s390: add support for vector extension

The vector extension introduces 32 128-bit vector registers and a set of
instruction to operate on the vector registers.

The kernel can control the use of vector registers for the problem state
program with a bit in control register 0. Once enabled for a process the
kernel needs to retain the content of the vector registers on context
switch. The signal frame is extended to include the vector registers.
Two new register sets NT_S390_VXRS_LOW and NT_S390_VXRS_HIGH are added
to the regset interface for the debugger and core dumps.
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>

arch/s390/include/asm/elf.h

@@ -102,6 +102,7 @@
 #define HWCAP_S390_ETF3EH	256
 #define HWCAP_S390_HIGH_GPRS	512
 #define HWCAP_S390_TE		1024
+#define HWCAP_S390_VXRS		2048
 
 /*
  * These are used to set parameters in the core dumps.

arch/s390/include/asm/lowcore.h

@@ -310,7 +310,10 @@ struct _lowcore {
 
 	/* Extended facility list */
 	__u64	stfle_fac_list[32];		/* 0x0f00 */
-	__u8	pad_0x1000[0x11b8-0x1000];	/* 0x1000 */
+	__u8	pad_0x1000[0x11b0-0x1000];	/* 0x1000 */
+
+	/* Pointer to vector register save area */
+	__u64	vector_save_area_addr;		/* 0x11b0 */
 
 	/* 64 bit extparam used for pfault/diag 250: defined by architecture */
 	__u64	ext_params2;			/* 0x11B8 */
@@ -334,9 +337,10 @@ struct _lowcore {
 
 	/* Transaction abort diagnostic block */
 	__u8	pgm_tdb[256];			/* 0x1800 */
+	__u8	pad_0x1900[0x1c00-0x1900];	/* 0x1900 */
 
-	/* align to the top of the prefix area */
-	__u8	pad_0x1900[0x2000-0x1900];	/* 0x1900 */
+	/* Software defined save area for vector registers */
+	__u8	vector_save_area[1024];		/* 0x1c00 */
 } __packed;
 
 #endif /* CONFIG_32BIT */

arch/s390/include/asm/nmi.h

@@ -38,7 +38,7 @@ struct mci {
 	__u32 pm :  1; /* 22 psw program mask and cc validity */
 	__u32 ia :  1; /* 23 psw instruction address validity */
 	__u32 fa :  1; /* 24 failing storage address validity */
-	__u32	 :  1; /* 25 */
+	__u32 vr :  1; /* 25 vector register validity */
 	__u32 ec :  1; /* 26 external damage code validity */
 	__u32 fp :  1; /* 27 floating point register validity */
 	__u32 gr :  1; /* 28 general register validity */

arch/s390/include/asm/processor.h

@@ -117,6 +117,7 @@ struct thread_struct {
 	int ri_signum;
 #ifdef CONFIG_64BIT
 	unsigned char trap_tdb[256];	/* Transaction abort diagnose block */
+	__vector128 *vxrs;		/* Vector register save area */
 #endif
 };
 

arch/s390/include/asm/setup.h

@@ -56,6 +56,7 @@ extern void detect_memory_memblock(void);
 #define MACHINE_FLAG_TOPOLOGY	(1UL << 14)
 #define MACHINE_FLAG_TE		(1UL << 15)
 #define MACHINE_FLAG_TLB_LC	(1UL << 17)
+#define MACHINE_FLAG_VX		(1UL << 18)
 
