xentrace: Skip to low cpu when throwing away portions of the circular buffer

Skip to the next "low" cpu when throwing away portions of the circular
memory buffer.  This makes subsequent analysis easier.

Signed-off-by: George Dunlap <george.dunlap@eu.citrix.com>

1 files changed, 15 insertions, 0 deletions

diff --git a/tools/xentrace/xentrace.c b/tools/xentrace/xentrace.c
index 18283f3e97..cce5469c47 100644
--- a/tools/xentrace/xentrace.c
+++ b/tools/xentrace/xentrace.c
@@ -149,6 +149,7 @@ void membuf_reserve_window(unsigned cpu, unsigned long window_size)
 {
     struct cpu_change_record *rec;
     long need_to_consume, free, freed;
+    int last_cpu = -1;
 
     if ( membuf.pending_size > 0 )
     {
@@ -193,11 +194,25 @@ void membuf_reserve_window(unsigned cpu, unsigned long window_size)
 
         if ( need_to_consume > 0 )
         {
+            last_cpu = rec->data.cpu;
             MEMBUF_CONS_INCREMENT(freed);
             need_to_consume -= freed;
         }
     } while( need_to_consume > 0 );
 
+    /* For good tsc consistency, we need to start at a low-cpu buffer.  Keep
+     * skipping until the cpu goes down or stays the same. */
+    rec = (struct cpu_change_record *)MEMBUF_POINTER(membuf.cons);
+    while ( rec->data.cpu > last_cpu )
+    {
+        last_cpu = rec->data.cpu; 
+
+        freed = sizeof(*rec) + rec->data.window_size;
+        
+        MEMBUF_CONS_INCREMENT(freed);
+        rec = (struct cpu_change_record *)MEMBUF_POINTER(membuf.cons);
+    }
+
 start_window:
     /*
      * Start writing "pending" data.  Update prod once all this data is