s390/fpu: remove anonymous union from struct fpu

The anonymous union within struct fpu contains a floating point register
array and a vector register array. Given that the vector register is always
present remove the floating point register array. For configurations
without vector registers save the floating point register contents within
their corresponding vector register location.

This allows to remove the union, and also to simplify ptrace and perf code.
Signed-off-by: Heiko Carstens <hca@linux.ibm.com>

arch/s390/include/asm/fpu-types.h

@@ -12,23 +12,15 @@
 #include <asm/sigcontext.h>
 
 struct fpu {
-	__u32 fpc;		/* Floating-point control */
-	union {
-		/* Floating-point register save area */
-		freg_t fprs[__NUM_FPRS];
-		/* Vector register save area */
-		__vector128 vxrs[__NUM_VXRS];
-	};
+	u32 fpc;
+	__vector128 vxrs[__NUM_VXRS] __aligned(8);
 };
 
 /* In-kernel FPU state structure */
 struct kernel_fpu {
 	int	    mask;
 	u32	    fpc;
-	union {
-		freg_t fprs[__NUM_FPRS];
-		__vector128 vxrs[__NUM_VXRS];
-	};
+	__vector128 vxrs[__NUM_VXRS] __aligned(8);
 };
 
 #define DECLARE_KERNEL_FPU_ONSTACK(name)	\

arch/s390/include/asm/fpu.h

@@ -98,44 +98,68 @@ static __always_inline void load_vx_regs(__vector128 *vxrs)
 	fpu_vlm(16, 31, &vxrs[16]);
 }
 
+static __always_inline void __save_fp_regs(freg_t *fprs, unsigned int offset)
+{
+	fpu_std(0, &fprs[0 * offset]);
+	fpu_std(1, &fprs[1 * offset]);
+	fpu_std(2, &fprs[2 * offset]);
+	fpu_std(3, &fprs[3 * offset]);
+	fpu_std(4, &fprs[4 * offset]);
+	fpu_std(5, &fprs[5 * offset]);
+	fpu_std(6, &fprs[6 * offset]);
+	fpu_std(7, &fprs[7 * offset]);
+	fpu_std(8, &fprs[8 * offset]);
+	fpu_std(9, &fprs[9 * offset]);
+	fpu_std(10, &fprs[10 * offset]);
+	fpu_std(11, &fprs[11 * offset]);
+	fpu_std(12, &fprs[12 * offset]);
+	fpu_std(13, &fprs[13 * offset]);
+	fpu_std(14, &fprs[14 * offset]);
+	fpu_std(15, &fprs[15 * offset]);
+}
+
+static __always_inline void __load_fp_regs(freg_t *fprs, unsigned int offset)
+{
+	fpu_ld(0, &fprs[0 * offset]);
+	fpu_ld(1, &fprs[1 * offset]);
+	fpu_ld(2, &fprs[2 * offset]);
+	fpu_ld(3, &fprs[3 * offset]);
+	fpu_ld(4, &fprs[4 * offset]);
+	fpu_ld(5, &fprs[5 * offset]);
+	fpu_ld(6, &fprs[6 * offset]);
+	fpu_ld(7, &fprs[7 * offset]);
+	fpu_ld(8, &fprs[8 * offset]);
+	fpu_ld(9, &fprs[9 * offset]);
+	fpu_ld(10, &fprs[10 * offset]);
+	fpu_ld(11, &fprs[11 * offset]);
+	fpu_ld(12, &fprs[12 * offset]);
+	fpu_ld(13, &fprs[13 * offset]);
+	fpu_ld(14, &fprs[14 * offset]);
+	fpu_ld(15, &fprs[15 * offset]);
+}
+
 static __always_inline void save_fp_regs(freg_t *fprs)
 {
-	fpu_std(0, &fprs[0]);
-	fpu_std(1, &fprs[1]);
-	fpu_std(2, &fprs[2]);
-	fpu_std(3, &fprs[3]);
-	fpu_std(4, &fprs[4]);
-	fpu_std(5, &fprs[5]);
-	fpu_std(6, &fprs[6]);
-	fpu_std(7, &fprs[7]);
-	fpu_std(8, &fprs[8]);
-	fpu_std(9, &fprs[9]);
-	fpu_std(10, &fprs[10]);
-	fpu_std(11, &fprs[11]);
-	fpu_std(12, &fprs[12]);
-	fpu_std(13, &fprs[13]);
-	fpu_std(14, &fprs[14]);
-	fpu_std(15, &fprs[15]);
+	__save_fp_regs(fprs, sizeof(freg_t) / sizeof(freg_t));
 }
 
