ppc64: Allow user to change MSR_FE0 and MSR_FE1 - from ppc32

arch/ppc64/kernel/head.S

@@ -1421,7 +1421,7 @@ _STATIC(load_up_fpu)
 	addi	r4,r4,THREAD	       /* want THREAD of last_task_used_math */
 	SAVE_32FPRS(0, r4)
 	mffs	fr0
-	stfd	fr0,THREAD_FPSCR-4(r4)
+	stfd	fr0,THREAD_FPSCR(r4)
 	ld	r5,PT_REGS(r4)
 	ld	r4,_MSR-STACK_FRAME_OVERHEAD(r5)
 	li	r20,MSR_FP|MSR_FE0|MSR_FE1
@@ -1430,10 +1430,12 @@ _STATIC(load_up_fpu)
 1:
 #endif /* CONFIG_SMP */
 	/* enable use of FP after return */
-	ori	r23,r23,MSR_FP|MSR_FE0|MSR_FE1
-	ld	r4, PACACURRENT(r13)
+	ld	r4,PACACURRENT(r13)
 	addi	r5,r4,THREAD		/* Get THREAD */
-	lfd	fr0,THREAD_FPSCR-4(r5)
+	lwz	r4,THREAD_FPEXC_MODE(r5)
+	ori	r23,r23,MSR_FP
+	or	r23,r23,r4
+	lfd	fr0,THREAD_FPSCR(r5)
 	mtfsf	0xff,fr0
 	REST_32FPRS(0, r5)
 #ifndef CONFIG_SMP
@@ -1476,7 +1478,7 @@ _GLOBAL(giveup_fpu)
 	cmpi	0,r5,0
 	SAVE_32FPRS(0, r3)
 	mffs	fr0
-	stfd	fr0,THREAD_FPSCR-4(r3)
+	stfd	fr0,THREAD_FPSCR(r3)
 	beq	1f
 	ld	r4,_MSR-STACK_FRAME_OVERHEAD(r5)
 	li	r3,MSR_FP|MSR_FE0|MSR_FE1

arch/ppc64/kernel/misc.S

@@ -460,21 +460,21 @@ _GLOBAL(_get_HID0)
 	blr
 
 _GLOBAL(cvt_fd)
-	lfd	0,-4(r5)	/* load up fpscr value */
+	lfd	0,0(r5)		/* load up fpscr value */
 	mtfsf	0xff,0
 	lfs	0,0(r3)
 	stfd	0,0(r4)
 	mffs	0		/* save new fpscr value */
-	stfd	0,-4(r5)
+	stfd	0,0(r5)
 	blr
 
 _GLOBAL(cvt_df)
-	lfd	0,-4(r5)	/* load up fpscr value */
+	lfd	0,0(r5)		/* load up fpscr value */
 	mtfsf	0xff,0
 	lfd	0,0(r3)
 	stfs	0,0(r4)
 	mffs	0		/* save new fpscr value */
-	stfd	0,-4(r5)
+	stfd	0,0(r5)
 	blr
 
 /*

arch/ppc64/kernel/mk_defs.c

@@ -51,6 +51,7 @@ main(void)
 	/* task_struct->thread */
 	DEFINE(THREAD, offsetof(struct task_struct, thread));
 	DEFINE(PT_REGS, offsetof(struct thread_struct, regs));
+	DEFINE(THREAD_FPEXC_MODE, offsetof(struct thread_struct, fpexc_mode));
 	DEFINE(THREAD_FPR0, offsetof(struct thread_struct, fpr[0]));
 	DEFINE(THREAD_FPSCR, offsetof(struct thread_struct, fpscr));
 	DEFINE(KSP, offsetof(struct thread_struct, ksp));

arch/ppc64/kernel/process.c

@@ -163,8 +163,7 @@ release_thread(struct task_struct *t)
  */
 int
 copy_thread(int nr, unsigned long clone_flags, unsigned long usp,
-	    unsigned long unused,
-	    struct task_struct *p, struct pt_regs *regs)
+	    unsigned long unused, struct task_struct *p, struct pt_regs *regs)
 {
 	struct pt_regs *childregs, *kregs;
 	extern void ret_from_fork(void);
@@ -208,17 +207,6 @@ copy_thread(int nr, unsigned long clone_flags, unsigned long usp,
  	 */
 	kregs->nip = *((unsigned long *)ret_from_fork);
 
