| Message ID | 20201028055742.74941-2-ljp@linux.ibm.com (mailing list archive) |
|---|---|
| State | Not Applicable |
| Delegated to: | Netdev Maintainers |
| Headers | show |
| Series | ibmvnic: fixes in reset path | expand |
On Wed, 28 Oct 2020 00:57:41 -0500 Lijun Pan wrote: > We need to notify peers only when failover and migration happen. > It is unnecessary to call that in other events like > FATAL, NON_FATAL, CHANGE_PARAM, and TIMEOUT resets > since in those scenarios the MAC address and ip address mapping > does not change. Originally all the resets except CHANGE_PARAM > are processed by do_reset such that we need to find out > failover and migration cases in do_reset and call notifier functions. > We only need to notify peers in do_reset and do_hard_reset. > We don't need notify peers in do_change_param_reset since it is > a CHANGE_PARAM reset. In a nested reset case, it will finally > call into do_hard_reset with reasons other than failvoer and > migration. So, we don't need to check the reset reason in > do_hard_reset and just call notifier functions anyway. You're completely undoing the commit you linked to: commit 61d3e1d9bc2a1910d773cbf4ed6f587a7a6166b5 Author: Nathan Fontenot <nfont@linux.vnet.ibm.com> Date: Mon Jun 12 20:47:45 2017 -0400 ibmvnic: Remove netdev notify for failover resets When handling a driver reset due to a failover of the backing server on the vios, doing the netdev_notify_peers() can cause network traffic to stall or halt. Remove the netdev notify call for failover resets. Signed-off-by: Nathan Fontenot <nfont@linux.vnet.ibm.com> Signed-off-by: David S. Miller <davem@davemloft.net> diff --git a/drivers/net/ethernet/ibm/ibmvnic.c b/drivers/net/ethernet/ibm/ibmvnic.c index fd3ef3005fb0..59ea7a5ae776 100644 --- a/drivers/net/ethernet/ibm/ibmvnic.c +++ b/drivers/net/ethernet/ibm/ibmvnic.c @@ -1364,7 +1364,9 @@ static int do_reset(struct ibmvnic_adapter *adapter, for (i = 0; i < adapter->req_rx_queues; i++) napi_schedule(&adapter->napi[i]); - netdev_notify_peers(netdev); + if (adapter->reset_reason != VNIC_RESET_FAILOVER) + netdev_notify_peers(netdev); + return 0; } But you don't seem to address why this change was unnecessary. AFAIK you're saying "we only need this event for FAILOVER and MOBILITY" but the previous commit _excluded_ FAILOVER for some vague reason. If the previous commit was incorrect you need to explain that in the commit message. > netdev_notify_peers calls below two functions with rtnl lock(). > call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); > call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev); > When netdev_notify_peers was substituted in > commit 986103e7920c ("net/ibmvnic: Fix RTNL deadlock during device reset"), > call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev) was missed. That should be a separate patch. > Fixes: 61d3e1d9bc2a ("ibmvnic: Remove netdev notify for failover resets") > Fixes: 986103e7920c ("net/ibmvnic: Fix RTNL deadlock during device > reset") Please don't line-wrap fixes tags. > Suggested-by: Brian King <brking@linux.vnet.ibm.com> > Suggested-by: Pradeep Satyanarayana <pradeeps@linux.vnet.ibm.com> > Signed-off-by: Dany Madden <drt@linux.ibm.com> > Signed-off-by: Lijun Pan <ljp@linux.ibm.com> > --- > drivers/net/ethernet/ibm/ibmvnic.c | 8 +++++++- > 1 file changed, 7 insertions(+), 1 deletion(-) > > diff --git a/drivers/net/ethernet/ibm/ibmvnic.c b/drivers/net/ethernet/ibm/ibmvnic.c > index 1b702a43a5d0..718da39f5ae4 100644 > --- a/drivers/net/ethernet/ibm/ibmvnic.c > +++ b/drivers/net/ethernet/ibm/ibmvnic.c > @@ -2067,8 +2067,11 @@ static int do_reset(struct ibmvnic_adapter *adapter, > for (i = 0; i < adapter->req_rx_queues; i++) > napi_schedule(&adapter->napi[i]); > > - if (adapter->reset_reason != VNIC_RESET_FAILOVER) > + if (adapter->reset_reason == VNIC_RESET_FAILOVER || > + adapter->reset_reason == VNIC_RESET_MOBILITY) { > call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, netdev); > + call_netdevice_notifiers(NETDEV_RESEND_IGMP, netdev); > + } > > rc = 0; > > @@ -2138,6 +2141,9 @@ static int do_hard_reset(struct ibmvnic_adapter *adapter, > if (rc) > return IBMVNIC_OPEN_FAILED; > > + call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, netdev); > + call_netdevice_notifiers(NETDEV_RESEND_IGMP, netdev); > + > return 0; > } >
On 2020-10-30 13:27, Jakub Kicinski wrote: > On Wed, 28 Oct 2020 00:57:41 -0500 Lijun Pan wrote: >> We need to notify peers only when failover and migration happen. >> It is unnecessary to call that in other events like >> FATAL, NON_FATAL, CHANGE_PARAM, and TIMEOUT resets >> since in those scenarios the MAC address and ip address mapping >> does not change. Originally all the resets except CHANGE_PARAM >> are processed by do_reset such that we need to find out >> failover and migration cases in do_reset and call notifier functions. >> We only need to notify peers in do_reset and do_hard_reset. >> We don't need notify peers in do_change_param_reset since it is >> a CHANGE_PARAM reset. In a nested reset case, it will finally >> call into do_hard_reset with reasons other than failvoer and >> migration. So, we don't need to check the reset reason in >> do_hard_reset and just call notifier functions anyway. > > You're completely undoing the commit you linked to: Testing is underway. We will clarify the description in the next version. Thank you for your review and feedback. Dany > > commit 61d3e1d9bc2a1910d773cbf4ed6f587a7a6166b5 > Author: Nathan Fontenot <nfont@linux.vnet.ibm.com> > Date: Mon Jun 12 20:47:45 2017 -0400 > > ibmvnic: Remove netdev notify for failover resets > > When handling a driver reset due to a failover of the backing > server on the vios, doing the netdev_notify_peers() can cause > network traffic to stall or halt. Remove the netdev notify call > for failover resets. > > Signed-off-by: Nathan Fontenot <nfont@linux.vnet.ibm.com> > Signed-off-by: David S. Miller <davem@davemloft.net> > > diff --git a/drivers/net/ethernet/ibm/ibmvnic.c > b/drivers/net/ethernet/ibm/ibmvnic.c > index fd3ef3005fb0..59ea7a5ae776 100644 > --- a/drivers/net/ethernet/ibm/ibmvnic.c > +++ b/drivers/net/ethernet/ibm/ibmvnic.c > @@ -1364,7 +1364,9 @@ static int do_reset(struct ibmvnic_adapter > *adapter, > for (i = 0; i < adapter->req_rx_queues; i++) > napi_schedule(&adapter->napi[i]); > > - netdev_notify_peers(netdev); > + if (adapter->reset_reason != VNIC_RESET_FAILOVER) > + netdev_notify_peers(netdev); > + > return 0; > } > > But you don't seem to address why this change was unnecessary. > > AFAIK you're saying "we only need this event for FAILOVER and MOBILITY" > but the previous commit _excluded_ FAILOVER for some vague reason. > > If the previous commit was incorrect you need to explain that in the > commit message. > >> netdev_notify_peers calls below two functions with rtnl lock(). >> call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); >> call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev); >> When netdev_notify_peers was substituted in >> commit 986103e7920c ("net/ibmvnic: Fix RTNL deadlock during device >> reset"), >> call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev) was missed. > > That should be a separate patch. > >> Fixes: 61d3e1d9bc2a ("ibmvnic: Remove netdev notify for failover >> resets") >> Fixes: 986103e7920c ("net/ibmvnic: Fix RTNL deadlock during device >> reset") > > Please don't line-wrap fixes tags. > >> Suggested-by: Brian King <brking@linux.vnet.ibm.com> >> Suggested-by: Pradeep Satyanarayana <pradeeps@linux.vnet.ibm.com> >> Signed-off-by: Dany Madden <drt@linux.ibm.com> >> Signed-off-by: Lijun Pan <ljp@linux.ibm.com> >> --- >> drivers/net/ethernet/ibm/ibmvnic.c | 8 +++++++- >> 1 file changed, 7 insertions(+), 1 deletion(-) >> >> diff --git a/drivers/net/ethernet/ibm/ibmvnic.c >> b/drivers/net/ethernet/ibm/ibmvnic.c >> index 1b702a43a5d0..718da39f5ae4 100644 >> --- a/drivers/net/ethernet/ibm/ibmvnic.c >> +++ b/drivers/net/ethernet/ibm/ibmvnic.c >> @@ -2067,8 +2067,11 @@ static int do_reset(struct ibmvnic_adapter >> *adapter, >> for (i = 0; i < adapter->req_rx_queues; i++) >> napi_schedule(&adapter->napi[i]); >> >> - if (adapter->reset_reason != VNIC_RESET_FAILOVER) >> + if (adapter->reset_reason == VNIC_RESET_FAILOVER || >> + adapter->reset_reason == VNIC_RESET_MOBILITY) { >> call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, netdev); >> + call_netdevice_notifiers(NETDEV_RESEND_IGMP, netdev); >> + } >> >> rc = 0; >> >> @@ -2138,6 +2141,9 @@ static int do_hard_reset(struct ibmvnic_adapter >> *adapter, >> if (rc) >> return IBMVNIC_OPEN_FAILED; >> >> + call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, netdev); >> + call_netdevice_notifiers(NETDEV_RESEND_IGMP, netdev); >> + >> return 0; >> } >>
diff --git a/drivers/net/ethernet/ibm/ibmvnic.c b/drivers/net/ethernet/ibm/ibmvnic.c index 1b702a43a5d0..718da39f5ae4 100644 --- a/drivers/net/ethernet/ibm/ibmvnic.c +++ b/drivers/net/ethernet/ibm/ibmvnic.c @@ -2067,8 +2067,11 @@ static int do_reset(struct ibmvnic_adapter *adapter, for (i = 0; i < adapter->req_rx_queues; i++) napi_schedule(&adapter->napi[i]); - if (adapter->reset_reason != VNIC_RESET_FAILOVER) + if (adapter->reset_reason == VNIC_RESET_FAILOVER || + adapter->reset_reason == VNIC_RESET_MOBILITY) { call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, netdev); + call_netdevice_notifiers(NETDEV_RESEND_IGMP, netdev); + } rc = 0; @@ -2138,6 +2141,9 @@ static int do_hard_reset(struct ibmvnic_adapter *adapter, if (rc) return IBMVNIC_OPEN_FAILED; + call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, netdev); + call_netdevice_notifiers(NETDEV_RESEND_IGMP, netdev); + return 0; }
We need to notify peers only when failover and migration happen. It is unnecessary to call that in other events like FATAL, NON_FATAL, CHANGE_PARAM, and TIMEOUT resets since in those scenarios the MAC address and ip address mapping does not change. Originally all the resets except CHANGE_PARAM are processed by do_reset such that we need to find out failover and migration cases in do_reset and call notifier functions. We only need to notify peers in do_reset and do_hard_reset. We don't need notify peers in do_change_param_reset since it is a CHANGE_PARAM reset. In a nested reset case, it will finally call into do_hard_reset with reasons other than failvoer and migration. So, we don't need to check the reset reason in do_hard_reset and just call notifier functions anyway. netdev_notify_peers calls below two functions with rtnl lock(). call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, dev); call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev); When netdev_notify_peers was substituted in commit 986103e7920c ("net/ibmvnic: Fix RTNL deadlock during device reset"), call_netdevice_notifiers(NETDEV_RESEND_IGMP, dev) was missed. Fixes: 61d3e1d9bc2a ("ibmvnic: Remove netdev notify for failover resets") Fixes: 986103e7920c ("net/ibmvnic: Fix RTNL deadlock during device reset") Suggested-by: Brian King <brking@linux.vnet.ibm.com> Suggested-by: Pradeep Satyanarayana <pradeeps@linux.vnet.ibm.com> Signed-off-by: Dany Madden <drt@linux.ibm.com> Signed-off-by: Lijun Pan <ljp@linux.ibm.com> --- drivers/net/ethernet/ibm/ibmvnic.c | 8 +++++++- 1 file changed, 7 insertions(+), 1 deletion(-)