 static __always_inline void load_fp_regs(freg_t *fprs)
 {
-	fpu_ld(0, &fprs[0]);
-	fpu_ld(1, &fprs[1]);
-	fpu_ld(2, &fprs[2]);
-	fpu_ld(3, &fprs[3]);
-	fpu_ld(4, &fprs[4]);
-	fpu_ld(5, &fprs[5]);
-	fpu_ld(6, &fprs[6]);
-	fpu_ld(7, &fprs[7]);
-	fpu_ld(8, &fprs[8]);
-	fpu_ld(9, &fprs[9]);
-	fpu_ld(10, &fprs[10]);
-	fpu_ld(11, &fprs[11]);
-	fpu_ld(12, &fprs[12]);
-	fpu_ld(13, &fprs[13]);
-	fpu_ld(14, &fprs[14]);
-	fpu_ld(15, &fprs[15]);
+	__load_fp_regs(fprs, sizeof(freg_t) / sizeof(freg_t));
+}
+
+static __always_inline void save_fp_regs_vx(__vector128 *vxrs)
+{
+	freg_t *fprs = (freg_t *)&vxrs[0].high;
+
+	__save_fp_regs(fprs, sizeof(__vector128) / sizeof(freg_t));
+}
+
+static __always_inline void load_fp_regs_vx(__vector128 *vxrs)
+{
+	freg_t *fprs = (freg_t *)&vxrs[0].high;
+
+	__load_fp_regs(fprs, sizeof(__vector128) / sizeof(freg_t));
 }
 
 static inline void kernel_fpu_begin(struct kernel_fpu *state, int flags)
@@ -170,7 +194,7 @@ static inline void save_kernel_fpu_regs(struct thread_struct *thread)
 	if (likely(cpu_has_vx()))
 		save_vx_regs(state->vxrs);
 	else
-		save_fp_regs(state->fprs);
+		save_fp_regs_vx(state->vxrs);
 }
 
 static inline void restore_kernel_fpu_regs(struct thread_struct *thread)
@@ -183,7 +207,7 @@ static inline void restore_kernel_fpu_regs(struct thread_struct *thread)
 	if (likely(cpu_has_vx()))
 		load_vx_regs(state->vxrs);
 	else
-		load_fp_regs(state->fprs);
+		load_fp_regs_vx(state->vxrs);
 }
 
 static inline void convert_vx_to_fp(freg_t *fprs, __vector128 *vxrs)
@@ -206,19 +230,13 @@ static inline void fpregs_store(_s390_fp_regs *fpregs, struct fpu *fpu)
 {
 	fpregs->pad = 0;
 	fpregs->fpc = fpu->fpc;
-	if (cpu_has_vx())
-		convert_vx_to_fp((freg_t *)&fpregs->fprs, fpu->vxrs);
-	else
-		memcpy((freg_t *)&fpregs->fprs, fpu->fprs, sizeof(fpregs->fprs));
+	convert_vx_to_fp((freg_t *)&fpregs->fprs, fpu->vxrs);
 }
 
 static inline void fpregs_load(_s390_fp_regs *fpregs, struct fpu *fpu)
 {
 	fpu->fpc = fpregs->fpc;
-	if (cpu_has_vx())
-		convert_fp_to_vx(fpu->vxrs, (freg_t *)&fpregs->fprs);
-	else
-		memcpy(fpu->fprs, (freg_t *)&fpregs->fprs, sizeof(fpregs->fprs));
+	convert_fp_to_vx(fpu->vxrs, (freg_t *)&fpregs->fprs);
 }
 
