Implemented system_time() for x86_64.

* Uses 64-bit multiplication, special handling for CPUs clocked < 1 GHz
  in system_time_nsecs() not required like on x86.
* Tested against a straight conversion of the x86 version, noticably
  faster with a large number of system_time() calls.

headers/private/kernel/arch/x86/arch_cpu.h

@@ -367,9 +367,14 @@ extern "C" {
 
 struct arch_thread;
 
-
+#ifdef __x86_64__
+void __x86_setup_system_time(uint64 conversionFactor,
+	uint64 conversionFactorNsecs);
+#else
 void __x86_setup_system_time(uint32 conversionFactor,
 	uint32 conversionFactorNsecs, bool conversionFactorNsecsShift);
+#endif
+
 void x86_context_switch(struct arch_thread* oldState,
 	struct arch_thread* newState);
 void x86_userspace_thread_exit(void);

src/system/kernel/arch/x86/arch_cpu.cpp

@@ -820,6 +820,12 @@ arch_cpu_init(kernel_args* args)
 	uint32 conversionFactor = args->arch_args.system_time_cv_factor;
 	uint64 conversionFactorNsecs = (uint64)conversionFactor * 1000;
 
+#ifdef __x86_64__
+	// The x86_64 system_time() implementation uses 64-bit multiplication and
+	// therefore shifting is not necessary for low frequencies (it's also not
+	// too likely that there'll be any x86_64 CPUs clocked under 1GHz).
+	__x86_setup_system_time(conversionFactor, conversionFactorNsecs);
+#else
 	if (conversionFactorNsecs >> 32 != 0) {
 		// the TSC frequency is < 1 GHz, which forces us to shift the factor
 		__x86_setup_system_time(conversionFactor, conversionFactorNsecs >> 16,
@@ -828,6 +834,7 @@ arch_cpu_init(kernel_args* args)
 		// the TSC frequency is >= 1 GHz
 		__x86_setup_system_time(conversionFactor, conversionFactorNsecs, false);
 	}
+#endif
 
 	return B_OK;
 }

src/system/libroot/os/arch/x86_64/system_time_asm.S

@@ -7,143 +7,65 @@
  * Distributed under the terms of the NewOS License.
  */
 
+
 #include <asm_defs.h>
 
-FUNCTION(__x86_setup_system_time):
-	ret
-FUNCTION_END(__x86_setup_system_time)
 
-// TODO: Implement these.
-
-/* int64 system_time(); */
-FUNCTION(system_time):
-	ud2a
-	ret
-FUNCTION_END(system_time)
-
-
-/* int64 system_time_nsecs(); */
-FUNCTION(system_time_nsecs):
-	ud2a
-	ret
-FUNCTION_END(system_time_nsecs)
-
-#if 0
 /* saves the conversion factor needed for system_time */
-.lcomm cv_factor				4
-.lcomm cv_factor_nsecs			4
-.lcomm cv_factor_nsecs_shift	1
+.lcomm cv_factor				8
+.lcomm cv_factor_nsecs			8
 
 
 .text
 
 
 FUNCTION(__x86_setup_system_time):
-	movl	4(%esp), %eax
-	movl	%eax, cv_factor
-	movl	8(%esp), %eax
-	movl	%eax, cv_factor_nsecs
-	movb	12(%esp), %al
-	movb	%al, cv_factor_nsecs_shift
+	movq	%rdi, cv_factor
+	movq	%rsi, cv_factor_nsecs
 	ret
 FUNCTION_END(__x86_setup_system_time)
 
 
 /* int64 system_time(); */
 FUNCTION(system_time):
-	pushl	%ebx
-	pushl	%ecx
-	movl	cv_factor, %ebx
+	movq	cv_factor, %rcx
 
-	/* load 64-bit factor into %eax (low), %edx (high) */
-	rdtsc		/* time in %edx,%eax */
+	// Load 64-bit TSC into %eax (low), %edx (high).
+	rdtsc
 
