Fix deadlock in ECC error shutdown path

There is a scenario wherein if a PNOR ECC UE occurs when attempting to service a page fault, the kernel will deadlock when it attempts to kill the task that triggered the fault. This is due to the kill routine being called while holding the VMM spinlock and then collecting a backtrace, which also performs a nested acquire of the same spinlock. This fix inhibits invoking the backtrace if the kill routine holds the VMM spinlock. Change-Id: I75bf1f248740a08fd485379d88e146096edf65a9 CQ: SW461429 Reviewed-on: http://rchgit01.rchland.ibm.com/gerrit1/75264 Tested-by: Jenkins Server <pfd-jenkins+hostboot@us.ibm.com> Tested-by: Jenkins OP Build CI <op-jenkins+hostboot@us.ibm.com> Tested-by: Jenkins OP HW <op-hw-jenkins+hostboot@us.ibm.com> Tested-by: FSP CI Jenkins <fsp-CI-jenkins+hostboot@us.ibm.com> Reviewed-by: Ilya Smirnov <ismirno@us.ibm.com> Reviewed-by: Michael Baiocchi <mbaiocch@us.ibm.com> Reviewed-by: Matthew Raybuck <matthew.raybuck@ibm.com> Reviewed-by: William G. Hoffa <wghoffa@us.ibm.com>
open-power · Apr 1, 2019 · de2312a · de2312a
1 parent f55c462
commit de2312a
Show file tree

Hide file tree

Showing 2 changed files with 9 additions and 3 deletions.
diff --git a/src/include/kernel/vmmmgr.H b/src/include/kernel/vmmmgr.H
@@ -5,7 +5,7 @@
 /*                                                                        */
 /* OpenPOWER HostBoot Project                                             */
 /*                                                                        */
-/* Contributors Listed Below - COPYRIGHT 2010,2018                        */
+/* Contributors Listed Below - COPYRIGHT 2010,2019                        */
 /* [+] International Business Machines Corp.                              */
 /*                                                                        */
 /*                                                                        */
@@ -298,6 +298,7 @@ class VmmManager
     public:
         friend class Block;
         friend class StackSegment;
+        friend class MessageHandler;
 
 };
 

diff --git a/src/kernel/msghandler.C b/src/kernel/msghandler.C
@@ -33,7 +33,7 @@
 #include <kernel/console.H>
 #include <kernel/doorbell.H>
 #include <kernel/misc.H>
-
+#include <kernel/vmmmgr.H>
 
 void MessageHandler::sendMessage(msg_sys_types_t i_type, void* i_key,
                                  void* i_data, task_t* i_task)
@@ -186,7 +186,12 @@ int MessageHandler::recvMessage(msg_t* i_msg)
             printk("Unhandled msg rc %d (%s) for key %p on task %d @ %p\n",
                     msg_rc, ErrnoToString(msg_rc), key, deferred_task->tid,
                     deferred_task->context.nip);
-            KernelMisc::printkBacktrace(deferred_task);
+            // Kernel will deadlock if the message handler has the VMM spinlock
+            // locked and then attempts to print the backtrace
+            if(VmmManager::getLock() != iv_lock)
+            {
+                KernelMisc::printkBacktrace(deferred_task);
+            }
             MAGIC_INSTRUCTION(MAGIC_BREAK_ON_ERROR);
             endTaskList.insert(deferred_task);
         }