-	/*
-	 * copy fpu info - assume lazy fpu switch now always
-	 *  -- Cort
-	 */
-	if (regs->msr & MSR_FP) {
-		giveup_fpu(current);
-		childregs->msr &= ~(MSR_FP | MSR_FE0 | MSR_FE1);
-	}
-	memcpy(&p->thread.fpr, &current->thread.fpr, sizeof(p->thread.fpr));
-	p->thread.fpscr = current->thread.fpscr;
-
 	return 0;
 }
 
@@ -247,10 +235,38 @@ void start_thread(struct pt_regs *regs, unsigned long nip, unsigned long sp)
 	current->thread.fpscr = 0;
 }
 
+/* XXX temporary */
+#define PR_FP_EXC_PRECISE	3	/* precise exception mode */
+
+int set_fpexc_mode(struct task_struct *tsk, unsigned int val)
+{
+	struct pt_regs *regs = tsk->thread.regs;
+
+	if (val > PR_FP_EXC_PRECISE)
+		return -EINVAL;
+	tsk->thread.fpexc_mode = __pack_fe01(val);
+	if (regs != NULL && (regs->msr & MSR_FP) != 0)
+		regs->msr = (regs->msr & ~(MSR_FE0|MSR_FE1))
+			| tsk->thread.fpexc_mode;
+	return 0;
+}
+
+int get_fpexc_mode(struct task_struct *tsk, unsigned long adr)
+{
+	unsigned int val;
+
+	val = __unpack_fe01(tsk->thread.fpexc_mode);
+	return put_user(val, (unsigned int *) adr);
+}
+
 int sys_clone(int p1, int p2, int p3, int p4, int p5, int p6,
 	      struct pt_regs *regs)
 {
 	struct task_struct *p;
+
+	if (regs->msr & MSR_FP)
+		giveup_fpu(current);
+
 	p = do_fork(p1 & ~CLONE_IDLETASK, regs->gpr[1], regs, 0);
 	return IS_ERR(p) ? PTR_ERR(p) : p->pid;
 }
@@ -259,6 +275,10 @@ int sys_fork(int p1, int p2, int p3, int p4, int p5, int p6,
 	     struct pt_regs *regs)
 {
 	struct task_struct *p;
+
+	if (regs->msr & MSR_FP)
+		giveup_fpu(current);
+
 	p = do_fork(SIGCHLD, regs->gpr[1], regs, 0);
 	return IS_ERR(p) ? PTR_ERR(p) : p->pid;
 }
@@ -267,6 +287,10 @@ int sys_vfork(int p1, int p2, int p3, int p4, int p5, int p6,
 			 struct pt_regs *regs)
 {
 	struct task_struct *p;
+
+	if (regs->msr & MSR_FP)
+		giveup_fpu(current);
+
 	p = do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->gpr[1], regs, 0);
 	return IS_ERR(p) ? PTR_ERR(p) : p->pid;
 }

arch/ppc64/kernel/ptrace32.c

@@ -177,10 +177,7 @@ int sys32_ptrace(long request, long pid, unsigned long addr, unsigned long data)
 		if (numReg >= PT_FPR0) {
 			if (child->thread.regs->msr & MSR_FP)
 				giveup_fpu(child);
-		        if (numReg == PT_FPSCR) 
-			        tmp = ((unsigned int *)child->thread.fpscr);
-		        else 
-			        tmp = ((unsigned long int *)child->thread.fpr)[numReg - PT_FPR0];
+			tmp = ((unsigned long int *)child->thread.fpr)[numReg - PT_FPR0];
 		} else { /* register within PT_REGS struct */
 			tmp = get_reg(child, numReg);
 		} 

arch/ppc64/kernel/signal.c

@@ -55,7 +55,19 @@
  * able to continue 'til the next breakpoint from within the signal
  * handler, even if the handler returns.
  */
+#if 0
 #define MSR_USERCHANGE	(MSR_FE0 | MSR_FE1)
+#else
+/*
+ * glibc tries to set FE0/FE1 via a signal handler. Since it only ever
+ * sets both bits and this is the default setting we now disable this
+ * behaviour. This is done to insure the new prctl which alters FE0/FE1 does
+ * not get overriden by glibc. Setting and clearing FE0/FE1 via signal
+ * handler has always been bogus since load_up_fpu used to set FE0/FE1
+ * unconditionally.
+ */
+#define MSR_USERCHANGE	0
+#endif
 