 #endif /* _ASM_S390_FPU_H */

arch/s390/kernel/fpu.c

@@ -24,7 +24,7 @@ void __kernel_fpu_begin(struct kernel_fpu *state, int flags)
 		fpu_stfpc(&state->fpc);
 	if (!cpu_has_vx()) {
 		if (flags & KERNEL_VXR_LOW)
-			save_fp_regs(state->fprs);
+			save_fp_regs_vx(state->vxrs);
 		return;
 	}
 	mask = flags & KERNEL_VXR;
@@ -73,7 +73,7 @@ void __kernel_fpu_end(struct kernel_fpu *state, int flags)
 		fpu_lfpc(&state->fpc);
 	if (!cpu_has_vx()) {
 		if (flags & KERNEL_VXR_LOW)
-			load_fp_regs(state->fprs);
+			load_fp_regs_vx(state->vxrs);
 		return;
 	}
 	mask = flags & KERNEL_VXR;
@@ -115,7 +115,7 @@ void __load_user_fpu_regs(void)
 	if (likely(cpu_has_vx()))
 		load_vx_regs(state->vxrs);
 	else
-		load_fp_regs(state->fprs);
+		load_fp_regs_vx(state->vxrs);
 	clear_thread_flag(TIF_FPU);
 }
 
@@ -143,7 +143,7 @@ void save_user_fpu_regs(void)
 	if (likely(cpu_has_vx()))
 		save_vx_regs(state->vxrs);
 	else
-		save_fp_regs(state->fprs);
+		save_fp_regs_vx(state->vxrs);
 	set_thread_flag(TIF_FPU);
 out:
 	local_irq_restore(flags);

arch/s390/kernel/perf_regs.c

@@ -19,10 +19,7 @@ u64 perf_reg_value(struct pt_regs *regs, int idx)
 			return 0;
 
 		idx -= PERF_REG_S390_FP0;
-		if (cpu_has_vx())
-			fp = *(freg_t *)(current->thread.ufpu.vxrs + idx);
-		else
-			fp = current->thread.ufpu.fprs[idx];
+		fp = *(freg_t *)(current->thread.ufpu.vxrs + idx);
 		return fp.ui;
 	}
 

arch/s390/kernel/ptrace.c

@@ -252,17 +252,10 @@ static unsigned long __peek_user(struct task_struct *child, addr_t addr)
 
 	} else if (addr < offsetof(struct user, regs.fp_regs) + sizeof(s390_fp_regs)) {
 		/*
-		 * floating point regs. are either in child->thread.ufpu
-		 * or the child->thread.ufpu.vxrs array
+		 * floating point regs. are in the child->thread.ufpu.vxrs array
 		 */
 		offset = addr - offsetof(struct user, regs.fp_regs.fprs);
-		if (cpu_has_vx())
-			tmp = *(addr_t *)
-			       ((addr_t)child->thread.ufpu.vxrs + 2 * offset);
-		else
-			tmp = *(addr_t *)
-			       ((addr_t)child->thread.ufpu.fprs + offset);
-
+		tmp = *(addr_t *)((addr_t)child->thread.ufpu.vxrs + 2 * offset);
 	} else if (addr < offsetof(struct user, regs.per_info) + sizeof(per_struct)) {
 		/*
 		 * Handle access to the per_info structure.
@@ -400,17 +393,10 @@ static int __poke_user(struct task_struct *child, addr_t addr, addr_t data)
 
