From d4ba50477e6f0092a3b83528c02c0f30a2b708a3 Mon Sep 17 00:00:00 2001
From: Yasufumi Kinoshita <yasufumi.kinoshita@oracle.com>
Date: Wed, 30 Mar 2016 14:32:20 +1100
Subject: [PATCH] Some POWER specific optimizations

Bug#18842925 : SET THREAD PRIORITY IN INNODB MUTEX SPINLOOP
Like "pause" instruction for hyper-threading at Intel CPUs,
POWER has special instructions only for hinting priority of hardware-threads.

Approved by Sunny in rb#6256

Backport of the 5.7 fix - https://github.com/mysql/mysql-server/commit/c92102a6ef0f280bfb56e5585fca0d0cdcc34890
(excluded cache line size patch)

Suggestion by Stewart Smith
---
 config.h.cmake                   |  1 +
 configure.cmake                  | 10 ++++++++++
 storage/innobase/include/ut0ut.h |  8 ++++++++
 storage/innobase/ut/ut0ut.cc     |  4 ++++
 storage/xtradb/include/ut0ut.h   |  8 ++++++++
 storage/xtradb/ut/ut0ut.cc       |  4 ++++
 6 files changed, 35 insertions(+)

diff --git a/config.h.cmake b/config.h.cmake
index ce6f14723b887..493f90edc8cff 100644
--- a/config.h.cmake
+++ b/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
diff --git a/configure.cmake b/configure.cmake
index a1fdb98389d67..acb80965a6425 100644
--- a/configure.cmake
+++ b/configure.cmake
@@ -855,6 +855,16 @@ IF(NOT CMAKE_CROSSCOMPILING AND NOT MSVC)
     }
    " HAVE_FAKE_PAUSE_INSTRUCTION)
   ENDIF()
+  IF (NOT HAVE_PAUSE_INSTRUCTION)
+    CHECK_C_SOURCE_COMPILES("
+    int main()
+    {
+     __asm__ __volatile__ (\"or 1,1,1\");
+     __asm__ __volatile__ (\"or 2,2,2\");
+     return 0;
+    }
+    " HAVE_HMT_PRIORITY_INSTRUCTION)
+  ENDIF()
 ENDIF()
   
 CHECK_SYMBOL_EXISTS(tcgetattr "termios.h" HAVE_TCGETATTR 1)
diff --git a/storage/innobase/include/ut0ut.h b/storage/innobase/include/ut0ut.h
index 338cdf115d75a..03e4b5df8337f 100644
--- a/storage/innobase/include/ut0ut.h
+++ b/storage/innobase/include/ut0ut.h
@@ -92,6 +92,14 @@ struct ut_when_dtor {
 #  define UT_RELAX_CPU() __asm__ __volatile__ ("":::"memory")
 # endif
 
+# if defined(HAVE_HMT_PRIORITY_INSTRUCTION)
+#  define UT_LOW_PRIORITY_CPU() __asm__ __volatile__ ("or 1,1,1")
+#  define UT_RESUME_PRIORITY_CPU() __asm__ __volatile__ ("or 2,2,2")
+# 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.
diff --git a/storage/innobase/ut/ut0ut.cc b/storage/innobase/ut/ut0ut.cc
index a0fc154ce6b94..85e77b6cced6b 100644
--- a/storage/innobase/ut/ut0ut.cc
+++ b/storage/innobase/ut/ut0ut.cc
@@ -401,6 +401,8 @@ ut_delay(
 {
 	ulint	i, j;
 
+	UT_LOW_PRIORITY_CPU();
+
 	j = 0;
 
 	for (i = 0; i < delay * 50; i++) {
@@ -408,6 +410,8 @@ ut_delay(
 		UT_RELAX_CPU();
 	}
 
+	UT_RESUME_PRIORITY_CPU();
+
 	return(j);
 }
 #endif /* !UNIV_HOTBACKUP */
diff --git a/storage/xtradb/include/ut0ut.h b/storage/xtradb/include/ut0ut.h
index fbebbc68628f4..6ea23514994cd 100644
--- a/storage/xtradb/include/ut0ut.h
+++ b/storage/xtradb/include/ut0ut.h
@@ -89,6 +89,14 @@ struct ut_when_dtor {
 #  define UT_RELAX_CPU() __asm__ __volatile__ ("":::"memory")
 # endif
 
+# if defined(HAVE_HMT_PRIORITY_INSTRUCTION)
+#  define UT_LOW_PRIORITY_CPU() __asm__ __volatile__ ("or 1,1,1")
+#  define UT_RESUME_PRIORITY_CPU() __asm__ __volatile__ ("or 2,2,2")
+# 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.
diff --git a/storage/xtradb/ut/ut0ut.cc b/storage/xtradb/ut/ut0ut.cc
index 5e85594ef440a..42fc04e5f8169 100644
--- a/storage/xtradb/ut/ut0ut.cc
+++ b/storage/xtradb/ut/ut0ut.cc
@@ -402,6 +402,8 @@ ut_delay(
 {
 	ulint	i, j;
 
+	UT_LOW_PRIORITY_CPU();
+
 	j = 0;
 
 	for (i = 0; i < delay * 50; i++) {
@@ -409,6 +411,8 @@ ut_delay(
 		UT_RELAX_CPU();
 	}
 
+	UT_RESUME_PRIORITY_CPU();
+
 	return(j);
 }
 #endif /* !UNIV_HOTBACKUP */