Merge pull request #168 from grooverdan/10.1-MDEV-8684-UT_RELAX_CPU_isnt_relaxing

MDEV-8684 ut relax cpu isnt relaxing

config.h.cmake

@@ -225,6 +225,7 @@
 #cmakedefine HAVE_PREAD 1
 #cmakedefine HAVE_PAUSE_INSTRUCTION 1
 #cmakedefine HAVE_FAKE_PAUSE_INSTRUCTION 1
+#cmakedefine HAVE_HMT_PRIORITY_INSTRUCTION 1
 #cmakedefine HAVE_RDTSCLL 1
 #cmakedefine HAVE_READ_REAL_TIME 1
 #cmakedefine HAVE_PTHREAD_ATTR_CREATE 1

configure.cmake

@@ -855,6 +855,17 @@ IF(NOT CMAKE_CROSSCOMPILING AND NOT MSVC)
     }
    " HAVE_FAKE_PAUSE_INSTRUCTION)
   ENDIF()
+  IF (NOT HAVE_PAUSE_INSTRUCTION)
+    CHECK_C_SOURCE_COMPILES("
+    #include <sys/platform/ppc.h>
+    int main()
+    {
+     __ppc_set_ppr_low();
+     __ppc_set_ppr_med();
+     return 0;
+    }
+    " HAVE_HMT_PRIORITY_INSTRUCTION)
+  ENDIF()
 ENDIF()
   
 CHECK_SYMBOL_EXISTS(tcgetattr "termios.h" HAVE_TCGETATTR 1)

storage/innobase/include/ut0ut.h

@@ -88,15 +88,26 @@ struct ut_when_dtor {
    the YieldProcessor macro defined in WinNT.h. It is a CPU architecture-
    independent way by using YieldProcessor. */
 #  define UT_RELAX_CPU() YieldProcessor()
-# elif defined(HAVE_ATOMIC_BUILTINS)
+# elif defined(__powerpc__)
+#include <sys/platform/ppc.h>
 #  define UT_RELAX_CPU() do { \
-     volatile lint	volatile_var; \
-     os_compare_and_swap_lint(&volatile_var, 0, 1); \
+     volatile lint      volatile_var = __ppc_get_timebase(); \
    } while (0)
 # else
 #  define UT_RELAX_CPU() ((void)0) /* avoid warning for an empty statement */
 # endif
 
+#define UT_COMPILER_BARRIER() __asm__ __volatile__ ("":::"memory")
+
+# if defined(HAVE_HMT_PRIORITY_INSTRUCTION)
+#include <sys/platform/ppc.h>
+#  define UT_LOW_PRIORITY_CPU() __ppc_set_ppr_low()
+#  define UT_RESUME_PRIORITY_CPU() __ppc_set_ppr_med()
+# else
+#  define UT_LOW_PRIORITY_CPU() ((void)0)
+#  define UT_RESUME_PRIORITY_CPU() ((void)0)
+# endif
+
 /*********************************************************************//**
 Delays execution for at most max_wait_us microseconds or returns earlier
 if cond becomes true.
@@ -342,7 +353,7 @@ Runs an idle loop on CPU. The argument gives the desired delay
 in microseconds on 100 MHz Pentium + Visual C++.
 @return	dummy value */
 UNIV_INTERN
-ulint
+void
 ut_delay(
 /*=====*/
 	ulint	delay);	/*!< in: delay in microseconds on 100 MHz Pentium */

storage/innobase/ut/ut0ut.cc

@@ -45,9 +45,6 @@ Created 5/11/1994 Heikki Tuuri
 # include <string>
 #endif /* UNIV_HOTBACKUP */
 
-/** A constant to prevent the compiler from optimizing ut_delay() away. */
-UNIV_INTERN ibool	ut_always_false	= FALSE;
-
 #ifdef __WIN__
 /*****************************************************************//**
 NOTE: The Windows epoch starts from 1601/01/01 whereas the Unix
@@ -397,25 +394,21 @@ Runs an idle loop on CPU. The argument gives the desired delay
 in microseconds on 100 MHz Pentium + Visual C++.
 @return	dummy value */
 UNIV_INTERN
-ulint
+void
 ut_delay(
 /*=====*/
 	ulint	delay)	/*!< in: delay in microseconds on 100 MHz Pentium */
 {
-	ulint	i, j;
+	ulint	i;
 
