| Message ID | 1359147253-7204-1-git-send-email-nhorman@tuxdriver.com |
|---|---|
| State | Superseded, archived |
| Delegated to: | David Miller |
| Headers | show |
On Fri, 2013-01-25 at 15:54 -0500, Neil Horman wrote: > vmxnet3 fails to set netif_carrier_off on probe, meaning that when an interface > is opened the __LINK_STATE_NOCARRIER bit is already cleared, and so > /sys/class/net/<ifname>/operstate remains in the unknown state. Correct this by > setting netif_carrier_off on probe, like other drivers do. > > Also, while we're at it, lets remove the netif_carrier_ok checks from the > link_state_update function, as that check is atomically contained within the > netif_carrier_[on|off] functions anyway > > Tested successfully by myself > > Signed-off-by: Neil Horman <nhorman@tuxdriver.com> > CC: "David S. Miller" <davem@davemloft.net> > CC: "VMware, Inc." <pv-drivers@vmware.com> > --- > drivers/net/vmxnet3/vmxnet3_drv.c | 7 +++---- > 1 file changed, 3 insertions(+), 4 deletions(-) > > diff --git a/drivers/net/vmxnet3/vmxnet3_drv.c b/drivers/net/vmxnet3/vmxnet3_drv.c > index b1c90f8..66c26a9 100644 > --- a/drivers/net/vmxnet3/vmxnet3_drv.c > +++ b/drivers/net/vmxnet3/vmxnet3_drv.c [...] > @@ -3067,6 +3065,7 @@ vmxnet3_probe_device(struct pci_dev *pdev, > goto err_register; > } > > + netif_carrier_off(netdev); > vmxnet3_check_link(adapter, false); > return 0; > You should do this before calling register_netdev(), otherwise the link state change can race with vmxnet3_open(). Ben.
On Tuesday, January 29, 2013 05:54:11 PM Ben Hutchings wrote: > On Fri, 2013-01-25 at 15:54 -0500, Neil Horman wrote: > > vmxnet3 fails to set netif_carrier_off on probe, meaning that when an > > interface is opened the __LINK_STATE_NOCARRIER bit is already cleared, > > and so /sys/class/net/<ifname>/operstate remains in the unknown state. > > Correct this by setting netif_carrier_off on probe, like other drivers > > do. > > > > Also, while we're at it, lets remove the netif_carrier_ok checks from the > > link_state_update function, as that check is atomically contained within > > the netif_carrier_[on|off] functions anyway > > > > Tested successfully by myself > > > > Signed-off-by: Neil Horman <nhorman@tuxdriver.com> > > CC: "David S. Miller" <davem@davemloft.net> > > CC: "VMware, Inc." <pv-drivers@vmware.com> > > --- > > > > drivers/net/vmxnet3/vmxnet3_drv.c | 7 +++---- > > 1 file changed, 3 insertions(+), 4 deletions(-) > > > > diff --git a/drivers/net/vmxnet3/vmxnet3_drv.c > > b/drivers/net/vmxnet3/vmxnet3_drv.c index b1c90f8..66c26a9 100644 > > --- a/drivers/net/vmxnet3/vmxnet3_drv.c > > +++ b/drivers/net/vmxnet3/vmxnet3_drv.c > > [...] > > > @@ -3067,6 +3065,7 @@ vmxnet3_probe_device(struct pci_dev *pdev, > > > > goto err_register; > > > > } > > > > + netif_carrier_off(netdev); > > > > vmxnet3_check_link(adapter, false); > > return 0; > > You should do this before calling register_netdev(), otherwise the link > state change can race with vmxnet3_open(). Hmm, is it safe to do before the network device is registered? The rest of the drivers (e1000, tg3, etc) seem to be doing netif_carrier_off() in open() rather than probe() and then do link checks. Thanks, Dmitry -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Tue, Jan 29, 2013 at 10:54:28AM -0800, Dmitry Torokhov wrote: > On Tuesday, January 29, 2013 05:54:11 PM Ben Hutchings wrote: > > On Fri, 2013-01-25 at 15:54 -0500, Neil Horman wrote: > > > vmxnet3 fails to set netif_carrier_off on probe, meaning that when an > > > interface is opened the __LINK_STATE_NOCARRIER bit is already cleared, > > > and so /sys/class/net/<ifname>/operstate remains in the unknown state. > > > Correct this by setting netif_carrier_off on probe, like other drivers > > > do. > > > > > > Also, while we're at it, lets remove the netif_carrier_ok checks from the > > > link_state_update function, as