-	movl	%edx, %ecx	/* save high half */
-	mull	%ebx 		/* truncate %eax, but keep %edx */
-	movl	%ecx, %eax
-	movl	%edx, %ecx	/* save high half of low */
-	mull	%ebx /*, %eax*/
-	/* now compute  [%edx, %eax] + [%ecx], propagating carry */
-	subl	%ebx, %ebx	/* need zero to propagate carry */
-	addl	%ecx, %eax
-	adc		%ebx, %edx
-	popl	%ecx
-	popl	%ebx
+	// Convert into a single 64-bit value.
+	shl		$32, %rdx
+	orq		%rdx, %rax
+
+	// Multiply by conversion factor, result in %rax (low), %rdx (high).
+	mulq	%rcx
+
+	// Shift the result right by 32 bits.
+	shr		$32, %rax
+	shl		$32, %rdx
+	orq		%rdx, %rax
 	ret
 FUNCTION_END(system_time)
 
 
 /* int64 system_time_nsecs(); */
 FUNCTION(system_time_nsecs):
-	testb	$0, cv_factor_nsecs_shift
-	jne		1f
+	// Same algorithm as system_time(), just with a different factor.
+	movq	cv_factor_nsecs, %rcx
 
-	/* same algorithm as system_time(), just with a different factor */
+	// Load 64-bit TSC into %eax (low), %edx (high).
+	rdtsc
 
-	pushl	%ebx
-	pushl	%ecx
-	movl	cv_factor_nsecs, %ebx
+	// Convert into a single 64-bit value.
+	shl		$32, %rdx
+	orq		%rdx, %rax
 
-	/* load 64-bit factor into %eax (low), %edx (high) */
-	rdtsc		/* time in %edx,%eax */
+	// Multiply by conversion factor, result in %rax (low), %rdx (high).
+	mulq	%rcx
 
-	movl	%edx, %ecx	/* save high half */
-	mull	%ebx 		/* truncate %eax, but keep %edx */
-	movl	%ecx, %eax
-	movl	%edx, %ecx	/* save high half of low */
-	mull	%ebx /*, %eax*/
-	/* now compute  [%edx, %eax] + [%ecx], propagating carry */
-	subl	%ebx, %ebx	/* need zero to propagate carry */
-	addl	%ecx, %eax
-	adc		%ebx, %edx
-	popl	%ecx
-	popl	%ebx
-	ret
-
-1:
-	/* TSC frequency is less than 1 GHz -- we shift everything up 16 bit */
-
-	pushl	%ebx
-	pushl	%ecx
-	pushl	%esi
-	movl	cv_factor_nsecs, %ebx
-
-	/* load 64-bit factor into %eax (low), %edx (high) */
-	rdtsc		/* time in %edx,%eax */
-
-	/* save high half */
-	movl	%edx, %ecx
-
-	/* multiply low half by conversion factor */
-	mull	%ebx
-
-	/* save result */
-	movl	%eax, %esi	/* low half -> %esi */
-	movl	%ecx, %eax
-	movl	%edx, %ecx	/* high half -> %ecx */
-
-	/* multiply high half by conversion factor */
-	mull	%ebx
-
-	/* now compute  [%edx, %eax] + [%ecx], propagating carry */
-	xorl	%ebx, %ebx	/* need zero to propagate carry */
-	addl	%ecx, %eax
-	adc		%ebx, %edx
-
-	/* shift the result left 16 bit */
-	shl		$16, %edx
-	movl	%eax, %ebx
-	shr		$16, %ebx
-	orw		%bx, %dx
-	shl		$16, %eax
-
-	/* add the high 16 bit of the low half of the low product */
-	shr		$16, %esi
-	orw		%si, %ax
-
-	popl	%esi
-	popl	%ecx
-	popl	%ebx
+	// Shift the result right by 32 bits.
+	shr		$32, %rax
+	shl		$32, %rdx
+	orq		%rdx, %rax
 	ret
 FUNCTION_END(system_time_nsecs)
-#endif