-	j = 0;
+	UT_LOW_PRIORITY_CPU();
 
 	for (i = 0; i < delay * 50; i++) {
-		j += i;
 		UT_RELAX_CPU();
+		UT_COMPILER_BARRIER();
 	}
 
-	if (ut_always_false) {
-		ut_always_false = (ibool) j;
-	}
-
-	return(j);
+	UT_RESUME_PRIORITY_CPU();
 }
 #endif /* !UNIV_HOTBACKUP */
 

storage/xtradb/include/ut0ut.h

@@ -80,20 +80,31 @@ struct ut_when_dtor {
 
 # elif defined(HAVE_FAKE_PAUSE_INSTRUCTION)
 #  define UT_RELAX_CPU() __asm__ __volatile__ ("rep; nop")
-# elif defined(HAVE_ATOMIC_BUILTINS)
-#  define UT_RELAX_CPU() do { \
-     volatile lint	volatile_var; \
-     os_compare_and_swap_lint(&volatile_var, 0, 1); \
-   } while (0)
 # elif defined(HAVE_WINDOWS_ATOMICS)
    /* In the Win32 API, the x86 PAUSE instruction is executed by calling
    the YieldProcessor macro defined in WinNT.h. It is a CPU architecture-
    independent way by using YieldProcessor. */
 #  define UT_RELAX_CPU() YieldProcessor()
+# elif defined(__powerpc__)
+#include <sys/platform/ppc.h>
+#  define UT_RELAX_CPU() do { \
+     volatile lint      volatile_var = __ppc_get_timebase(); \
+   } while (0)
 # else
 #  define UT_RELAX_CPU() ((void)0) /* avoid warning for an empty statement */
 # endif
 
+#define UT_COMPILER_BARRIER() __asm__ __volatile__ ("":::"memory")
+
+# if defined(HAVE_HMT_PRIORITY_INSTRUCTION)
+#include <sys/platform/ppc.h>
+#  define UT_LOW_PRIORITY_CPU() __ppc_set_ppr_low()
+#  define UT_RESUME_PRIORITY_CPU() __ppc_set_ppr_med()
+# else
+#  define UT_LOW_PRIORITY_CPU() ((void)0)
+#  define UT_RESUME_PRIORITY_CPU() ((void)0)
+# endif
+
 /*********************************************************************//**
 Delays execution for at most max_wait_us microseconds or returns earlier
 if cond becomes true.
@@ -334,7 +345,7 @@ Runs an idle loop on CPU. The argument gives the desired delay
 in microseconds on 100 MHz Pentium + Visual C++.
 @return	dummy value */
 UNIV_INTERN
-ulint
+void
 ut_delay(
 /*=====*/
 	ulint	delay);	/*!< in: delay in microseconds on 100 MHz Pentium */

storage/xtradb/ut/ut0ut.cc

@@ -46,9 +46,6 @@ Created 5/11/1994 Heikki Tuuri
 # include <string>
 #endif /* UNIV_HOTBACKUP */
 
-/** A constant to prevent the compiler from optimizing ut_delay() away. */
-UNIV_INTERN ibool	ut_always_false	= FALSE;
-
 #ifdef __WIN__
 /*****************************************************************//**
 NOTE: The Windows epoch starts from 1601/01/01 whereas the Unix
@@ -398,25 +395,21 @@ Runs an idle loop on CPU. The argument gives the desired delay
 in microseconds on 100 MHz Pentium + Visual C++.
 @return	dummy value */
 UNIV_INTERN
-ulint
+void
 ut_delay(
 /*=====*/
 	ulint	delay)	/*!< in: delay in microseconds on 100 MHz Pentium */
 {
-	ulint	i, j;
+	ulint	i;
 
-	j = 0;
+	UT_LOW_PRIORITY_CPU();
 
 	for (i = 0; i < delay * 50; i++) {
-		j += i;
 		UT_RELAX_CPU();
+		UT_COMPILER_BARRIER();
 	}
 
-	if (ut_always_false) {
-		ut_always_false = (ibool) j;
-	}
-
-	return(j);
+	UT_RESUME_PRIORITY_CPU();
 }
 #endif /* !UNIV_HOTBACKUP */