From: Simon Marlow <marlowsd@gmail.com>
Date: Tue, 9 Mar 2010 09:58:31 +0000 (+0000)
Subject: Fix a rare deadlock when the IO manager thread is slow to start up
X-Git-Url: http://git.megacz.com/?a=commitdiff_plain;h=f8a01fb39f3756bec409c9b752fd4f8088e130bf;p=ghc-hetmet.git

Fix a rare deadlock when the IO manager thread is slow to start up
This fixes occasional failures of ffi002(threaded1) on a loaded
machine.
---

diff --git a/rts/Capability.c b/rts/Capability.c
index c37ec4e..cf85372 100644
--- a/rts/Capability.c
+++ b/rts/Capability.c
@@ -786,10 +786,18 @@ shutdownCapability (Capability *cap, Task *task, rtsBool safe)
 		       "thread(s) are involved in foreign calls, yielding");
             cap->running_task = NULL;
 	    RELEASE_LOCK(&cap->lock);
+            // The IO manager thread might have been slow to start up,
+            // so the first attempt to kill it might not have
+            // succeeded.  Just in case, try again - the kill message
+            // will only be sent once.
+            //
+            // To reproduce this deadlock: run ffi002(threaded1)
+            // repeatedly on a loaded machine.
+            ioManagerDie();
             yieldThread();
             continue;
         }
-            
+
         traceEventShutdown(cap);
 	RELEASE_LOCK(&cap->lock);
 	break;