x86: Pave the way for xsave/xrstor support

* i386/i386/fpu.h (xgetbv, get_xcr0, xsetbv, set_xcr0): New inline
functions.
(xsave, xrstor): New macros.
* i386/i386/locore.S (cpu_features): Extend to two words.
(cpu_features_edx, cpu_features_ecx): New labels.
(discover_x86_cpu_type): Also save ecx cpuid(1) report.
* x86_64/locore.S: Likewise.
* i386/i386/locore.h (cpu_features): Extend to two words.
(CPU_FEATURE_XSAVE): New macro.
* i386/i386/proc_reg.h (CR4_OSXSAVE): New macro.

4 files changed, 38 insertions, 2 deletions

diff --git a/i386/i386/fpu.h b/i386/i386/fpu.h
index caade5d4..4050cce4 100644
--- a/i386/i386/fpu.h
+++ b/i386/i386/fpu.h
@@ -73,6 +73,35 @@
 #define	fxrstor(state) \
 	asm volatile("fxrstor %0" : : "m" (state))
 
+static inline uint64_t xgetbv(uint32_t n) {
+	uint32_t eax, edx;
+	asm volatile("xgetbv" : "=a" (eax), "=d" (edx) : "c" (n));
+	return eax + ((uint64_t) edx << 32);
+}
+
+static inline uint64_t get_xcr0(void) {
+	return xgetbv(0);
+}
+
+static inline void xsetbv(uint32_t n, uint64_t value) {
+	uint32_t eax, edx;
+
+	eax = value;
+	edx = value >> 32;
+
+	asm volatile("xsetbv" : : "c" (n), "a" (eax), "d" (edx));
+}
+
+static inline void set_xcr0(uint64_t value) {
+	xsetbv(0, value);
+}
+
+#define	xsave(state) \
+	asm volatile("xsave %0" : "=m" (*state))
+
+#define	xrstor(state) \
+	asm volatile("xrstor %0" : : "m" (state))
+
 #define fwait() \
     	asm("fwait");
 
diff --git a/i386/i386/locore.S b/i386/i386/locore.S
index a0960a26..bee3630c 100644
--- a/i386/i386/locore.S
+++ b/i386/i386/locore.S
@@ -1177,6 +1177,9 @@ syscall_addr:
 
 	.data
 DATA(cpu_features)
+DATA(cpu_features_edx)
+	.long	0
+DATA(cpu_features_ecx)
 	.long	0
 	.text
 
@@ -1226,7 +1229,8 @@ ENTRY(discover_x86_cpu_type)
 	   use it to find out what we are. */
 0:	movl	$1,%eax			/* Fetch CPU type info ... */
 	cpuid				/*  ... into eax */
-	movl	%edx,cpu_features	/* Keep a copy */
+	movl	%ecx,cpu_features_ecx	/* Keep a copy */
+	movl	%edx,cpu_features_edx	/* Keep a copy */
 	shrl	$8,%eax			/* Slide family bits down */
 	andl	$15,%eax		/* And select them */
 
diff --git a/i386/i386/locore.h b/i386/i386/locore.h
index 6948f72d..00da07ad 100644
--- a/i386/i386/locore.h
+++ b/i386/i386/locore.h
@@ -58,7 +58,7 @@ extern void cpu_shutdown (void);
 
 extern int syscall (void);
 
-extern unsigned int cpu_features[1];
+extern unsigned int cpu_features[2];
 
 #define CPU_FEATURE_FPU		 0
 #define CPU_FEATURE_VME		 1
@@ -89,6 +89,7 @@ extern unsigned int cpu_features[1];
 #define CPU_FEATURE_HTT		28
 #define CPU_FEATURE_TM		29
 #define CPU_FEATURE_PBE		31
+#define CPU_FEATURE_XSAVE	(1*32 + 26)
 
 #define CPU_HAS_FEATURE(feature) (cpu_features[(feature) / 32] & (1 << ((feature) % 32)))
 
diff --git a/i386/i386/proc_reg.h b/i386/i386/proc_reg.h
index c787ef26..624819c6 100644
--- a/i386/i386/proc_reg.h
+++ b/i386/i386/proc_reg.h
@@ -68,6 +68,8 @@
 					 * and FXRSTOR instructions */
 #define	CR4_OSXMMEXCPT	0x0400		/* Operating System Support for Unmasked
 					 * SIMD Floating-Point Exceptions */
+#define	CR4_OSXSAVE	0x40000		/* Operating System Support for XSAVE
+					 * and XRSTOR instructions */
 
 #ifndef	__ASSEMBLER__
 #ifdef	__GNUC__