Don't hold the pinger_sem in ptlrpc_pinger_sending_on_import.

It's an inherently racy operation in the first place, and until the pinger
thread is fixed not to hold the pinger_sem while it's waiting for an upcall
to complete, it prevents ptlrpcd from needlessly blocking.

diff --git a/lustre/ChangeLog b/lustre/ChangeLog
index e37117f..000b5c6 100644 (file)
--- a/lustre/ChangeLog
+++ b/lustre/ChangeLog
@@ -39,6 +39,7 @@ tbd         Cluster File Systems, Inc. <info@clusterfs.com>
        - allow readahead allocations to fail when low on memory (5383)
        - mmap locking landed again, after considerable improvement (2828)
        - fix lustre/lustrefs init scripts for SuSE (patch from Scali, 5702)
+       - don't hold the pinger_sem in ptlrpc_pinger_sending_on_import
 
 2004-11-23  Cluster File Systems, Inc. <info@clusterfs.com>
        * version 1.4.0

diff --git a/lustre/ptlrpc/pinger.c b/lustre/ptlrpc/pinger.c
index 88f9694..299fc34 100644 (file)
--- a/lustre/ptlrpc/pinger.c
+++ b/lustre/ptlrpc/pinger.c
@@ -235,9 +235,7 @@ int ptlrpc_stop_pinger(void)
 
 void ptlrpc_pinger_sending_on_import(struct obd_import *imp)
 {
-        down(&pinger_sem);
         imp->imp_next_ping = jiffies + (obd_timeout * HZ);
-        up(&pinger_sem);
 }
 
 int ptlrpc_pinger_add_import(struct obd_import *imp)