x86: Extend XSAVE support to unbound state

* i386/i386/fpu.c (fp_xsave_size): New variable.
(init_fpu): Save XSAVE size to fp_xsave_size.
(fpu_module_init): Pass fp_xsave_size as size to kmem_cache_init.
(fpu_set_state, fp_load, fp_state_alloc): Use fp_xsave_size to clear
ifps.
* i386/include/mach/i386/fp_reg.h (struct i386_xfp_save): Replace static
fp_yreg_word with extended field.
* i386/i386/thread.h (struct i386_fpsave_state): Make fp_valid field
first in the structure to let the extended finish the structure.

3 files changed, 12 insertions, 23 deletions

diff --git a/i386/i386/fpu.c b/i386/i386/fpu.c
index 517764d6..3961f17b 100644
--- a/i386/i386/fpu.c
+++ b/i386/i386/fpu.c
@@ -71,6 +71,7 @@
 
 int		fp_kind = FP_387;	/* 80387 present */
 uint64_t	fp_xsave_support;	/* Bitmap of supported XSAVE save areas */
+unsigned	fp_xsave_size = sizeof(struct i386_fpsave_state);
 struct kmem_cache	ifps_cache;	/* cache for FPU save area */
 static unsigned long	mxcsr_feature_mask = 0xffffffff;	/* Always AND user-provided mxcsr with this security mask */
 
@@ -168,22 +169,12 @@ init_fpu(void)
 		    eax = 0xd;
 		    ecx = 0x0;
 		    cpuid(eax, ebx, ecx, edx);
+		    fp_xsave_size = offsetof(struct i386_fpsave_state, xfp_save_state) + ebx;
 
-		    if (ebx > sizeof(struct i386_xfp_save)) {
-			    /* TODO: rather make struct unbound and set size
-			     * in fpu_module_init */
-			    printf("XSAVE area size %u larger than provisioned "
-				   "%u, not enabling XSAVE\n",
-				   ebx, sizeof(struct i386_xfp_save));
-#ifndef MACH_RING1
-			    set_cr4(get_cr4() & ~(CR4_OSFXSR | CR4_OSXSAVE));
-#endif /* MACH_RING1 */
-		    } else {
-			    fp_kind = FP_387X;
-		    }
+		    fp_kind = FP_387X;
 		}
 
-		if (fp_kind == FP_387 && CPU_HAS_FEATURE(CPU_FEATURE_FXSR)) {
+		else if (CPU_HAS_FEATURE(CPU_FEATURE_FXSR)) {
 #ifndef MACH_RING1
 		    set_cr4(get_cr4() | CR4_OSFXSR);
 #endif /* MACH_RING1 */
@@ -226,7 +217,7 @@ void
 fpu_module_init(void)
 {
 	kmem_cache_init(&ifps_cache, "i386_fpsave_state",
-			sizeof(struct i386_fpsave_state),
+			fp_xsave_size,
 			alignof(struct i386_fpsave_state),
 			NULL, 0);
 }
@@ -398,7 +389,7 @@ ASSERT_IPL(SPL0);
 	    /*
 	     * Ensure that reserved parts of the environment are 0.
 	     */
-	    memset(ifps, 0, sizeof(*ifps));
+	    memset(ifps, 0, fp_xsave_size);
 
 	    if (fp_kind == FP_387X || fp_kind == FP_387FX) {
 		int i;
@@ -842,7 +833,7 @@ ASSERT_IPL(SPL0);
 	ifps = pcb->ims.ifps;
 	if (ifps == 0) {
 	    ifps = (struct i386_fpsave_state *) kmem_cache_alloc(&ifps_cache);
-	    memset(ifps, 0, sizeof *ifps);
+	    memset(ifps, 0, fp_xsave_size);
 	    pcb->ims.ifps = ifps;
 	    fpinit(thread);
 #if 1
@@ -895,7 +886,7 @@ fp_state_alloc(void)
 	struct i386_fpsave_state *ifps;
 
 	ifps = (struct i386_fpsave_state *)kmem_cache_alloc(&ifps_cache);
-	memset(ifps, 0, sizeof *ifps);
+	memset(ifps, 0, fp_xsave_size);
 	pcb->ims.ifps = ifps;
 
 	ifps->fp_valid = TRUE;
diff --git a/i386/i386/thread.h b/i386/i386/thread.h
index 3d4808c7..4a9c1987 100644
--- a/i386/i386/thread.h
+++ b/i386/i386/thread.h
@@ -129,6 +129,8 @@ struct i386_kernel_state {
  */
 
 struct i386_fpsave_state {
+	boolean_t		fp_valid;
+
 	union {
 		struct {
 			struct i386_fp_save	fp_save_state;
@@ -136,7 +138,6 @@ struct i386_fpsave_state {
 		};
 		struct i386_xfp_save	xfp_save_state;
 	};
-	boolean_t		fp_valid;
 };
 
 /*
diff --git a/i386/include/mach/i386/fp_reg.h b/i386/include/mach/i386/fp_reg.h
index 5992ea78..41301ec4 100644
--- a/i386/include/mach/i386/fp_reg.h
+++ b/i386/include/mach/i386/fp_reg.h
@@ -79,12 +79,9 @@ struct i386_xfp_save {
 					/* space for 16 128-bit XMM registers */
 	unsigned int	padding[24];
 	struct i386_xfp_xstate_header header;
-
-	unsigned char	fp_yreg_word[16][16];
-					/* space for the high part of the
-					 * 16 256-bit YMM registers */
+	unsigned char	extended[0];	/* Extended region */
 } __attribute__((packed, aligned(64)));
-_Static_assert(sizeof(struct i386_xfp_save) == 512 + 8*8 + 16*16);
+_Static_assert(sizeof(struct i386_xfp_save) == 512 + 8*8);
 
 /*
  * Control register