[PATCH net] mlx4: xdp_prog becomes inactive after ethtool '-L' or '-G'

If the rx-queues ever get re-initialized (e.g. by changing the
number of rx-queues with ethtool -L), the existing xdp_prog becomes
inactive.

The bug is that the xdp_prog ptr has not been carried over from
the old rx-queues to the new rx-queues

Fixes: 47a38e155037 ("net/mlx4_en: add support for fast rx drop bpf program")
Signed-off-by: Martin KaFai Lau <redacted>
---
 drivers/net/ethernet/mellanox/mlx4/en_ethtool.c |  4 +-
 drivers/net/ethernet/mellanox/mlx4/en_netdev.c  | 52 ++++++++++++++++++++-----
 drivers/net/ethernet/mellanox/mlx4/mlx4_en.h    |  3 +-
 3 files changed, 47 insertions(+), 12 deletions(-)

diff --git a/drivers/net/ethernet/mellanox/mlx4/en_ethtool.c b/drivers/net/ethernet/mellanox/mlx4/en_ethtool.c
index d5a9372ed84d..9aa422691954 100644
--- a/drivers/net/ethernet/mellanox/mlx4/en_ethtool.c
+++ b/drivers/net/ethernet/mellanox/mlx4/en_ethtool.c

@@ -1099,7 +1099,7 @@ static int mlx4_en_set_ringparam(struct net_device *dev,
 	memcpy(&new_prof, priv->prof, sizeof(struct mlx4_en_port_profile));
 	new_prof.tx_ring_size = tx_size;
 	new_prof.rx_ring_size = rx_size;
-	err = mlx4_en_try_alloc_resources(priv, tmp, &new_prof);
+	err = mlx4_en_try_alloc_resources(priv, tmp, &new_prof, true);
 	if (err)
 		goto out;
 

@@ -1774,7 +1774,7 @@ static int mlx4_en_set_channels(struct net_device *dev,
 	new_prof.tx_ring_num[TX_XDP] = xdp_count;
 	new_prof.rx_ring_num = channel->rx_count;
 
-	err = mlx4_en_try_alloc_resources(priv, tmp, &new_prof);
+	err = mlx4_en_try_alloc_resources(priv, tmp, &new_prof, true);
 	if (err)
 		goto out;
 

diff --git a/drivers/net/ethernet/mellanox/mlx4/en_netdev.c b/drivers/net/ethernet/mellanox/mlx4/en_netdev.c
index 761f8b12399c..f4179086b3c6 100644
--- a/drivers/net/ethernet/mellanox/mlx4/en_netdev.c
+++ b/drivers/net/ethernet/mellanox/mlx4/en_netdev.c

@@ -2184,23 +2184,57 @@ static void mlx4_en_update_priv(struct mlx4_en_priv *dst,
 
 int mlx4_en_try_alloc_resources(struct mlx4_en_priv *priv,
 				struct mlx4_en_priv *tmp,
-				struct mlx4_en_port_profile *prof)
+				struct mlx4_en_port_profile *prof,
+				bool carry_xdp_prog)
 {
-	int t;
+	struct bpf_prog *xdp_prog = NULL;
+	int err;
+	int i;
 
 	mlx4_en_copy_priv(tmp, priv, prof);
 
+	if (carry_xdp_prog) {
+		/* All rx_rings has the same xdp_prog.  Pick the first one */
+		xdp_prog = rcu_dereference_protected(
+			priv->rx_ring[0]->xdp_prog,
+			lockdep_is_held(&priv->mdev->state_lock));
+
+		if (xdp_prog) {
+			xdp_prog = bpf_prog_add(xdp_prog, tmp->rx_ring_num);
+			if (IS_ERR(xdp_prog)) {
+				err = PTR_ERR(xdp_prog);
+				xdp_prog = NULL;
+				goto err_free;
+			}
+		}
+	}
+
 	if (mlx4_en_alloc_resources(tmp)) {
 		en_warn(priv,
 			"%s: Resource allocation failed, using previous configuration\n",
 			__func__);
-		for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
-			kfree(tmp->tx_ring[t]);
-			kfree(tmp->tx_cq[t]);
-		}
-		return -ENOMEM;
+		err = -ENOMEM;
+		goto err_free;
+	}
+
+	if (xdp_prog) {
+		for (i = 0; i < tmp->rx_ring_num; i++)
+			rcu_assign_pointer(tmp->rx_ring[i]->xdp_prog,
+					   xdp_prog);
 	}
+
 	return 0;
+
+err_free:
+	if (xdp_prog)
+		bpf_prog_sub(xdp_prog, tmp->rx_ring_num);
+
+	for (i = 0; i < MLX4_EN_NUM_TX_TYPES; i++) {
+		kfree(tmp->tx_ring[i]);
+		kfree(tmp->tx_cq[i]);
+	}
+
+	return err;
 }
 
 void mlx4_en_safe_replace_resources(struct mlx4_en_priv *priv,

@@ -2755,7 +2789,7 @@ static int mlx4_xdp_set(struct net_device *dev, struct bpf_prog *prog)
 		en_warn(priv, "Reducing the number of TX rings, to not exceed the max total rings number.\n");
 	}
 
