Re: [Libunwind-devel] Updated fast trace patch with initial performance results

[Arun Sharma]

On Tue, Apr 5, 2011 at 3:13 PM, Lassi Tuura <address@hidden> wrote:

> How far do we want to go in attempting to avoid the one calloc()? :-)
> Choices seem to be:
>  a. Use __thread, require per-thread wrapper callbacks from app
>  b. Use lock-free global cache stack, must still free 'unused' caches.
>  c. Use pthread_getspecific, deal with calloc from pthread_key_create,
>    maybe require app to call some init function once at 'safe' time if
>    it uses unw_backtrace?

It just occurred to me that we can use both __thread and pthread_setspecific()

--- a/src/x86_64/Gtrace.c
+++ b/src/x86_64/Gtrace.c
@@ -161,6 +161,8 @@ trace_cache_expand (unw_trace_cache_t *cache)
   return 0;
 }

+static __thread  unw_trace_cache_t *tls_cache;
+
 /* Get the frame cache for the current thread. Create it if there is none. */
 static unw_trace_cache_t *
 trace_cache_get (void)
@@ -169,10 +171,11 @@ trace_cache_get (void)
   if (pthread_once)
   {
     pthread_once(&trace_cache_once, &trace_cache_init_once);
-    if (! (cache = pthread_getspecific(trace_cache_key)))
+    if (! (cache = tls_cache))
     {
       cache = trace_cache_create();
       pthread_setspecific(trace_cache_key, cache);
+      tls_cache = cache;
     }
     Debug(5, "using cache %p\n", cache);
     return cache;

Hopefully, this will be faster than pthread_getspecific() and may
allow us to defer the call to pthread_setspecific() to a safer point.

 -Arun