 #define MACHINE_IS_VM		(S390_lowcore.machine_flags & MACHINE_FLAG_VM)
 #define MACHINE_IS_KVM		(S390_lowcore.machine_flags & MACHINE_FLAG_KVM)
@@ -78,6 +79,7 @@ extern void detect_memory_memblock(void);
 #define MACHINE_HAS_TOPOLOGY	(0)
 #define MACHINE_HAS_TE		(0)
 #define MACHINE_HAS_TLB_LC	(0)
+#define MACHINE_HAS_VX		(0)
 #else /* CONFIG_64BIT */
 #define MACHINE_HAS_IEEE	(1)
 #define MACHINE_HAS_CSP		(1)
@@ -90,6 +92,7 @@ extern void detect_memory_memblock(void);
 #define MACHINE_HAS_TOPOLOGY	(S390_lowcore.machine_flags & MACHINE_FLAG_TOPOLOGY)
 #define MACHINE_HAS_TE		(S390_lowcore.machine_flags & MACHINE_FLAG_TE)
 #define MACHINE_HAS_TLB_LC	(S390_lowcore.machine_flags & MACHINE_FLAG_TLB_LC)
+#define MACHINE_HAS_VX		(S390_lowcore.machine_flags & MACHINE_FLAG_VX)
 #endif /* CONFIG_64BIT */
 
 /*

arch/s390/include/asm/switch_to.h

@@ -103,6 +103,48 @@ static inline void restore_fp_regs(freg_t *fprs)
 	asm volatile("ld 15,%0" : : "Q" (fprs[15]));
 }
 
+static inline void save_vx_regs(__vector128 *vxrs)
+{
+	typedef struct { __vector128 _[__NUM_VXRS]; } addrtype;
+
+	asm volatile(
+		"	la	1,%0\n"
+		"	.word	0xe70f,0x1000,0x003e\n"	/* vstm 0,15,0(1) */
+		"	.word	0xe70f,0x1100,0x0c3e\n"	/* vstm 16,31,256(1) */
+		: "=Q" (*(addrtype *) vxrs) : : "1");
+}
+
+static inline void restore_vx_regs(__vector128 *vxrs)
+{
+	typedef struct { __vector128 _[__NUM_VXRS]; } addrtype;
+
+	asm volatile(
+		"	la	1,%0\n"
+		"	.word	0xe70f,0x1000,0x0036\n"	/* vlm 0,15,0(1) */
+		"	.word	0xe70f,0x1100,0x0c36\n"	/* vlm 16,31,256(1) */
+		: : "Q" (*(addrtype *) vxrs) : "1");
+}
+
+static inline void save_fp_vx_regs(struct task_struct *task)
+{
+#ifdef CONFIG_64BIT
+	if (task->thread.vxrs)
+		save_vx_regs(task->thread.vxrs);
+	else
+#endif
+	save_fp_regs(task->thread.fp_regs.fprs);
+}
+
+static inline void restore_fp_vx_regs(struct task_struct *task)
+{
+#ifdef CONFIG_64BIT
+	if (task->thread.vxrs)
+		restore_vx_regs(task->thread.vxrs);
+	else
+#endif
+	restore_fp_regs(task->thread.fp_regs.fprs);
+}
+
 static inline void save_access_regs(unsigned int *acrs)
 {
 	typedef struct { int _[NUM_ACRS]; } acrstype;
@@ -120,16 +162,16 @@ static inline void restore_access_regs(unsigned int *acrs)
 #define switch_to(prev,next,last) do {					\
 	if (prev->mm) {							\
 		save_fp_ctl(&prev->thread.fp_regs.fpc);			\
-		save_fp_regs(prev->thread.fp_regs.fprs);		\
+		save_fp_vx_regs(prev);					\
 		save_access_regs(&prev->thread.acrs[0]);		\
 		save_ri_cb(prev->thread.ri_cb);				\
 	}								\
 	if (next->mm) {							\
+		update_cr_regs(next);					\
 		restore_fp_ctl(&next->thread.fp_regs.fpc);		\
-		restore_fp_regs(next->thread.fp_regs.fprs);		\
+		restore_fp_vx_regs(next);				\
 		restore_access_regs(&next->thread.acrs[0]);		\
 		restore_ri_cb(next->thread.ri_cb, prev->thread.ri_cb);	\
-		update_cr_regs(next);					\
 	}								\
 	prev = __switch_to(prev,next);					\
 } while (0)