that check is atomically contained within > > > the netif_carrier_[on|off] functions anyway > > > > > > Tested successfully by myself > > > > > > Signed-off-by: Neil Horman <nhorman@tuxdriver.com> > > > CC: "David S. Miller" <davem@davemloft.net> > > > CC: "VMware, Inc." <pv-drivers@vmware.com> > > > --- > > > > > > drivers/net/vmxnet3/vmxnet3_drv.c | 7 +++---- > > > 1 file changed, 3 insertions(+), 4 deletions(-) > > > > > > diff --git a/drivers/net/vmxnet3/vmxnet3_drv.c > > > b/drivers/net/vmxnet3/vmxnet3_drv.c index b1c90f8..66c26a9 100644 > > > --- a/drivers/net/vmxnet3/vmxnet3_drv.c > > > +++ b/drivers/net/vmxnet3/vmxnet3_drv.c > > > > [...] > > > > > @@ -3067,6 +3065,7 @@ vmxnet3_probe_device(struct pci_dev *pdev, > > > > > > goto err_register; > > > > > > } > > > > > > + netif_carrier_off(netdev); > > > > > > vmxnet3_check_link(adapter, false); > > > return 0; > > > > You should do this before calling register_netdev(), otherwise the link > > state change can race with vmxnet3_open(). > > Hmm, is it safe to do before the network device is registered? The rest > of the drivers (e1000, tg3, etc) seem to be doing netif_carrier_off() > in open() rather than probe() and then do link checks. > It appears to be safe to me, at least for the netif_carrier_off case, as theres a check to effecitvely just set the NOCARRIER bit if the device is in UNREGISTERED state. Neil > Thanks, > Dmitry > > -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Tue, 2013-01-29 at 14:09 -0500, Neil Horman wrote: > On Tue, Jan 29, 2013 at 10:54:28AM -0800, Dmitry Torokhov wrote: > > On Tuesday, January 29, 2013 05:54:11 PM Ben Hutchings wrote: > > > On Fri, 2013-01-25 at 15:54 -0500, Neil Horman wrote: > > > > vmxnet3 fails to set netif_carrier_off on probe, meaning that when an > > > > interface is opened the __LINK_STATE_NOCARRIER bit is already cleared, > > > > and so /sys/class/net/<ifname>/operstate remains in the unknown state. > > > > Correct this by setting netif_carrier_off on probe, like other drivers > > > > do. > > > > > > > > Also, while we're at it, lets remove the netif_carrier_ok checks from the > > > > link_state_update function, as that check is atomically contained within > > > > the netif_carrier_[on|off] functions anyway > > > > > > > > Tested successfully by myself > > > > > > > > Signed-off-by: Neil Horman <nhorman@tuxdriver.com> > > > > CC: "David S. Miller" <davem@davemloft.net> > > > > CC: "VMware, Inc." <pv-drivers@vmware.com> > > > > --- > > > > > > > > drivers/net/vmxnet3/vmxnet3_drv.c | 7 +++---- > > > > 1 file changed, 3 insertions(+), 4 deletions(-) > > > > > > > > diff --git a/drivers/net/vmxnet3/vmxnet3_drv.c > > > > b/drivers/net/vmxnet3/vmxnet3_drv.c index b1c90f8..66c26a9 100644 > > > > --- a/drivers/net/vmxnet3/vmxnet3_drv.c > > > > +++ b/drivers/net/vmxnet3/vmxnet3_drv.c > > > > > > [...] > > > > > > > @@ -3067,6 +3065,7 @@ vmxnet3_probe_device(struct pci_dev *pdev, > > > > > > > > goto err_register; > > > > > > > > } > > > > > > > > + netif_carrier_off(netdev); > > > > > > > > vmxnet3_check_link(adapter, false); > > > > return 0; > > > > > > You should do this before calling register_netdev(), otherwise the link > > > state change can race with vmxnet3_open(). > > > > Hmm, is it safe to do before the network device is registered? The rest > > of the drivers (e1000, tg3, etc) seem to be doing netif_carrier_off() > > in open() rather than probe() and then do link checks. > > > It appears to be safe to me, at least for the netif_carrier_off case, as theres > a check to effecitvely just set the NOCARRIER bit if the device is in > UNREGISTERED state. Right, I made this work properly a little while ago (commit 8f4cccbbd92f). Previous best practice was to call netif_carrier_off() after registration, but userland could still see the wrong operstate because linkwatch updates it asynchronously. Ben.