-	err = mlx4_en_try_alloc_resources(priv, tmp, &new_prof);
+	err = mlx4_en_try_alloc_resources(priv, tmp, &new_prof, false);
 	if (err) {
 		if (prog)
 			bpf_prog_sub(prog, priv->rx_ring_num - 1);

@@ -3499,7 +3533,7 @@ int mlx4_en_reset_config(struct net_device *dev,
 	memcpy(&new_prof, priv->prof, sizeof(struct mlx4_en_port_profile));
 	memcpy(&new_prof.hwtstamp_config, &ts_config, sizeof(ts_config));
 
-	err = mlx4_en_try_alloc_resources(priv, tmp, &new_prof);
+	err = mlx4_en_try_alloc_resources(priv, tmp, &new_prof, true);
 	if (err)
 		goto out;
 

diff --git a/drivers/net/ethernet/mellanox/mlx4/mlx4_en.h b/drivers/net/ethernet/mellanox/mlx4/mlx4_en.h
index ba1c6cd0cc79..cec59bc264c9 100644
--- a/drivers/net/ethernet/mellanox/mlx4/mlx4_en.h
+++ b/drivers/net/ethernet/mellanox/mlx4/mlx4_en.h

@@ -679,7 +679,8 @@ void mlx4_en_set_stats_bitmap(struct mlx4_dev *dev,
 
 int mlx4_en_try_alloc_resources(struct mlx4_en_priv *priv,
 				struct mlx4_en_priv *tmp,
-				struct mlx4_en_port_profile *prof);
+				struct mlx4_en_port_profile *prof,
+				bool carry_xdp_prog);
 void mlx4_en_safe_replace_resources(struct mlx4_en_priv *priv,
 				    struct mlx4_en_priv *tmp);
 

-- 
2.5.1

From: Martin KaFai Lau <redacted>
Date: Fri, 27 Jan 2017 23:40:51 -0800

If the rx-queues ever get re-initialized (e.g. by changing the
number of rx-queues with ethtool -L), the existing xdp_prog becomes
inactive.

The bug is that the xdp_prog ptr has not been carried over from
the old rx-queues to the new rx-queues

Fixes: 47a38e155037 ("net/mlx4_en: add support for fast rx drop bpf program")
Signed-off-by: Martin KaFai Lau <redacted>

Mellanox folks, please review.

Hi Martin,

Thanks for your patch.

It looks good to me, in general.
I just have one small comment below.

On 28/01/2017 9:40 AM, Martin KaFai Lau wrote:

If the rx-queues ever get re-initialized (e.g. by changing the
number of rx-queues with ethtool -L), the existing xdp_prog becomes
inactive.

The bug is that the xdp_prog ptr has not been carried over from
the old rx-queues to the new rx-queues

Fixes: 47a38e155037 ("net/mlx4_en: add support for fast rx drop bpf program")
Signed-off-by: Martin KaFai Lau <redacted>
---

...

Why do you prefer dealing with xdp_prog in two stages? You can handle it 
all at once, after "mlx4_en_alloc_resources()" succeeds.

+
  	if (mlx4_en_alloc_resources(tmp)) {
  		en_warn(priv,
  			"%s: Resource allocation failed, using previous configuration\n",
  			__func__);
-		for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
-			kfree(tmp->tx_ring[t]);
-			kfree(tmp->tx_cq[t]);
-		}
-		return -ENOMEM;
+		err = -ENOMEM;
+		goto err_free;
+	}
+
+	if (xdp_prog) {
+		for (i = 0; i < tmp->rx_ring_num; i++)
+			rcu_assign_pointer(tmp->rx_ring[i]->xdp_prog,
+					   xdp_prog);
  	}
+
  	return 0;
+
+err_free:
+	if (xdp_prog)
+		bpf_prog_sub(xdp_prog, tmp->rx_ring_num);
+
+	for (i = 0; i < MLX4_EN_NUM_TX_TYPES; i++) {
+		kfree(tmp->tx_ring[i]);
+		kfree(tmp->tx_cq[i]);
+	}
+
+	return err;
  }
  
  void mlx4_en_safe_replace_resources(struct mlx4_en_priv *priv,

Regards,
Tariq Toukan.

From: Tariq Toukan <redacted>
Date: Mon, 30 Jan 2017 19:18:28 +0200

It looks good to me, in general.
I just have one small comment below.