arch/s390/include/uapi/asm/sigcontext.h

@@ -7,10 +7,14 @@
 #define _ASM_S390_SIGCONTEXT_H
 
 #include <linux/compiler.h>
+#include <linux/types.h>
 
-#define __NUM_GPRS 16
-#define __NUM_FPRS 16
-#define __NUM_ACRS 16
+#define __NUM_GPRS		16
+#define __NUM_FPRS		16
+#define __NUM_ACRS		16
+#define __NUM_VXRS		32
+#define __NUM_VXRS_LOW		16
+#define __NUM_VXRS_HIGH		16
 
 #ifndef __s390x__
 
@@ -59,6 +63,16 @@ typedef struct
 	_s390_fp_regs     fpregs;
 } _sigregs;
 
+typedef struct
+{
+#ifndef __s390x__
+	unsigned long gprs_high[__NUM_GPRS];
+#endif
+	unsigned long long vxrs_low[__NUM_VXRS_LOW];
+	__vector128 vxrs_high[__NUM_VXRS_HIGH];
+	unsigned char __reserved[128];
+} _sigregs_ext;
+
 struct sigcontext
 {
 	unsigned long	oldmask[_SIGCONTEXT_NSIG_WORDS];

arch/s390/include/uapi/asm/types.h

@@ -17,6 +17,10 @@
 typedef unsigned long addr_t; 
 typedef __signed__ long saddr_t;
 
+typedef struct {
+	__u32 u[4];
+} __vector128;
+
 #endif /* __ASSEMBLY__ */
 
 #endif /* _UAPI_S390_TYPES_H */

arch/s390/include/uapi/asm/ucontext.h

@@ -7,10 +7,15 @@
 #ifndef _ASM_S390_UCONTEXT_H
 #define _ASM_S390_UCONTEXT_H
 
-#define UC_EXTENDED	0x00000001
-
-#ifndef __s390x__
+#define UC_GPRS_HIGH	1	/* uc_mcontext_ext has valid high gprs */
+#define UC_VXRS		2	/* uc_mcontext_ext has valid vector regs */
 
+/*
+ * The struct ucontext_extended describes how the registers are stored
+ * on a rt signal frame. Please note that the structure is not fixed,
+ * if new CPU registers are added to the user state the size of the
+ * struct ucontext_extended will increase.
+ */
 struct ucontext_extended {
 	unsigned long	  uc_flags;
 	struct ucontext  *uc_link;
@@ -19,11 +24,9 @@ struct ucontext_extended {
 	sigset_t	  uc_sigmask;
 	/* Allow for uc_sigmask growth.  Glibc uses a 1024-bit sigset_t.  */
 	unsigned char	  __unused[128 - sizeof(sigset_t)];
-	unsigned long	  uc_gprs_high[16];
+	_sigregs_ext	  uc_mcontext_ext;
 };
 
-#endif
-
 struct ucontext {
 	unsigned long	  uc_flags;
 	struct ucontext  *uc_link;

arch/s390/kernel/compat_linux.h

@@ -50,6 +50,14 @@ typedef struct
 	_s390_fp_regs32     fpregs;
 } _sigregs32;
 
+typedef struct
+{
+	__u32 gprs_high[__NUM_GPRS];
+	__u64 vxrs_low[__NUM_VXRS_LOW];
+	__vector128 vxrs_high[__NUM_VXRS_HIGH];
+	__u8 __reserved[128];
+} _sigregs_ext32;
+
 #define _SIGCONTEXT_NSIG32	64
 #define _SIGCONTEXT_NSIG_BPW32	32
 #define __SIGNAL_FRAMESIZE32	96
@@ -72,6 +80,7 @@ struct ucontext32 {
 	compat_sigset_t		uc_sigmask;
 	/* Allow for uc_sigmask growth.  Glibc uses a 1024-bit sigset_t.  */
 	unsigned char		__unused[128 - sizeof(compat_sigset_t)];
+	_sigregs_ext32		uc_mcontext_ext;
 };
 
