[PATCH 01/02] RDMA/nes: atomic counters for cm listener create and destroy

[Faisal Latif]

Running long hour iterative MPI tests, sometimes ethtool statistics
CM Destroy Listener count is more than CM Create Listener.
This inconsistency is fixed by making counter variable atomic.

Signed-off-by: Faisal Latif faisal.la...@intel.com
---
 drivers/infiniband/hw/nes/nes.h |4 ++--
 drivers/infiniband/hw/nes/nes_cm.c  |8 
 drivers/infiniband/hw/nes/nes_nic.c |4 ++--
 3 files changed, 8 insertions(+), 8 deletions(-)

diff --git a/drivers/infiniband/hw/nes/nes.h b/drivers/infiniband/hw/nes/nes.h
index 9884056..cffdac4 100644
--- a/drivers/infiniband/hw/nes/nes.h
+++ b/drivers/infiniband/hw/nes/nes.h
@@ -193,8 +193,8 @@ extern u32 cm_packets_created;
 extern u32 cm_packets_received;
 extern u32 cm_packets_dropped;
 extern u32 cm_packets_retrans;
-extern u32 cm_listens_created;
-extern u32 cm_listens_destroyed;
+extern atomic_t cm_listens_created;
+extern atomic_t cm_listens_destroyed;
 extern u32 cm_backlog_drops;
 extern atomic_t cm_loopbacks;
 extern atomic_t cm_nodes_created;
diff --git a/drivers/infiniband/hw/nes/nes_cm.c 
b/drivers/infiniband/hw/nes/nes_cm.c
index 39468c2..debd92c 100644
--- a/drivers/infiniband/hw/nes/nes_cm.c
+++ b/drivers/infiniband/hw/nes/nes_cm.c
@@ -67,8 +67,8 @@ u32 cm_packets_dropped;
 u32 cm_packets_retrans;
 u32 cm_packets_created;
 u32 cm_packets_received;
-u32 cm_listens_created;
-u32 cm_listens_destroyed;
+atomic_t cm_listens_created;
+atomic_t cm_listens_destroyed;
 u32 cm_backlog_drops;
 atomic_t cm_loopbacks;
 atomic_t cm_nodes_created;
@@ -1042,7 +1042,7 @@ static int mini_cm_dec_refcnt_listen(struct nes_cm_core 
*cm_core,
kfree(listener);
listener = NULL;
ret = 0;
-   cm_listens_destroyed++;
+   atomic_inc(cm_listens_destroyed);
} else {
spin_unlock_irqrestore(cm_core-listen_list_lock, flags);
}
@@ -3172,7 +3172,7 @@ int nes_create_listen(struct iw_cm_id *cm_id, int backlog)
g_cm_core-api-stop_listener(g_cm_core, (void 
*)cm_node);
return err;
}
-   cm_listens_created++;
+   atomic_inc(cm_listens_created);
}
 
cm_id-add_ref(cm_id);
diff --git a/drivers/infiniband/hw/nes/nes_nic.c 
b/drivers/infiniband/hw/nes/nes_nic.c
index ab11027..3d550dc 100644
--- a/drivers/infiniband/hw/nes/nes_nic.c
+++ b/drivers/infiniband/hw/nes/nes_nic.c
@@ -1230,8 +1230,8 @@ static void nes_netdev_get_ethtool_stats(struct 
net_device *netdev,
target_stat_values[++index] = cm_packets_received;
target_stat_values[++index] = cm_packets_dropped;
target_stat_values[++index] = cm_packets_retrans;
-   target_stat_values[++index] = cm_listens_created;
-   target_stat_values[++index] = cm_listens_destroyed;
+   target_stat_values[++index] = atomic_read(cm_listens_created);
+   target_stat_values[++index] = atomic_read(cm_listens_destroyed);
target_stat_values[++index] = cm_backlog_drops;
target_stat_values[++index] = atomic_read(cm_loopbacks);
target_stat_values[++index] = atomic_read(cm_nodes_created);
-- 
1.5.3.3

--
To unsubscribe from this list: send the line unsubscribe linux-rdma in
the body of a message to majord...@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

[PATCH 02/02] RDMA/nes: listener destroyed during loopback setup crash

[Faisal Latif]

When listener is destroyed and where is MPA response pending for
loopback connection, the active side cm_node gets destroyed in
cm_event_connect_error() and again in nes_accept()/nes_reject().
Incrementing cm_node's refcount to not be destroyed by cm_event_connect_error().

Signed-off-by: Faisal Latif faisal.la...@intel.com
---
 drivers/infiniband/hw/nes/nes_cm.c |3 ++-
 1 files changed, 2 insertions(+), 1 deletions(-)

diff --git a/drivers/infiniband/hw/nes/nes_cm.c 
b/drivers/infiniband/hw/nes/nes_cm.c
index debd92c..2a49ee4 100644
--- a/drivers/infiniband/hw/nes/nes_cm.c
+++ b/drivers/infiniband/hw/nes/nes_cm.c
@@ -1011,9 +1011,10 @@ static int mini_cm_dec_refcnt_listen(struct nes_cm_core 
*cm_core,
event.cm_info.loc_port =
 loopback-loc_port;
event.cm_info.cm_id = loopback-cm_id;
+   add_ref_cm_node(loopback);
+   loopback-state = NES_CM_STATE_CLOSED;
cm_event_connect_error(event);
cm_node-state = 
NES_CM_STATE_LISTENER_DESTROYED;
-   loopback-state = NES_CM_STATE_CLOSED;
 
rem_ref_cm_node(cm_node-cm_core,
 cm_node);
-- 
1.5.3.3

--
To unsubscribe from this list: send the line unsubscribe linux-rdma in
the body of a message to majord...@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

RE: [PATCH 02/02] RDMA/nes: listener destroyed during loopback setup crash

[Sean Hefty]

@@ -1011,9 +1011,10 @@ static int mini_cm_dec_refcnt_listen(struct nes_cm_core
*cm_core,
   event.cm_info.loc_port =
loopback-loc_port;
   event.cm_info.cm_id = loopback-cm_id;
+  add_ref_cm_node(loopback);
+  loopback-state = NES_CM_STATE_CLOSED;

It seems odd to increment a reference count in a function with the name
'dec_refcnt'.  Is there a better name for mini_cm_dec_refcnt_listen?

--
To unsubscribe from this list: send the line unsubscribe linux-rdma in
the body of a message to majord...@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH] mlx4_ib: simplify retrieval of ib_device

[Roland Dreier]

thanks, applied
-- 
Roland Dreier  rola...@cisco.com
For corporate legal information go to:
http://www.cisco.com/web/about/doing_business/legal/cri/index.html
--
To unsubscribe from this list: send the line unsubscribe linux-rdma in
the body of a message to majord...@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH 2/2] ib/ehca: allow access for ib_query_qp()

[Roland Dreier]

applied 1  2, thanks
-- 
Roland Dreier  rola...@cisco.com
For corporate legal information go to:
http://www.cisco.com/web/about/doing_business/legal/cri/index.html
--
To unsubscribe from this list: send the line unsubscribe linux-rdma in
the body of a message to majord...@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html