Bring back support for UNW_CACHE_PER_THREAD.

Needs to be build with --enable-per-thread-cache. Default caching policy
is also UNW_CACHE_PER_THREAD than.

configure.ac

@@ -298,7 +298,16 @@ fi
 AC_SUBST([LIBLZMA])
 AM_CONDITIONAL(HAVE_LZMA, test x$enable_minidebuginfo = xyes)
 
+AC_MSG_CHECKING([whether to support UNW_CACHE_PER_THREAD])
+AC_ARG_ENABLE([per-thread-cache],
+AS_HELP_STRING([--enable-per-thread-cache], [build with support for UNW_CACHE_PER_THREAD (which imposes a hight TLS memory usage) (default: disabled)]))
+AC_MSG_RESULT([$enable_per_thread_cache])
+AS_IF([test x$enable_per_thread_cache = xyes], [
   LIBUNWIND___THREAD
+  AS_IF([test x$libc_cv_gcc___thread = xno], [
+    AC_MSG_FAILURE([UNW_CACHE_PER_THREAD requires __thread])
+  ])
+])
 
 AC_MSG_CHECKING([for Intel compiler])
 AC_COMPILE_IFELSE([AC_LANG_PROGRAM([[]], [[#ifndef __INTEL_COMPILER

include/dwarf.h

@@ -37,13 +37,6 @@ struct elf_dyn_info;
 # include "config.h"
 #endif
 
-#ifdef HAVE___THREAD
-  /* For now, turn off per-thread caching.  It uses up too much TLS
-     memory per thread even when the thread never uses libunwind at
-     all.  */
-# undef HAVE___THREAD
-#endif
-
 #ifndef UNW_REMOTE_ONLY
   #if defined(HAVE_LINK_H)
     #include <link.h>

include/libunwind_i.h

@@ -37,11 +37,10 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.  */
 
 #include "compiler.h"
 
-#ifdef HAVE___THREAD
+#if defined(HAVE___THREAD) && HAVE___THREAD
-  /* For now, turn off per-thread caching.  It uses up too much TLS
+#define UNWI_DEFAULT_CACHING_POLICY UNW_CACHE_PER_THREAD
-     memory per thread even when the thread never uses libunwind at
+#else
-     all.  */
+#define UNWI_DEFAULT_CACHING_POLICY UNW_CACHE_GLOBAL
-# undef HAVE___THREAD
 #endif
 
 /* Platform-independent libunwind-internal declarations.  */

include/tdep-tilegx/libunwind_i.h

@@ -36,10 +36,6 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.  */
 #include "mempool.h"
 #include "dwarf.h"
 
-#ifdef HAVE___THREAD
-# undef HAVE___THREAD
-#endif
-
 typedef struct
 {
   /* no Tilegx-specific fast trace */

src/aarch64/Ginit.c

@@ -172,7 +172,7 @@ HIDDEN void
 aarch64_local_addr_space_init (void)
 {
   memset (&local_addr_space, 0, sizeof (local_addr_space));
-  local_addr_space.caching_policy = UNW_CACHE_GLOBAL;
+  local_addr_space.caching_policy = UNWI_DEFAULT_CACHING_POLICY;
   local_addr_space.acc.find_proc_info = dwarf_find_proc_info;
   local_addr_space.acc.put_unwind_info = put_unwind_info;
   local_addr_space.acc.get_dyn_info_list_addr = get_dyn_info_list_addr;

src/arm/Ginit.c

@@ -220,7 +220,7 @@ HIDDEN void
 arm_local_addr_space_init (void)
 {
   memset (&local_addr_space, 0, sizeof (local_addr_space));
-  local_addr_space.caching_policy = UNW_CACHE_GLOBAL;
+  local_addr_space.caching_policy = UNWI_DEFAULT_CACHING_POLICY;
   local_addr_space.acc.find_proc_info = arm_find_proc_info;
   local_addr_space.acc.put_unwind_info = arm_put_unwind_info;
   local_addr_space.acc.get_dyn_info_list_addr = get_dyn_info_list_addr;

src/dwarf/Gparser.c

@@ -606,7 +606,17 @@ get_rs_cache (unw_addr_space_t as, intrmask_t *saved_maskp)
   if (caching == UNW_CACHE_NONE)
     return NULL;
 
