From da390045aa0c06d91b9b3224e831a488bcb0d274 Mon Sep 17 00:00:00 2001
From: Wez Furlong <wez@netevil.org>
Date: Fri, 19 Feb 2010 15:14:55 +0000
Subject: [PATCH] improve compatibility with PPC platforms.

---
 misc.c       |  3 +++
 sol_compat.h | 55 ++++++++++++++++++++++++++++++++++++++++++++++++++--
 umem.c       |  5 +----
 3 files changed, 57 insertions(+), 6 deletions(-)

diff --git a/misc.c b/misc.c
index 1a2437f..3fe3e94 100644
--- a/misc.c
+++ b/misc.c
@@ -66,6 +66,9 @@
 #define	ERR_SIZE 8192		/* must be a power of 2 */
 
 static mutex_t umem_error_lock = DEFAULTMUTEX;
+#ifdef NEED_64_LOCK
+pthread_mutex_t umem_ppc_64inc_lock = PTHREAD_MUTEX_INITIALIZER;
+#endif
 
 static char umem_error_buffer[ERR_SIZE] = "";
 static uint_t umem_error_begin = 0;
diff --git a/sol_compat.h b/sol_compat.h
index 215296a..b2e3e25 100644
--- a/sol_compat.h
+++ b/sol_compat.h
@@ -59,7 +59,7 @@ static INLINE hrtime_t gethrtime(void) {
 static INLINE thread_t _thr_self(void) {
   return thr_self();
 }
-#if defined(_MACH_PORT_T)
+#if defined(__MACH__)
 #define CPUHINT() (pthread_mach_thread_np(pthread_self()))
 #endif
 # define thr_sigsetmask            pthread_sigmask
@@ -114,6 +114,12 @@ static INLINE int thr_create(void *stack_base,
 # ifdef _WIN32
 #  define ec_atomic_inc(a)		InterlockedIncrement(a)
 #  define ec_atomic_inc64(a)    InterlockedIncrement64(a)
+# elif defined(__MACH__)
+#  include <libkern/OSAtomic.h>
+#  define ec_atomic_inc(x) OSAtomicIncrement32Barrier((int32_t*)x)
+#  if !defined(__ppc__)
+#   define ec_atomic_inc64(x) OSAtomicIncrement64Barrier((int64_t*)x)
+#  endif
 # elif (defined(__i386__) || defined(__x86_64__)) && defined(__GNUC__)
 static INLINE uint_t ec_atomic_cas(uint_t *mem, uint_t with, uint_t cmp)
 {
@@ -124,6 +130,40 @@ static INLINE uint_t ec_atomic_cas(uint_t *mem, uint_t with, uint_t cmp)
         : "memory");
   return prev;
 }
+static INLINE uint64_t ec_atomic_cas64(uint64_t *mem, uint64_t with,
+  uint64_t cmp)
+{
+  uint64_t prev;
+#  if defined(__x86_64__)
+  __asm__ volatile ("lock; cmpxchgq %1, %2"
+    : "=a" (prev)
+    : "r" (with), "m" (*(mem)), "0" (cmp)
+    : "memory");
+#  else
+  __asm__ volatile (
+    "pushl %%ebx;"
+    "mov 4+%1,%%ecx;"
+    "mov %1,%%ebx;"
+    "lock;"
+    "cmpxchg8b (%3);"
+    "popl %%ebx"
+    : "=A" (prev)
+    : "m" (with), "A" (cmp), "r" (mem)
+    : "%ecx", "memory");
+#  endif
+  return prev;
+}
+static INLINE uint64_t ec_atomic_inc64(uint64_t *mem)
+{
+  register uint64_t last;
+  do {
+    last = *mem;
+  } while (ec_atomic_cas64(mem, last+1, last) != last);
+  return ++last;
+}
+#  define ec_atomic_inc64 ec_atomic_inc64
+# else
+#  error no atomic solution for your platform
 # endif
 
 # ifndef ec_atomic_inc
@@ -139,7 +179,18 @@ static INLINE uint_t ec_atomic_inc(uint_t *mem)
 # ifndef ec_atomic_inc64
    /* yeah, it's not great.  It's only used to bump failed allocation
     * counts, so it is not critical right now. */
-#  define ec_atomic_inc64(a)  (*a)++
+extern pthread_mutex_t umem_ppc_64inc_lock;
+static INLINE uint64_t ec_atomic_inc64(uint64_t *val)
+{
+  uint64_t rval;
+  pthread_mutex_lock(&umem_ppc_64inc_lock);
+  rval = *val + 1;
+  *val = rval;
+  pthread_mutex_unlock(&umem_ppc_64inc_lock);
+  return rval;
+}
+#  define ec_atomic_inc64 ec_atomic_inc64
+#  define NEED_64_LOCK 1
 # endif
 
 #endif
diff --git a/umem.c b/umem.c
index 9553417..064db65 100644
--- a/umem.c
+++ b/umem.c
@@ -518,13 +518,10 @@ umem_log_header_t *umem_failure_log;
 umem_log_header_t *umem_slab_log;
 
 extern thread_t _thr_self(void);
-#if defined(__MACH__) || defined(__FreeBSD__)
+#ifndef CPUHINT
 # define CPUHINT()	((int)(_thr_self()))
 #endif
 
-#ifndef CPUHINT
-#define	CPUHINT()		(_thr_self())
-#endif
 
 #define	CPUHINT_MAX()		INT_MAX