sh: Move over and enable FPU support for SH-5.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>

4 files changed, 177 insertions, 6 deletions

diff --git a/arch/sh/kernel/cpu/sh5/Makefile b/arch/sh/kernel/cpu/sh5/Makefile
index 7ea10732b5c..6d388e8d899 100644
--- a/arch/sh/kernel/cpu/sh5/Makefile
+++ b/arch/sh/kernel/cpu/sh5/Makefile
@@ -1,3 +1,4 @@
 obj-y := entry.o switchto.o
 
 obj-$(CONFIG_KALLSYMS)		+= unwind.o
+obj-$(CONFIG_SH_FPU)		+= fpu.o
diff --git a/arch/sh/kernel/cpu/sh5/fpu.c b/arch/sh/kernel/cpu/sh5/fpu.c
new file mode 100644
index 00000000000..d3f5e7468df
--- /dev/null
+++ b/arch/sh/kernel/cpu/sh5/fpu.c
@@ -0,0 +1,170 @@
+/*
+ * This file is subject to the terms and conditions of the GNU General Public
+ * License.  See the file "COPYING" in the main directory of this archive
+ * for more details.
+ *
+ * arch/sh64/kernel/fpu.c
+ *
+ * Copyright (C) 2001  Manuela Cirronis, Paolo Alberelli
+ * Copyright (C) 2002  STMicroelectronics Limited
+ *   Author : Stuart Menefy
+ *
+ * Started from SH4 version:
+ *   Copyright (C) 1999, 2000  Kaz Kojima & Niibe Yutaka
+ *
+ */
+
+#include <linux/sched.h>
+#include <linux/signal.h>
+#include <asm/processor.h>
+#include <asm/user.h>
+#include <asm/io.h>
+
+/*
+ * Initially load the FPU with signalling NANS.  This bit pattern
+ * has the property that no matter whether considered as single or as
+ * double precision, it still represents a signalling NAN.
+ */
+#define sNAN64		0xFFFFFFFFFFFFFFFFULL
+#define sNAN32		0xFFFFFFFFUL
+
+static union sh_fpu_union init_fpuregs = {
+	.hard = {
+	  .fp_regs = { [0 ... 63] = sNAN32 },
+	  .fpscr = FPSCR_INIT
+	}
+};
+
+inline void fpsave(struct sh_fpu_hard_struct *fpregs)
+{
+	asm volatile("fst.p     %0, (0*8), fp0\n\t"
+		     "fst.p     %0, (1*8), fp2\n\t"
+		     "fst.p     %0, (2*8), fp4\n\t"
+		     "fst.p     %0, (3*8), fp6\n\t"
+		     "fst.p     %0, (4*8), fp8\n\t"
+		     "fst.p     %0, (5*8), fp10\n\t"
+		     "fst.p     %0, (6*8), fp12\n\t"
+		     "fst.p     %0, (7*8), fp14\n\t"
+		     "fst.p     %0, (8*8), fp16\n\t"
+		     "fst.p     %0, (9*8), fp18\n\t"
+		     "fst.p     %0, (10*8), fp20\n\t"
+		     "fst.p     %0, (11*8), fp22\n\t"
+		     "fst.p     %0, (12*8), fp24\n\t"
+		     "fst.p     %0, (13*8), fp26\n\t"
+		     "fst.p     %0, (14*8), fp28\n\t"
+		     "fst.p     %0, (15*8), fp30\n\t"
+		     "fst.p     %0, (16*8), fp32\n\t"
+		     "fst.p     %0, (17*8), fp34\n\t"
+		     "fst.p     %0, (18*8), fp36\n\t"
+		     "fst.p     %0, (19*8), fp38\n\t"
+		     "fst.p     %0, (20*8), fp40\n\t"
+		     "fst.p     %0, (21*8), fp42\n\t"
+		     "fst.p     %0, (22*8), fp44\n\t"
+		     "fst.p     %0, (23*8), fp46\n\t"
+		     "fst.p     %0, (24*8), fp48\n\t"
+		     "fst.p     %0, (25*8), fp50\n\t"
+		     "fst.p     %0, (26*8), fp52\n\t"
+		     "fst.p     %0, (27*8), fp54\n\t"
+		     "fst.p     %0, (28*8), fp56\n\t"
+		     "fst.p     %0, (29*8), fp58\n\t"
+		     "fst.p     %0, (30*8), fp60\n\t"
+		     "fst.p     %0, (31*8), fp62\n\t"
+
+		     "fgetscr   fr63\n\t"
+		     "fst.s     %0, (32*8), fr63\n\t"
+		: /* no output */
+		: "r" (fpregs)
+		: "memory");
+}
+
+
+static inline void
+fpload(struct sh_fpu_hard_struct *fpregs)
+{
+	asm volatile("fld.p     %0, (0*8), fp0\n\t"
+		     "fld.p     %0, (1*8), fp2\n\t"
+		     "fld.p     %0, (2*8), fp4\n\t"
+		     "fld.p     %0, (3*8), fp6\n\t"
+		     "fld.p     %0, (4*8), fp8\n\t"
+		     "fld.p     %0, (5*8), fp10\n\t"
+		     "fld.p     %0, (6*8), fp12\n\t"
