Hello Roland,
We have found several panics and memory leak for IPoIB-CM. I reviewed
the code and caught some issues. I will start to submit a list of patches
for review. i have tested the combined patch for connectX, it passed 24
hours 16 netperf/netserver bidirctional TCP, UDP stream test for 2.7.26-rc6
kernel.
I turned on debug, I found that same QP context being destoryed twice
for nonSRQ connection. I reviewed the code and found that there is a window
the list could be added after the reap call, so checking the QP context
status is needed.
My linux email client is broken, so I attach each patch here.
Address a possible race
-------------------
Signed-off-by: Shirley Ma <[EMAIL PROTECTED]>
drivers/infiniband/ulp/ipoib/ipoib_cm.c | 9 ++++++---
1 files changed, 6 insertions(+), 3 deletions(-)
diff --git a/drivers/infiniband/ulp/ipoib/ipoib_cm.c
b/drivers/infiniband/ulp/ipoib/ipoib_cm.c
index 97e67d3..0886ee7 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib_cm.c
+++ b/drivers/infiniband/ulp/ipoib/ipoib_cm.c
@@ -559,9 +559,12 @@ void ipoib_cm_handle_rx_wc(struct net_device *dev, struct
ib_wc *wc)
else {
if (!--p->recv_count) {
spin_lock_irqsave(&priv->lock, flags);
- list_move(&p->list, &priv->cm.rx_reap_list);
- spin_unlock_irqrestore(&priv->lock, flags);
- queue_work(ipoib_workqueue, &priv->cm.rx_reap_task);
+ if (p->state == IPOIB_CM_RX_LIVE) {
+ list_move(&p->list, &priv->cm.rx_reap_list);
+ spin_unlock_irqrestore(&priv->lock, flags);
+ queue_work(ipoib_workqueue,
&priv->cm.rx_reap_task);
+ } else
+ spin_unlock_irqrestore(&priv->lock, flags);
}
return;
}
(See attached file: nonsrq_last_recv_race.patch)
Thanks
Shirley
nonsrq_last_recv_race.patch
Description: Binary data
_______________________________________________ general mailing list [email protected] http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