+#if defined(HAVE___THREAD) && HAVE___THREAD
+  if (likely (caching == UNW_CACHE_PER_THREAD))
+    {
+      static __thread struct dwarf_rs_cache tls_cache __attribute__((tls_model("initial-exec")));
+      Debug (16, "using TLS cache\n");
+      cache = &tls_cache;
+    }
+  else
+#else
   if (likely (caching == UNW_CACHE_GLOBAL))
+#endif
     {
       Debug (16, "acquiring lock\n");
       lock_acquire (&cache->lock, *saved_maskp);
@@ -615,6 +625,8 @@ get_rs_cache (unw_addr_space_t as, intrmask_t *saved_maskp)
   if ((atomic_read (&as->cache_generation) != atomic_read (&cache->generation))
        || !cache->hash)
     {
+      /* cache_size is only set in the global_cache, copy it over before flushing */
+      cache->log_size = as->global_cache.log_size;
       if (dwarf_flush_rs_cache (cache) < 0)
         return NULL;
       cache->generation = as->cache_generation;

src/hppa/Ginit.c

@@ -179,7 +179,7 @@ HIDDEN void
 hppa_local_addr_space_init (void)
 {
   memset (&local_addr_space, 0, sizeof (local_addr_space));
-  local_addr_space.caching_policy = UNW_CACHE_GLOBAL;
+  local_addr_space.caching_policy = UNWI_DEFAULT_CACHING_POLICY;
   local_addr_space.acc.find_proc_info = dwarf_find_proc_info;
   local_addr_space.acc.put_unwind_info = put_unwind_info;
   local_addr_space.acc.get_dyn_info_list_addr = get_dyn_info_list_addr;

src/ia64/Ginit.c

@@ -361,7 +361,7 @@ ia64_local_addr_space_init (void)
 #elif defined(__hpux)
   local_addr_space.abi = ABI_HPUX;
 #endif
-  local_addr_space.caching_policy = UNW_CACHE_GLOBAL;
+  local_addr_space.caching_policy = UNWI_DEFAULT_CACHING_POLICY;
   local_addr_space.acc.find_proc_info = tdep_find_proc_info;
   local_addr_space.acc.put_unwind_info = put_unwind_info;
   local_addr_space.acc.get_dyn_info_list_addr = get_dyn_info_list_addr;

src/ia64/Gscript.c

@@ -45,7 +45,7 @@ enum ia64_script_insn_opcode
     IA64_INSN_MOVE_SCRATCH_NO_NAT /* like above, but clear NaT info */
   };
 
-#ifdef HAVE___THREAD
+#if defined(HAVE___THREAD) && HAVE___THREAD
 static __thread struct ia64_script_cache ia64_per_thread_cache =
   {
 #ifdef HAVE_ATOMIC_OPS_H
@@ -105,7 +105,7 @@ get_script_cache (unw_addr_space_t as, intrmask_t *saved_maskp)
   if (!spin_trylock_irqsave (&cache->busy, *saved_maskp))
     return NULL;
 #else
-# ifdef HAVE___THREAD
+# if defined(HAVE___THREAD) && HAVE___THREAD
   if (as->caching_policy == UNW_CACHE_PER_THREAD)
     cache = &ia64_per_thread_cache;
 # endif

src/mi/Gset_cache_size.c

@@ -38,6 +38,12 @@ unw_set_cache_size (unw_addr_space_t as, size_t size, int flag)
   if (flag != 0)
     return -1;
 
+  /* Currently not supported for per-thread cache due to memory leak */
+  /* A pthread-key destructor would work, but is not signal safe */
+#if defined(HAVE___THREAD) && HAVE___THREAD
+  return -1;
+#endif
+
   /* Round up to next power of two, slowly but portably */
   while(power < size)
     {

src/mi/Gset_caching_policy.c

@@ -31,7 +31,7 @@ unw_set_caching_policy (unw_addr_space_t as, unw_caching_policy_t policy)
   if (!tdep_init_done)
     tdep_init ();
 
-#ifndef HAVE___THREAD
+#if !(defined(HAVE___THREAD) && HAVE___THREAD)
   if (policy == UNW_CACHE_PER_THREAD)
     policy = UNW_CACHE_GLOBAL;
 #endif

src/mips/Ginit.c

@@ -195,7 +195,7 @@ mips_local_addr_space_init (void)
 # error Unsupported ABI
 #endif
   local_addr_space.addr_size = sizeof (void *);
-  local_addr_space.caching_policy = UNW_CACHE_GLOBAL;
+  local_addr_space.caching_policy = UNWI_DEFAULT_CACHING_POLICY;
   local_addr_space.acc.find_proc_info = dwarf_find_proc_info;
   local_addr_space.acc.put_unwind_info = put_unwind_info;
   local_addr_space.acc.get_dyn_info_list_addr = get_dyn_info_list_addr;

