Nap instruction sometimes executed with wrong permissions.

    When executing the nap instruction, all thread state can be
    lost.  This was causing r0 to be lost, which contained the
    system-call number for the idle thread to request permission
    to execute nap and as a result 'task-yield' was executed
    instead of 'cpu-nap' after the idle thread took its first
    nap.

    Re-arranged the sreset code that handles nap-wakeup to deal
    with thread-state being lost when entering nap and instead
    using the previously saved task-state from the 'cpu-nap'
    system call.

Change-Id: Id7468a8577c4d7b273b23bc97e7dd040555e7b67
Reviewed-on: http://gfw160.austin.ibm.com:8080/gerrit/1567
Tested-by: Jenkins Server
Reviewed-by: A. Patrick Williams III <iawillia@us.ibm.com>

1 files changed, 1 insertions, 2 deletions

diff --git a/src/sys/prof/idletask.C b/src/sys/prof/idletask.C
index 1c38835ce..8f79b94f7 100644
--- a/src/sys/prof/idletask.C
+++ b/src/sys/prof/idletask.C
@@ -43,8 +43,7 @@ void TaskManager::idleTaskLoop(void* unused)
         setThreadPriorityLow();
 
         // Request privilege escalation for nap.
-        register uint64_t syscall = MISC_CPUNAP;
-        asm volatile("or 3, %0, %0; sc" :: "r" (syscall) : "r3", "cc");
+        asm volatile("li 3, %0; sc" :: "K" (MISC_CPUNAP) : "r3", "cc");
 
         // Execute nap.
         nap();