* [PATCH net] mlx4: xdp_prog becomes inactive after ethtool '-L' or '-G'
@ 2017-01-28 7:40 Martin KaFai Lau
2017-01-30 14:56 ` David Miller
2017-01-30 17:18 ` Tariq Toukan
0 siblings, 2 replies; 6+ messages in thread
From: Martin KaFai Lau @ 2017-01-28 7:40 UTC (permalink / raw)
To: netdev; +Cc: Brenden Blanco, Saeed Mahameed, Tariq Toukan, Kernel Team
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 <kafai@fb.com>
---
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
^ permalink raw reply related [flat|nested] 6+ messages in thread
* Re: [PATCH net] mlx4: xdp_prog becomes inactive after ethtool '-L' or '-G'
2017-01-28 7:40 [PATCH net] mlx4: xdp_prog becomes inactive after ethtool '-L' or '-G' Martin KaFai Lau
@ 2017-01-30 14:56 ` David Miller
2017-01-30 17:18 ` Tariq Toukan
1 sibling, 0 replies; 6+ messages in thread
From: David Miller @ 2017-01-30 14:56 UTC (permalink / raw)
To: kafai; +Cc: netdev, bblanco, saeedm, tariqt, kernel-team
From: Martin KaFai Lau <kafai@fb.com>
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 <kafai@fb.com>
Mellanox folks, please review.
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [PATCH net] mlx4: xdp_prog becomes inactive after ethtool '-L' or '-G'
2017-01-28 7:40 [PATCH net] mlx4: xdp_prog becomes inactive after ethtool '-L' or '-G' Martin KaFai Lau
2017-01-30 14:56 ` David Miller
@ 2017-01-30 17:18 ` Tariq Toukan
2017-01-31 18:11 ` David Miller
2017-02-01 5:54 ` Martin KaFai Lau
1 sibling, 2 replies; 6+ messages in thread
From: Tariq Toukan @ 2017-01-30 17:18 UTC (permalink / raw)
To: Martin KaFai Lau, netdev
Cc: Brenden Blanco, Saeed Mahameed, Tariq Toukan, Kernel Team
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 <kafai@fb.com>
> ---
...
> 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 (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.
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [PATCH net] mlx4: xdp_prog becomes inactive after ethtool '-L' or '-G'
2017-01-30 17:18 ` Tariq Toukan
@ 2017-01-31 18:11 ` David Miller
2017-01-31 21:32 ` Martin KaFai Lau
2017-02-01 5:54 ` Martin KaFai Lau
1 sibling, 1 reply; 6+ messages in thread
From: David Miller @ 2017-01-31 18:11 UTC (permalink / raw)
To: tariqt; +Cc: kafai, netdev, bblanco, saeedm, kernel-team
From: Tariq Toukan <tariqt@mellanox.com>
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.
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [PATCH net] mlx4: xdp_prog becomes inactive after ethtool '-L' or '-G'
2017-01-31 18:11 ` David Miller
@ 2017-01-31 21:32 ` Martin KaFai Lau
0 siblings, 0 replies; 6+ messages in thread
From: Martin KaFai Lau @ 2017-01-31 21:32 UTC (permalink / raw)
To: David Miller; +Cc: tariqt, netdev, bblanco, saeedm, kernel-team
On Tue, Jan 31, 2017 at 01:11:40PM -0500, David Miller wrote:
> From: Tariq Toukan <tariqt@mellanox.com>
> 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.
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
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [PATCH net] mlx4: xdp_prog becomes inactive after ethtool '-L' or '-G'
2017-01-30 17:18 ` Tariq Toukan
2017-01-31 18:11 ` David Miller
@ 2017-02-01 5:54 ` Martin KaFai Lau
1 sibling, 0 replies; 6+ messages in thread
From: Martin KaFai Lau @ 2017-02-01 5:54 UTC (permalink / raw)
To: Tariq Toukan; +Cc: netdev, Brenden Blanco, Saeed Mahameed, Kernel Team
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:
> > 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 <kafai@fb.com>
> > ---
> ...
> > 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
> > +
> > 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.
^ permalink raw reply [flat|nested] 6+ messages in thread
end of thread, other threads:[~2017-02-01 5:54 UTC | newest]
Thread overview: 6+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2017-01-28 7:40 [PATCH net] mlx4: xdp_prog becomes inactive after ethtool '-L' or '-G' Martin KaFai Lau
2017-01-30 14:56 ` David Miller
2017-01-30 17:18 ` Tariq Toukan
2017-01-31 18:11 ` David Miller
2017-01-31 21:32 ` Martin KaFai Lau
2017-02-01 5:54 ` Martin KaFai Lau
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.