From d545103be504fc8ec6a23c4ad5ff08297b6fc1fa Mon Sep 17 00:00:00 2001
From: Dan Crowell <dcrowell@us.ibm.com>
Date: Thu, 29 Mar 2012 10:01:24 -0500
Subject: Fix FSI FFDC collection thread bug

Fixing a possible multi-threading issue in how the FSI driver
collects FFDC.  I am keeping track of the Task that hit the
error to avoid hitting the mutex twice.

Change-Id: Ic48cc046e1a23c67df84802c51307be645a06a8d
RTC: 35287
Reviewed-on: http://gfw160.austin.ibm.com:8080/gerrit/806
Tested-by: Jenkins Server
Reviewed-by: A. Patrick Williams III <iawillia@us.ibm.com>
---
 src/usr/fsi/fsidd.C | 19 +++++++++++--------
 src/usr/fsi/fsidd.H |  6 ++++--
 2 files changed, 15 insertions(+), 10 deletions(-)

(limited to 'src')

diff --git a/src/usr/fsi/fsidd.C b/src/usr/fsi/fsidd.C
index 94423c3da..59d4f8438 100644
--- a/src/usr/fsi/fsidd.C
+++ b/src/usr/fsi/fsidd.C
@@ -626,6 +626,7 @@ errlHndl_t FsiDD::initializeHardware()
  */
 FsiDD::FsiDD()
 :iv_master(NULL)
+,iv_ffdcTask(0)
 {
     TRACFCOMP(g_trac_fsi, "FsiDD::FsiDD()>");
 
@@ -714,10 +715,10 @@ errlHndl_t FsiDD::read(const FsiAddrInfo_t& i_addrInfo,
 
         // atomic section >>
         l_mutex
-            = (i_addrInfo.opbTarg)->getHbMutexAttr<TARGETING::ATTR_FSI_MASTER_MUTEX>();
+          = (i_addrInfo.opbTarg)->getHbMutexAttr<TARGETING::ATTR_FSI_MASTER_MUTEX>();
 
-        // skip the mutex lock if we're already inside a previous operation
-        if( !iv_ffdcCollection )
+        if( (iv_ffdcTask != 0)  // performance hack for typical case
+            && (iv_ffdcTask != task_gettid()) )
         {
             mutex_lock(l_mutex);
             need_unlock = true;
@@ -794,8 +795,8 @@ errlHndl_t FsiDD::write(const FsiAddrInfo_t& i_addrInfo,
         l_mutex
             = (i_addrInfo.opbTarg)->getHbMutexAttr<TARGETING::ATTR_FSI_MASTER_MUTEX>();
 
-        // skip the mutex lock if we're already inside a previous operation
-        if( !iv_ffdcCollection )
+        if( (iv_ffdcTask != 0)  // performance hack for typical case
+            && (iv_ffdcTask != task_gettid()) )
         {
             mutex_lock(l_mutex);
             need_unlock = true;
@@ -870,9 +871,10 @@ errlHndl_t FsiDD::handleOpbErrors(const FsiAddrInfo_t& i_addrInfo,
                                         TWO_UINT32_TO_UINT64(i_opbStatReg,0));
 
         // Collect some FFDC but avoid an infinite loop
-        if( !iv_ffdcCollection )
+        if( iv_ffdcTask != 0 )
         {
-            iv_ffdcCollection = true;
+            iv_ffdcTask = task_gettid();
+
             uint32_t data = 0;
             errlHndl_t l_err2 = NULL;
 
@@ -917,7 +919,8 @@ errlHndl_t FsiDD::handleOpbErrors(const FsiAddrInfo_t& i_addrInfo,
             }
 
             //MAGIC_INSTRUCTION(MAGIC_BREAK);
-            iv_ffdcCollection = false;
+
+            iv_ffdcTask = 0;
         }
 
         l_err->collectTrace("FSI");
diff --git a/src/usr/fsi/fsidd.H b/src/usr/fsi/fsidd.H
index 69060544e..51dad9f54 100644
--- a/src/usr/fsi/fsidd.H
+++ b/src/usr/fsi/fsidd.H
@@ -29,6 +29,7 @@
 #include <errl/errlentry.H>
 #include <usr/devicefw/driverif.H>
 #include <attributestructs.H>
+#include <sys/task.h>
 
 
 /** @file fsidd.H
@@ -245,6 +246,7 @@ class FsiDD
      *
      * @param[in] i_addrInfo  FSI addressing information
      * @param[in] i_opbStatReg  OPB Status bits (OPB_REG_STAT[0:31])
+     * @param[in] i_relFsiAddr  Relative FSI Address that was being accessed
      *
      * @return errlHndl_t  NULL on success
      */
@@ -458,9 +460,9 @@ class FsiDD
     TARGETING::Target* iv_master;
 
     /**
-     * Flag to avoid infinite recursion
+     * Non-zero if a Task is currently collecting FFDC
      */
-    bool iv_ffdcCollection;
+    tid_t iv_ffdcTask;
 
   private:
 
-- 
cgit v1.2.1