Martin, please address Tariq's feedback.

Thank you.

On Tue, Jan 31, 2017 at 01:11:40PM -0500, David Miller wrote:

From: Tariq Toukan <redacted>
Date: Mon, 30 Jan 2017 19:18:28 +0200

quoted

It looks good to me, in general.
I just have one small comment below.

Martin, please address Tariq's feedback.

Sorry for the delay.  I am on PTO for this week and
have been travelling.

I will take a closer look when I get
a more stable intenet connection tonight.

Thanks,
--Martin

On Mon, Jan 30, 2017 at 07:18:28PM +0200, Tariq Toukan wrote:

Hi Martin,

Thanks for your patch.

It looks good to me, in general.
I just have one small comment below.

Thanks for your feedback and sorry for the delay.

On 28/01/2017 9:40 AM, Martin KaFai Lau wrote:

quoted

If the rx-queues ever get re-initialized (e.g. by changing the
number of rx-queues with ethtool -L), the existing xdp_prog becomes
inactive.

The bug is that the xdp_prog ptr has not been carried over from
the old rx-queues to the new rx-queues

Fixes: 47a38e155037 ("net/mlx4_en: add support for fast rx drop bpf program")
Signed-off-by: Martin KaFai Lau <redacted>
---

...

quoted

diff --git a/drivers/net/ethernet/mellanox/mlx4/en_netdev.c b/drivers/net/ethernet/mellanox/mlx4/en_netdev.c
index 761f8b12399c..f4179086b3c6 100644
--- a/drivers/net/ethernet/mellanox/mlx4/en_netdev.c
+++ b/drivers/net/ethernet/mellanox/mlx4/en_netdev.c

@@ -2184,23 +2184,57 @@ static void mlx4_en_update_priv(struct mlx4_en_priv *dst,
  int mlx4_en_try_alloc_resources(struct mlx4_en_priv *priv,
  				struct mlx4_en_priv *tmp,
-				struct mlx4_en_port_profile *prof)
+				struct mlx4_en_port_profile *prof,
+				bool carry_xdp_prog)
  {
-	int t;
+	struct bpf_prog *xdp_prog = NULL;
+	int err;
+	int i;
  	mlx4_en_copy_priv(tmp, priv, prof);
+	if (carry_xdp_prog) {
+		/* All rx_rings has the same xdp_prog.  Pick the first one */
+		xdp_prog = rcu_dereference_protected(
+			priv->rx_ring[0]->xdp_prog,
+			lockdep_is_held(&priv->mdev->state_lock));
+
+		if (xdp_prog) {
+			xdp_prog = bpf_prog_add(xdp_prog, tmp->rx_ring_num);
+			if (IS_ERR(xdp_prog)) {
+				err = PTR_ERR(xdp_prog);
+				xdp_prog = NULL;
+				goto err_free;
+			}
+		}
+	}

Why do you prefer dealing with xdp_prog in two stages? You can handle it all
at once, after "mlx4_en_alloc_resources()" succeeds.

If bpf_prog_add() did fail, resources allocated for tmp had to
be freed.  I was thinking it is not safe to call mlx4_en_free_resources()
at this point.  Since your feedback, I took another look and re-read the
'goto err' path in mlx4_en_alloc_resources(), I realized we can use
mlx4_en_free_resources() here for the bpf_prog_add() error case.  Hence,
agree with your suggestion.

A side note though:
after taking another look at mlx4_en_free_resources(), I might have found
another issue.  I need to run some tests to confirm first to avoid any false
alarm.

I will post v2.

Thanks,
--Martin

quoted

+
  	if (mlx4_en_alloc_resources(tmp)) {
  		en_warn(priv,
  			"%s: Resource allocation failed, using previous configuration\n",
  			__func__);
-		for (t = 0; t < MLX4_EN_NUM_TX_TYPES; t++) {
-			kfree(tmp->tx_ring[t]);
-			kfree(tmp->tx_cq[t]);
-		}
-		return -ENOMEM;
+		err = -ENOMEM;
+		goto err_free;
+	}
+
+	if (xdp_prog) {
+		for (i = 0; i < tmp->rx_ring_num; i++)
+			rcu_assign_pointer(tmp->rx_ring[i]->xdp_prog,
+					   xdp_prog);
  	}
+
  	return 0;
+
+err_free:
+	if (xdp_prog)
+		bpf_prog_sub(xdp_prog, tmp->rx_ring_num);
+
+	for (i = 0; i < MLX4_EN_NUM_TX_TYPES; i++) {
+		kfree(tmp->tx_ring[i]);
+		kfree(tmp->tx_cq[i]);
+	}
+
+	return err;
  }
  void mlx4_en_safe_replace_resources(struct mlx4_en_priv *priv,

Regards,
Tariq Toukan.