add the SSE support

=> context switch stores also the SSE registers

arch/x86/include/asm/processor.h

@@ -39,6 +39,45 @@
 extern "C" {
 #endif
 
+// feature list 1
+#define CPU_FEATURE_FPU		(1 << 0)
+#define CPU_FEATURE_MMX		(1 << 23)
+#define CPU_FEATURE_FXSR		(1 << 24)
+#define CPU_FEATURE_SSE		(1 << 25)
+#define CPU_FEATURE_SSE2		(1 << 26)
+
+// feature list 2
+#define CPU_FEATURE_AVX		(1 << 28)
+
+typedef struct {
+	uint32_t feature1, feature2;
+} cpu_info_t;
+
+extern cpu_info_t cpu_info;
+
+// determine the cpu features
+int cpu_detection(void);
+
+inline static uint32_t has_fpu(void)
+{
+	return (cpu_info.feature1 & CPU_FEATURE_FPU);
+}
+
+inline static uint32_t has_fxsr(void)
+{
+	return (cpu_info.feature1 & CPU_FEATURE_FXSR);
+}
+
+inline static uint32_t has_xmm(void)
+{
+	return (cpu_info.feature1 & CPU_FEATURE_SSE);
+}
+
+inline static uint32_t has_avx(void)
+{
+	return (cpu_info.feature2 & CPU_FEATURE_AVX);
+}
+
 /** @brief Read out time stamp counter
  *
  * The rdtsc asm command puts a 64 bit time stamp value
@@ -250,6 +289,8 @@ inline static int system_init(void)
 #ifdef CONFIG_PCI
 	pci_init();
 #endif
+	cpu_detection();
+
 	return 0;
 }
 

arch/x86/include/asm/tasks_types.h

@@ -22,6 +22,7 @@
 #define __ARCH_TASKS_TYPES__
 
 #include <metalsvm/stddef.h>
+#include <asm/processor.h>
 
 #ifdef __cplusplus
 extern "C" {
@@ -61,7 +62,10 @@ union fpu_state {
 };
 
 static inline void save_fpu_state(union fpu_state* state) {
-	asm volatile ("fsave %0; fwait" : "=m"((*state).fsave));
+	if (has_fxsr())
+		asm volatile ("fxsave %0; fnclex" : "=m"((*state).fxsave));
+	else
+		asm volatile ("fsave %0; fwait" : "=m"((*state).fsave));
 }
 
 #ifdef __cplusplus

arch/x86/kernel/isrs.c

@@ -165,6 +165,26 @@ void isrs_install(void)
 	irq_install_handler(7, fpu_handler);
 }
 
+static void fpu_init(union fpu_state* fpu)
+{
+	if (has_fxsr()) {
+		i387_fxsave_t* fx = &fpu->fxsave;
+
+                memset(fx, 0, sizeof(union fpu_state));
+                fx->cwd = 0x37f;
+                if (has_xmm)
+                        fx->mxcsr = 0x1f80;
+        } else {
+                i387_fsave_t *fp = &fpu->fsave;
+                memset(fp, 0, sizeof(union fpu_state));
+                fp->cwd = 0xffff037fu;
+                fp->swd = 0xffff0000u;
+                fp->twd = 0xffffffffu;
+                fp->fos = 0xffff0000u;
+        }
+
+}
+
 static void fpu_handler(struct state *s)
 {
 	task_t* task = per_core(current_task);
@@ -172,13 +192,16 @@ static void fpu_handler(struct state *s)
 	asm volatile ("clts"); // clear the TS flag of cr0
 	if (!(task->flags & TASK_FPU_INIT))  {
 		// use the FPU at the first time => Initialize FPU
-		asm volatile ("finit");
-		task->flags = task->flags|TASK_FPU_INIT|TASK_FPU_USED;
-	} else {
-		// restore the FPU context 
-		asm volatile ("frstor %0" :: "m"(task->fpu.fsave)); // restore fpu state
-		task->flags |= TASK_FPU_USED;
+		fpu_init(&task->fpu);
+		task->flags |= TASK_FPU_INIT;
 	}
+
+	// restore the FPU context 
+	if (has_fxsr())
+		asm volatile ("fxrstor %0" :: "m"(task->fpu.fxsave));
+	else
+		asm volatile ("frstor %0" :: "m"(task->fpu.fsave));
+	task->flags |= TASK_FPU_USED;
 }
 
 /** @brief Exception messages

arch/x86/kernel/processor.c

@@ -25,8 +25,31 @@
 #include <asm/RCCE_lib.h>
 #endif
 
+cpu_info_t cpu_info = { 0, 0 };
 static uint32_t cpu_freq = 0;
 
+int cpu_detection(void)
+{
+	uint32_t a, b, cr4;
+
+	cpuid(1, &a, &b, &cpu_info.feature2, &cpu_info.feature1);
+
+	cr4 = read_cr4();
+	if (has_fxsr())
+		cr4 |= 0x200; // set the OSFXSR bit
+	if (has_xmm())
+                cr4 |= 0x400; // set the OSXMMEXCPT bit
+	write_cr4(cr4);
+
+	if (has_avx())
+		kprintf("The CPU owns the Advanced Vector Extensions (AVX). However, MetalSVM doesn't support AVX!\n");
+
+	if (has_fpu())
+		asm volatile ("fninit");
+
+	return 0;
+}
+
 uint32_t detect_cpu_frequency(void)
 {
 #ifdef CONFIG_ROCKCREEK

kernel/tests.c

@@ -133,6 +133,7 @@ int test_init(void)
 	//create_user_task(NULL, "/bin/hello", argv);
 	create_user_task(NULL, "/bin/tests", argv);
 	//create_user_task(NULL, "/bin/jacobi", argv);
+	//create_user_task(NULL, "/bin/jacobi", argv);
 
 	return 0;
 }