From a6e9176bc547afeca88a9b5e4784f99d18b5ae66 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?J=C3=A9r=C3=B4me=20Duval?= <jerome.duval@gmail.com>
Date: Fri, 23 Sep 2022 18:55:15 +0200
Subject: [PATCH] x86_64: prefix rdtsc with lfence, or use rdtscp if available

a strong monotonicity is a property of system_time(). Performance should be
a bit impacted (around 10%).
removed omit-frame-pointer hints as the compiler generates the same without.

Change-Id: I920145d616604f32b38d9e134ce83f3e0e848bc7
Reviewed-on: https://review.haiku-os.org/c/haiku/+/5690
Reviewed-by: Adrien Destugues <pulkomandy@pulkomandy.tk>
Tested-by: Commit checker robot <no-reply+buildbot@haiku-os.org>
---
 .../libroot/os/arch/x86_64/system_time.cpp    | 79 ++++++++++++++++---
 1 file changed, 67 insertions(+), 12 deletions(-)

diff --git a/src/system/libroot/os/arch/x86_64/system_time.cpp b/src/system/libroot/os/arch/x86_64/system_time.cpp
index 5678d31323..aa02d85a98 100644
--- a/src/system/libroot/os/arch/x86_64/system_time.cpp
+++ b/src/system/libroot/os/arch/x86_64/system_time.cpp
@@ -4,6 +4,8 @@
  */
 
 
+#include <OS.h>
+
 #include <stdint.h>
 
 #include <x86intrin.h>
@@ -13,26 +15,79 @@ static uint64_t cv_factor;
 static uint64_t cv_factor_nsec;
 
 
-extern "C" void
-__x86_setup_system_time(uint64_t cv, uint64_t cv_nsec)
-{
-	cv_factor = cv;
-	cv_factor_nsec = cv_nsec;
-}
-
-
-extern "C" [[gnu::optimize("omit-frame-pointer")]] int64_t
-system_time()
+static int64_t
+__system_time_lfence()
 {
+	__builtin_ia32_lfence();
 	__uint128_t time = static_cast<__uint128_t>(__rdtsc()) * cv_factor;
 	return time >> 64;
 }
 
 
-extern "C" [[gnu::optimize("omit-frame-pointer")]] int64_t
-system_time_nsecs()
+static int64_t
+__system_time_rdtscp()
 {
+	uint32_t aux;
+	__uint128_t time = static_cast<__uint128_t>(__rdtscp(&aux)) * cv_factor;
+	return time >> 64;
+}
+
+
+static int64_t
+__system_time_nsecs_lfence()
+{
+	__builtin_ia32_lfence();
 	__uint128_t t = static_cast<__uint128_t>(__rdtsc()) * cv_factor_nsec;
 	return t >> 32;
 }
 
+
+static int64_t
+__system_time_nsecs_rdtscp()
+{
+	uint32_t aux;
+	__uint128_t t = static_cast<__uint128_t>(__rdtscp(&aux)) * cv_factor_nsec;
+	return t >> 32;
+}
+
+
+static int64_t (*sSystemTime)(void) = __system_time_lfence;
+static int64_t (*sSystemTimeNsecs)(void) = __system_time_nsecs_lfence;
+
+
+// from kernel/arch/x86/arch_cpu.h
+#define IA32_FEATURE_AMD_EXT_RDTSCP		(1 << 27) // rdtscp instruction
+
+
+extern "C" void
+__x86_setup_system_time(uint64_t cv, uint64_t cv_nsec)
+{
+	cv_factor = cv;
+	cv_factor_nsec = cv_nsec;
+
+	cpuid_info cpuInfo;
+	get_cpuid(&cpuInfo, 0x80000000, 0);
+	if (cpuInfo.eax_0.max_eax >= 0x80000001) {
+		get_cpuid(&cpuInfo, 0x80000001, 0);
+		if ((cpuInfo.regs.edx & IA32_FEATURE_AMD_EXT_RDTSCP)!= 0) {
+			sSystemTime = __system_time_rdtscp;
+			sSystemTimeNsecs = __system_time_nsecs_rdtscp;
+		}
+	}
+}
+
+
+extern "C" int64_t
+system_time()
+{
+	return sSystemTime();
+}
+
+
+
+extern "C" int64_t
+system_time_nsecs()
+{
+	return sSystemTimeNsecs();
+}
+