src/ppc32/Ginit.c

@@ -201,7 +201,7 @@ HIDDEN void
 ppc32_local_addr_space_init (void)
 {
   memset (&local_addr_space, 0, sizeof (local_addr_space));
-  local_addr_space.caching_policy = UNW_CACHE_GLOBAL;
+  local_addr_space.caching_policy = UNWI_DEFAULT_CACHING_POLICY;
   local_addr_space.acc.find_proc_info = dwarf_find_proc_info;
   local_addr_space.acc.put_unwind_info = put_unwind_info;
   local_addr_space.acc.get_dyn_info_list_addr = get_dyn_info_list_addr;

src/ppc64/Ginit.c

@@ -214,7 +214,7 @@ ppc64_local_addr_space_init (void)
 #else
   local_addr_space.abi = UNW_PPC64_ABI_ELFv1;
 #endif
-  local_addr_space.caching_policy = UNW_CACHE_GLOBAL;
+  local_addr_space.caching_policy = UNWI_DEFAULT_CACHING_POLICY;
   local_addr_space.acc.find_proc_info = dwarf_find_proc_info;
   local_addr_space.acc.put_unwind_info = put_unwind_info;
   local_addr_space.acc.get_dyn_info_list_addr = get_dyn_info_list_addr;

src/sh/Ginit.c

@@ -171,7 +171,7 @@ HIDDEN void
 sh_local_addr_space_init (void)
 {
   memset (&local_addr_space, 0, sizeof (local_addr_space));
-  local_addr_space.caching_policy = UNW_CACHE_GLOBAL;
+  local_addr_space.caching_policy = UNWI_DEFAULT_CACHING_POLICY;
   local_addr_space.acc.find_proc_info = dwarf_find_proc_info;
   local_addr_space.acc.put_unwind_info = put_unwind_info;
   local_addr_space.acc.get_dyn_info_list_addr = get_dyn_info_list_addr;

src/tilegx/Ginit.c

@@ -152,7 +152,7 @@ tilegx_local_addr_space_init (void)
 
   local_addr_space.abi = UNW_TILEGX_ABI_N64;
   local_addr_space.addr_size = sizeof (void *);
-  local_addr_space.caching_policy = UNW_CACHE_GLOBAL;
+  local_addr_space.caching_policy = UNWI_DEFAULT_CACHING_POLICY;
   local_addr_space.acc.find_proc_info = dwarf_find_proc_info;
   local_addr_space.acc.put_unwind_info = put_unwind_info;
   local_addr_space.acc.get_dyn_info_list_addr = get_dyn_info_list_addr;

src/x86/Ginit.c

@@ -228,7 +228,7 @@ HIDDEN void
 x86_local_addr_space_init (void)
 {
   memset (&local_addr_space, 0, sizeof (local_addr_space));
-  local_addr_space.caching_policy = UNW_CACHE_GLOBAL;
+  local_addr_space.caching_policy = UNWI_DEFAULT_CACHING_POLICY;
   local_addr_space.acc.find_proc_info = dwarf_find_proc_info;
   local_addr_space.acc.put_unwind_info = put_unwind_info;
   local_addr_space.acc.get_dyn_info_list_addr = get_dyn_info_list_addr;

src/x86_64/Ginit.c

@@ -273,7 +273,7 @@ HIDDEN void
 x86_64_local_addr_space_init (void)
 {
   memset (&local_addr_space, 0, sizeof (local_addr_space));
-  local_addr_space.caching_policy = UNW_CACHE_GLOBAL;
+  local_addr_space.caching_policy = UNWI_DEFAULT_CACHING_POLICY;
   local_addr_space.acc.find_proc_info = dwarf_find_proc_info;
   local_addr_space.acc.put_unwind_info = put_unwind_info;
   local_addr_space.acc.get_dyn_info_list_addr = get_dyn_info_list_addr;

tests/Gperf-simple.c

@@ -249,6 +249,8 @@ main (int argc, char **argv)
 
   measure_init ();
 
+  doit ("default         ");
+
   unw_set_caching_policy (unw_local_addr_space, UNW_CACHE_NONE);
   doit ("no cache        ");
 

tests/Gperf-trace.c

@@ -235,6 +235,8 @@ main (int argc, char **argv)
 
   measure_init ();
 
+  doit ("default         ");
+
   unw_set_caching_policy (unw_local_addr_space, UNW_CACHE_NONE);
   doit ("no cache        ");