net/tg3.c +11 -23
 	if (tg3_readphy(tp, reg, &val))
-		return -EIO;
+		val = -EIO;
+
+	spin_unlock(&tp->lock);
 
 	return val;
 }
@@ -914,14 +915,16 @@ static int tg3_mdio_read(struct mii_bus *bp, int mii_id, int reg)
 static int tg3_mdio_write(struct mii_bus *bp, int mii_id, int reg, u16 val)
 {
 	struct tg3 *tp = bp->priv;
+	u32 ret = 0;
 
-	if (tp->tg3_flags3 & TG3_FLG3_MDIOBUS_PAUSED)
-		return -EAGAIN;
+	spin_lock(&tp->lock);
 
 	if (tg3_writephy(tp, reg, val))
-		return -EIO;
+		ret = -EIO;
 
-	return 0;
+	spin_unlock(&tp->lock);
+
+	return ret;
 }
 
 static int tg3_mdio_reset(struct mii_bus *bp)
@@ -1011,12 +1014,6 @@ static void tg3_mdio_config_5785(struct tg3 *tp)
 
 static void tg3_mdio_start(struct tg3 *tp)
 {
-	if (tp->tg3_flags3 & TG3_FLG3_MDIOBUS_INITED) {
-		mutex_lock(&tp->mdio_bus->mdio_lock);
-		tp->tg3_flags3 &= ~TG3_FLG3_MDIOBUS_PAUSED;
-		mutex_unlock(&tp->mdio_bus->mdio_lock);
-	}
-
 	tp->mi_mode &= ~MAC_MI_MODE_AUTO_POLL;
 	tw32_f(MAC_MI_MODE, tp->mi_mode);
 	udelay(80);
@@ -1041,15 +1038,6 @@ static void tg3_mdio_start(struct tg3 *tp)
 		tg3_mdio_config_5785(tp);
 }
 
-static void tg3_mdio_stop(struct tg3 *tp)
-{
-	if (tp->tg3_flags3 & TG3_FLG3_MDIOBUS_INITED) {
-		mutex_lock(&tp->mdio_bus->mdio_lock);
-		tp->tg3_flags3 |= TG3_FLG3_MDIOBUS_PAUSED;
-		mutex_unlock(&tp->mdio_bus->mdio_lock);
-	}
-}
-
 static int tg3_mdio_init(struct tg3 *tp)
 {
 	int i;
@@ -1141,7 +1129,6 @@ static void tg3_mdio_fini(struct tg3 *tp)
 		tp->tg3_flags3 &= ~TG3_FLG3_MDIOBUS_INITED;
 		mdiobus_unregister(tp->mdio_bus);
 		mdiobus_free(tp->mdio_bus);
-		tp->tg3_flags3 &= ~TG3_FLG3_MDIOBUS_PAUSED;
 	}
 }
 
@@ -6392,8 +6379,6 @@ static int tg3_chip_reset(struct tg3 *tp)
 
 	tg3_nvram_lock(tp);
 
-	tg3_mdio_stop(tp);
-
 	tg3_ape_lock(tp, TG3_APE_LOCK_GRC);
 
 	/* No matching tg3_nvram_unlock() after this because
@@ -8698,6 +8683,8 @@ static int tg3_close(struct net_device *dev)
 
 	del_timer_sync(&tp->timer);
 
+	tg3_phy_stop(tp);
+
 	tg3_full_lock(tp, 1);
 #if 0
 	tg3_dump_state(tp);
drivers/net/tg3.h +0 -1
index 524691c..bab7940 100644
@@ -2748,7 +2748,6 @@ struct tg3 {
 #define TG3_FLG3_5701_DMA_BUG		0x00000008
 #define TG3_FLG3_USE_PHYLIB		0x00000010
 #define TG3_FLG3_MDIOBUS_INITED		0x00000020
-#define TG3_FLG3_MDIOBUS_PAUSED		0x00000040
 #define TG3_FLG3_PHY_CONNECTED		0x00000080
 #define TG3_FLG3_RGMII_STD_IBND_DISABLE	0x00000100
 #define TG3_FLG3_RGMII_EXT_IBND_RX_EN	0x00000200
#2 lore raw
Quoted 25 lines · Michael Chan
Quoted 23 lines · Matt Carlson

Felix, can you verify this solves your problem?

---

Felix Radensky noted that chip resets were generating stack trace dumps. This is because the driver is attempting to acquire the mdio bus mutex while holding the tp->lock spinlock. The fix is to change the code such that every phy access takes the tp->lock spinlock instead.

---
 drivers/net/tg3.c |   37 ++++++++++++-------------------------
 drivers/net/tg3.h |    1 -
 2 files changed, 12 insertions(+), 26 deletions(-)
drivers/net/tg3.c +0 -0
index f09bc5d..8afc60e 100644
@@ -902,11 +902,12 @@ static int tg3_mdio_read(struct mii_bus *bp, int mii_id, int reg)
 	struct tg3 *tp = bp->priv;
 	u32 val;
 
Quoted 7 lines

drivers/net/tg3.c static int tg3_mdio_read(struct mii_bus *bp, int mii_id, int reg) · line 906

Quoted 4 lines
drivers/net/tg3.c +1 -2
-	if (tp->tg3_flags3 & TG3_FLG3_MDIOBUS_PAUSED)
-		return -EAGAIN;
+	spin_lock(&tp->lock);
 

Matt, do we need spin_lock_bh() here? The tp->lock can be taken in NAPI poll BH context.

Yes. You are right. And the same change needs to be made in tg3_adjust_link(). V2 of the patch, coming up.