 struct stat64_emu31;

arch/s390/kernel/compat_signal.c

@@ -36,17 +36,16 @@ typedef struct
 	struct sigcontext32 sc;
 	_sigregs32 sregs;
 	int signo;
-	__u32 gprs_high[NUM_GPRS];
-	__u8 retcode[S390_SYSCALL_SIZE];
+	_sigregs_ext32 sregs_ext;
+	__u16 svc_insn;		/* Offset of svc_insn is NOT fixed! */
 } sigframe32;
 
 typedef struct 
 {
 	__u8 callee_used_stack[__SIGNAL_FRAMESIZE32];
-	__u8 retcode[S390_SYSCALL_SIZE];
+	__u16 svc_insn;
 	compat_siginfo_t info;
 	struct ucontext32 uc;
-	__u32 gprs_high[NUM_GPRS];
 } rt_sigframe32;
 
 int copy_siginfo_to_user32(compat_siginfo_t __user *to, const siginfo_t *from)
@@ -151,6 +150,38 @@ int copy_siginfo_from_user32(siginfo_t *to, compat_siginfo_t __user *from)
 	return err ? -EFAULT : 0;
 }
 
+/* Store registers needed to create the signal frame */
+static void store_sigregs(void)
+{
+	int i;
+
+	save_access_regs(current->thread.acrs);
+	save_fp_ctl(&current->thread.fp_regs.fpc);
+	if (current->thread.vxrs) {
+		save_vx_regs(current->thread.vxrs);
+		for (i = 0; i < __NUM_FPRS; i++)
+			current->thread.fp_regs.fprs[i] =
+				*(freg_t *)(current->thread.vxrs + i);
+	} else
+		save_fp_regs(current->thread.fp_regs.fprs);
+}
+
+/* Load registers after signal return */
+static void load_sigregs(void)
+{
+	int i;
+
+	restore_access_regs(current->thread.acrs);
+	/* restore_fp_ctl is done in restore_sigregs */
+	if (current->thread.vxrs) {
+		for (i = 0; i < __NUM_FPRS; i++)
+			*(freg_t *)(current->thread.vxrs + i) =
+				current->thread.fp_regs.fprs[i];
+		restore_vx_regs(current->thread.vxrs);
+	} else
+		restore_fp_regs(current->thread.fp_regs.fprs);
+}
+
 static int save_sigregs32(struct pt_regs *regs, _sigregs32 __user *sregs)
 {
 	_sigregs32 user_sregs;
@@ -163,11 +194,8 @@ static int save_sigregs32(struct pt_regs *regs, _sigregs32 __user *sregs)
 		(__u32)(regs->psw.mask & PSW_MASK_BA);
 	for (i = 0; i < NUM_GPRS; i++)
 		user_sregs.regs.gprs[i] = (__u32) regs->gprs[i];
-	save_access_regs(current->thread.acrs);
 	memcpy(&user_sregs.regs.acrs, current->thread.acrs,
 	       sizeof(user_sregs.regs.acrs));
-	save_fp_ctl(&current->thread.fp_regs.fpc);
-	save_fp_regs(current->thread.fp_regs.fprs);
 	memcpy(&user_sregs.fpregs, &current->thread.fp_regs,
 	       sizeof(user_sregs.fpregs));
 	if (__copy_to_user(sregs, &user_sregs, sizeof(_sigregs32)))
@@ -207,37 +235,67 @@ static int restore_sigregs32(struct pt_regs *regs,_sigregs32 __user *sregs)
 		regs->gprs[i] = (__u64) user_sregs.regs.gprs[i];
 	memcpy(&current->thread.acrs, &user_sregs.regs.acrs,
 	       sizeof(current->thread.acrs));
-	restore_access_regs(current->thread.acrs);
 
 	memcpy(&current->thread.fp_regs, &user_sregs.fpregs,
 	       sizeof(current->thread.fp_regs));
 