On Tuesday, January 29, 2013 07:17:10 PM Ben Hutchings wrote: > On Tue, 2013-01-29 at 14:09 -0500, Neil Horman wrote: > > On Tue, Jan 29, 2013 at 10:54:28AM -0800, Dmitry Torokhov wrote: > > > On Tuesday, January 29, 2013 05:54:11 PM Ben Hutchings wrote: > > > > On Fri, 2013-01-25 at 15:54 -0500, Neil Horman wrote: > > > > > vmxnet3 fails to set netif_carrier_off on probe, meaning that when > > > > > an > > > > > interface is opened the __LINK_STATE_NOCARRIER bit is already > > > > > cleared, > > > > > and so /sys/class/net/<ifname>/operstate remains in the unknown > > > > > state. > > > > > Correct this by setting netif_carrier_off on probe, like other > > > > > drivers > > > > > do. > > > > > > > > > > Also, while we're at it, lets remove the netif_carrier_ok checks > > > > > from the > > > > > link_state_update function, as that check is atomically contained > > > > > within > > > > > the netif_carrier_[on|off] functions anyway > > > > > > > > > > Tested successfully by myself > > > > > > > > > > Signed-off-by: Neil Horman <nhorman@tuxdriver.com> > > > > > CC: "David S. Miller" <davem@davemloft.net> > > > > > CC: "VMware, Inc." <pv-drivers@vmware.com> > > > > > --- > > > > > > > > > > drivers/net/vmxnet3/vmxnet3_drv.c | 7 +++---- > > > > > 1 file changed, 3 insertions(+), 4 deletions(-) > > > > > > > > > > diff --git a/drivers/net/vmxnet3/vmxnet3_drv.c > > > > > b/drivers/net/vmxnet3/vmxnet3_drv.c index b1c90f8..66c26a9 100644 > > > > > --- a/drivers/net/vmxnet3/vmxnet3_drv.c > > > > > +++ b/drivers/net/vmxnet3/vmxnet3_drv.c > > > > > > > > [...] > > > > > > > > > @@ -3067,6 +3065,7 @@ vmxnet3_probe_device(struct pci_dev *pdev, > > > > > > > > > > goto err_register; > > > > > > > > > > } > > > > > > > > > > + netif_carrier_off(netdev); > > > > > > > > > > vmxnet3_check_link(adapter, false); > > > > > return 0; > > > > > > > > You should do this before calling register_netdev(), otherwise the > > > > link > > > > state change can race with vmxnet3_open(). > > > > > > Hmm, is it safe to do before the network device is registered? The rest > > > of the drivers (e1000, tg3, etc) seem to be doing netif_carrier_off() > > > in open() rather than probe() and then do link checks. > > > > It appears to be safe to me, at least for the netif_carrier_off case, as > > theres a check to effecitvely just set the NOCARRIER bit if the device is > > in UNREGISTERED state. > > Right, I made this work properly a little while ago (commit > 8f4cccbbd92f). Previous best practice was to call netif_carrier_off() > after registration, but userland could still see the wrong operstate > because linkwatch updates it asynchronously. Ah, OK I see. Thank you for the explanation. Thanks, Dmitry -- To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/drivers/net/vmxnet3/vmxnet3_drv.c b/drivers/net/vmxnet3/vmxnet3_drv.c index b1c90f8..66c26a9 100644 --- a/drivers/net/vmxnet3/vmxnet3_drv.c +++ b/drivers/net/vmxnet3/vmxnet3_drv.c @@ -150,8 +150,7 @@ vmxnet3_check_link(struct vmxnet3_adapter *adapter, bool affectTxQueue) if (ret & 1) { /* Link is up. */ netdev_info(adapter->netdev, "NIC Link is Up %d Mbps\n", adapter->link_speed); - if (!netif_carrier_ok(adapter->netdev)) - netif_carrier_on(adapter->netdev); + netif_carrier_on(adapter->netdev); if (affectTxQueue) { for (i = 0; i < adapter->num_tx_queues; i++) @@ -160,8 +159,7 @@ vmxnet3_check_link(struct vmxnet3_adapter *adapter, bool affectTxQueue) } } else { netdev_info(adapter->netdev, "NIC Link is Down\n"); - if (netif_carrier_ok(adapter->netdev)) - netif_carrier_off(adapter->netdev); + netif_carrier_off(adapter->netdev); if (affectTxQueue) { for (i = 0; i < adapter->num_tx_queues; i++) @@ -3067,6 +3065,7 @@ vmxnet3_probe_device(struct pci_dev *pdev, goto err_register; } + netif_carrier_off(netdev); vmxnet3_check_link(adapter, false); return 0;
vmxnet3 fails to set netif_carrier_off on probe, meaning that when an interface is opened the __LINK_STATE_NOCARRIER bit is already cleared, and so /sys/class/net/<ifname>/operstate remains in the unknown state. Correct this by setting netif_carrier_off on probe, like other drivers do. Also, while we're at it, lets remove the netif_carrier_ok checks from the link_state_update function, as that check is atomically contained within the netif_carrier_[on|off] functions anyway Tested successfully by myself Signed-off-by: Neil Horman <nhorman@tuxdriver.com> CC: "David S. Miller" <davem@davemloft.net> CC: "VMware, Inc." <pv-drivers@vmware.com> --- drivers/net/vmxnet3/vmxnet3_drv.c | 7 +++---- 1 file changed, 3 insertions(+), 4 deletions(-)