+		     "fld.p     %0, (7*8), fp14\n\t"
+		     "fld.p     %0, (8*8), fp16\n\t"
+		     "fld.p     %0, (9*8), fp18\n\t"
+		     "fld.p     %0, (10*8), fp20\n\t"
+		     "fld.p     %0, (11*8), fp22\n\t"
+		     "fld.p     %0, (12*8), fp24\n\t"
+		     "fld.p     %0, (13*8), fp26\n\t"
+		     "fld.p     %0, (14*8), fp28\n\t"
+		     "fld.p     %0, (15*8), fp30\n\t"
+		     "fld.p     %0, (16*8), fp32\n\t"
+		     "fld.p     %0, (17*8), fp34\n\t"
+		     "fld.p     %0, (18*8), fp36\n\t"
+		     "fld.p     %0, (19*8), fp38\n\t"
+		     "fld.p     %0, (20*8), fp40\n\t"
+		     "fld.p     %0, (21*8), fp42\n\t"
+		     "fld.p     %0, (22*8), fp44\n\t"
+		     "fld.p     %0, (23*8), fp46\n\t"
+		     "fld.p     %0, (24*8), fp48\n\t"
+		     "fld.p     %0, (25*8), fp50\n\t"
+		     "fld.p     %0, (26*8), fp52\n\t"
+		     "fld.p     %0, (27*8), fp54\n\t"
+		     "fld.p     %0, (28*8), fp56\n\t"
+		     "fld.p     %0, (29*8), fp58\n\t"
+		     "fld.p     %0, (30*8), fp60\n\t"
+
+		     "fld.s     %0, (32*8), fr63\n\t"
+		     "fputscr   fr63\n\t"
+
+		     "fld.p     %0, (31*8), fp62\n\t"
+		: /* no output */
+		: "r" (fpregs) );
+}
+
+void fpinit(struct sh_fpu_hard_struct *fpregs)
+{
+	*fpregs = init_fpuregs.hard;
+}
+
+asmlinkage void
+do_fpu_error(unsigned long ex, struct pt_regs *regs)
+{
+	struct task_struct *tsk = current;
+
+	regs->pc += 4;
+
+	tsk->thread.trap_no = 11;
+	tsk->thread.error_code = 0;
+	force_sig(SIGFPE, tsk);
+}
+
+
+asmlinkage void
+do_fpu_state_restore(unsigned long ex, struct pt_regs *regs)
+{
+	void die(const char *str, struct pt_regs *regs, long err);
+
+	if (! user_mode(regs))
+		die("FPU used in kernel", regs, ex);
+
+	regs->sr &= ~SR_FD;
+
+	if (last_task_used_math == current)
+		return;
+
+	enable_fpu();
+	if (last_task_used_math != NULL) {
+		/* Other processes fpu state, save away */
+		fpsave(&last_task_used_math->thread.fpu.hard);
+        }
+        last_task_used_math = current;
+        if (used_math()) {
+                fpload(&current->thread.fpu.hard);
+        } else {
+		/* First time FPU user.  */
+		fpload(&init_fpuregs.hard);
+                set_used_math();
+        }
+	disable_fpu();
+}
+
diff --git a/arch/sh/kernel/process_64.c b/arch/sh/kernel/process_64.c
index 0761af4d2a4..0c2bc61b66b 100644
--- a/arch/sh/kernel/process_64.c
+++ b/arch/sh/kernel/process_64.c
@@ -469,9 +469,9 @@ int dump_fpu(struct pt_regs *regs, elf_fpregset_t *fpu)
 	fpvalid = !!tsk_used_math(tsk);
 	if (fpvalid) {
 		if (current == last_task_used_math) {
-			grab_fpu();
+			enable_fpu();
 			fpsave(&tsk->thread.fpu.hard);
-			release_fpu();
+			disable_fpu();
 			last_task_used_math = 0;
 			regs->sr |= SR_FD;
 		}
@@ -496,9 +496,9 @@ int copy_thread(int nr, unsigned long clone_flags, unsigned long usp,
 
 #ifdef CONFIG_SH_FPU
 	if(last_task_used_math == current) {
-		grab_fpu();
+		enable_fpu();
 		fpsave(&current->thread.fpu.hard);
-		release_fpu();
+		disable_fpu();
 		last_task_used_math = NULL;
 		regs->sr |= SR_FD;
 	}
diff --git a/arch/sh/kernel/signal_64.c b/arch/sh/kernel/signal_64.c
index 069fb608364..08f403e2366 100644
--- a/arch/sh/kernel/signal_64.c
+++ b/arch/sh/kernel/signal_64.c
@@ -211,9 +211,9 @@ setup_sigcontext_fpu(struct pt_regs *regs, struct sigcontext __user *sc)
 		return err;
 
 	if (current == last_task_used_math) {
-		grab_fpu();
+		enable_fpu();
 		fpsave(&current->thread.fpu.hard);
-		release_fpu();
+		disable_fpu();
 		last_task_used_math = NULL;
 		regs->sr |= SR_FD;
 	}