 	} else if (addr < offsetof(struct user, regs.fp_regs) + sizeof(s390_fp_regs)) {
 		/*
-		 * floating point regs. are either in child->thread.ufpu
-		 * or the child->thread.ufpu.vxrs array
+		 * floating point regs. are in the child->thread.ufpu.vxrs array
 		 */
 		offset = addr - offsetof(struct user, regs.fp_regs.fprs);
-		if (cpu_has_vx())
-			*(addr_t *)((addr_t)
-				child->thread.ufpu.vxrs + 2 * offset) = data;
-		else
-			*(addr_t *)((addr_t)
-				child->thread.ufpu.fprs + offset) = data;
-
+		*(addr_t *)((addr_t)child->thread.ufpu.vxrs + 2 * offset) = data;
 	} else if (addr < offsetof(struct user, regs.per_info) + sizeof(per_struct)) {
 		/*
 		 * Handle access to the per_info structure.
@@ -627,17 +613,10 @@ static u32 __peek_user_compat(struct task_struct *child, addr_t addr)
 
 	} else if (addr < offsetof(struct compat_user, regs.fp_regs) + sizeof(s390_fp_regs)) {
 		/*
-		 * floating point regs. are either in child->thread.ufpu
-		 * or the child->thread.ufpu.vxrs array
+		 * floating point regs. are in the child->thread.ufpu.vxrs array
 		 */
 		offset = addr - offsetof(struct compat_user, regs.fp_regs.fprs);
-		if (cpu_has_vx())
-			tmp = *(__u32 *)
-			       ((addr_t)child->thread.ufpu.vxrs + 2 * offset);
-		else
-			tmp = *(__u32 *)
-			       ((addr_t)child->thread.ufpu.fprs + offset);
-
+		tmp = *(__u32 *)((addr_t)child->thread.ufpu.vxrs + 2 * offset);
 	} else if (addr < offsetof(struct compat_user, regs.per_info) + sizeof(struct compat_per_struct_kernel)) {
 		/*
 		 * Handle access to the per_info structure.
@@ -753,17 +732,10 @@ static int __poke_user_compat(struct task_struct *child,
 
 	} else if (addr < offsetof(struct compat_user, regs.fp_regs) + sizeof(s390_fp_regs)) {
 		/*
-		 * floating point regs. are either in child->thread.ufpu
-		 * or the child->thread.ufpu.vxrs array
+		 * floating point regs. are in the child->thread.ufpu.vxrs array
 		 */
 		offset = addr - offsetof(struct compat_user, regs.fp_regs.fprs);
-		if (cpu_has_vx())
-			*(__u32 *)((addr_t)
-				child->thread.ufpu.vxrs + 2 * offset) = tmp;
-		else
-			*(__u32 *)((addr_t)
-				child->thread.ufpu.fprs + offset) = tmp;
-
+		*(__u32 *)((addr_t)child->thread.ufpu.vxrs + 2 * offset) = tmp;
 	} else if (addr < offsetof(struct compat_user, regs.per_info) + sizeof(struct compat_per_struct_kernel)) {
 		/*
 		 * Handle access to the per_info structure.
@@ -912,12 +884,7 @@ static int s390_fpregs_set(struct task_struct *target,
 
 	if (target == current)
 		save_user_fpu_regs();
-
-	if (cpu_has_vx())
-		convert_vx_to_fp(fprs, target->thread.ufpu.vxrs);
-	else
-		memcpy(&fprs, target->thread.ufpu.fprs, sizeof(fprs));
-
+	convert_vx_to_fp(fprs, target->thread.ufpu.vxrs);
 	if (count > 0 && pos < offsetof(s390_fp_regs, fprs)) {
 		u32 ufpc[2] = { target->thread.ufpu.fpc, 0 };
 		rc = user_regset_copyin(&pos, &count, &kbuf, &ubuf, &ufpc,
@@ -934,12 +901,7 @@ static int s390_fpregs_set(struct task_struct *target,
 					fprs, offsetof(s390_fp_regs, fprs), -1);
 	if (rc)
 		return rc;
-
-	if (cpu_has_vx())
-		convert_fp_to_vx(target->thread.ufpu.vxrs, fprs);
-	else
-		memcpy(target->thread.ufpu.fprs, &fprs, sizeof(fprs));
-
+	convert_fp_to_vx(target->thread.ufpu.vxrs, fprs);
 	return rc;
 }