 /*
  * When we have signals to deliver, we set up on the

arch/ppc64/kernel/signal32.c

@@ -39,7 +39,19 @@
  * able to continue 'til the next breakpoint from within the signal
  * handler, even if the handler returns.
  */
+#if 0
 #define MSR_USERCHANGE	(MSR_FE0 | MSR_FE1)
+#else
+/*
+ * glibc tries to set FE0/FE1 via a signal handler. Since it only ever
+ * sets both bits and this is the default setting we now disable this
+ * behaviour. This is done to insure the new prctl which alters FE0/FE1 does
+ * not get overriden by glibc. Setting and clearing FE0/FE1 via signal
+ * handler has always been bogus since load_up_fpu used to set FE0/FE1
+ * unconditionally.
+ */
+#define MSR_USERCHANGE	0
+#endif
 
 struct timespec32 {
 	s32    tv_sec;

arch/ppc64/kernel/traps.c

@@ -240,14 +240,12 @@ InstructionBreakpointException(struct pt_regs *regs)
 static void parse_fpe(struct pt_regs *regs)
 {
 	siginfo_t info;
-	unsigned int *tmp;
-	unsigned int fpscr;
+	unsigned long fpscr;
 
 	if (regs->msr & MSR_FP)
 		giveup_fpu(current);
 
-	tmp = &current->thread.fpscr;
-	fpscr = *tmp;
+	fpscr = current->thread.fpscr;
 
 	/* Invalid operation */
 	if ((fpscr & FPSCR_VE) && (fpscr & FPSCR_VX))

include/asm-ppc64/processor.h

@@ -652,6 +652,7 @@ struct thread_struct {
 	mm_segment_t	fs;		/* for get_fs() validation */
 	double		fpr[32];	/* Complete floating point set */
 	unsigned long	fpscr;		/* Floating point status */
+	unsigned int	fpexc_mode;	/* Floating-point exception mode */
 };
 
 #define INIT_SP		(sizeof(init_stack) + (unsigned long) &init_stack)
@@ -661,7 +662,8 @@ struct thread_struct {
 	(struct pt_regs *)INIT_SP - 1, /* regs */ \
 	KERNEL_DS, /*fs*/ \
 	{0}, /* fpr */ \
-	0 /* fpscr */ \
+	0, /* fpscr */ \
+	MSR_FE0|MSR_FE1, /* fpexc_mode */ \
 }
 
 /*
@@ -689,6 +691,23 @@ unsigned long get_wchan(struct task_struct *p);
 #define KSTK_EIP(tsk)  ((tsk)->thread.regs? (tsk)->thread.regs->nip: 0)
 #define KSTK_ESP(tsk)  ((tsk)->thread.regs? (tsk)->thread.regs->gpr[1]: 0)
 
+/* Get/set floating-point exception mode */
+#define GET_FPEXC_CTL(tsk, adr) get_fpexc_mode((tsk), (adr))
+#define SET_FPEXC_CTL(tsk, val) set_fpexc_mode((tsk), (val))
+
+extern int get_fpexc_mode(struct task_struct *tsk, unsigned long adr);
+extern int set_fpexc_mode(struct task_struct *tsk, unsigned int val);
+
+static inline unsigned int __unpack_fe01(unsigned long msr_bits)
+{
+	return ((msr_bits & MSR_FE0) >> 10) | ((msr_bits & MSR_FE1) >> 8);
+}
+
+static inline unsigned int __pack_fe01(unsigned int fpmode)
+{
+	return ((fpmode << 10) & MSR_FE0) | ((fpmode << 8) & MSR_FE1);
+}
+
 #define cpu_relax()     barrier()
 
 /*