-	restore_fp_regs(current->thread.fp_regs.fprs);
 	clear_pt_regs_flag(regs, PIF_SYSCALL); /* No longer in a system call */
 	return 0;
 }
 
-static int save_sigregs_gprs_high(struct pt_regs *regs, __u32 __user *uregs)
+static int save_sigregs_ext32(struct pt_regs *regs,
+			      _sigregs_ext32 __user *sregs_ext)
 {
 	__u32 gprs_high[NUM_GPRS];
+	__u64 vxrs[__NUM_VXRS_LOW];
 	int i;
 
+	/* Save high gprs to signal stack */
 	for (i = 0; i < NUM_GPRS; i++)
 		gprs_high[i] = regs->gprs[i] >> 32;
-	if (__copy_to_user(uregs, &gprs_high, sizeof(gprs_high)))
+	if (__copy_to_user(&sregs_ext->gprs_high, &gprs_high,
+			   sizeof(sregs_ext->gprs_high)))
 		return -EFAULT;
+
+	/* Save vector registers to signal stack */
+	if (current->thread.vxrs) {
+		for (i = 0; i < __NUM_VXRS_LOW; i++)
+			vxrs[i] = *((__u64 *)(current->thread.vxrs + i) + 1);
+		if (__copy_to_user(&sregs_ext->vxrs_low, vxrs,
+				   sizeof(sregs_ext->vxrs_low)) ||
+		    __copy_to_user(&sregs_ext->vxrs_high,
+				   current->thread.vxrs + __NUM_VXRS_LOW,
+				   sizeof(sregs_ext->vxrs_high)))
+			return -EFAULT;
+	}
 	return 0;
 }
 
-static int restore_sigregs_gprs_high(struct pt_regs *regs, __u32 __user *uregs)
+static int restore_sigregs_ext32(struct pt_regs *regs,
+				 _sigregs_ext32 __user *sregs_ext)
 {
 	__u32 gprs_high[NUM_GPRS];
+	__u64 vxrs[__NUM_VXRS_LOW];
 	int i;
 
-	if (__copy_from_user(&gprs_high, uregs, sizeof(gprs_high)))
+	/* Restore high gprs from signal stack */
+	if (__copy_from_user(&gprs_high, &sregs_ext->gprs_high,
+			     sizeof(&sregs_ext->gprs_high)))
 		return -EFAULT;
 	for (i = 0; i < NUM_GPRS; i++)
 		*(__u32 *)&regs->gprs[i] = gprs_high[i];
+
+	/* Restore vector registers from signal stack */
+	if (current->thread.vxrs) {
+		if (__copy_from_user(vxrs, &sregs_ext->vxrs_low,
+				     sizeof(sregs_ext->vxrs_low)) ||
+		    __copy_from_user(current->thread.vxrs + __NUM_VXRS_LOW,
+				     &sregs_ext->vxrs_high,
+				     sizeof(sregs_ext->vxrs_high)))
+			return -EFAULT;
+		for (i = 0; i < __NUM_VXRS_LOW; i++)
+			*((__u64 *)(current->thread.vxrs + i) + 1) = vxrs[i];
+	}
 	return 0;
 }
 
@@ -252,8 +310,9 @@ COMPAT_SYSCALL_DEFINE0(sigreturn)
 	set_current_blocked(&set);
 	if (restore_sigregs32(regs, &frame->sregs))
 		goto badframe;
-	if (restore_sigregs_gprs_high(regs, frame->gprs_high))
+	if (restore_sigregs_ext32(regs, &frame->sregs_ext))
 		goto badframe;
+	load_sigregs();
 	return regs->gprs[2];
 badframe:
 	force_sig(SIGSEGV, current);
@@ -269,12 +328,13 @@ COMPAT_SYSCALL_DEFINE0(rt_sigreturn)
 	if (__copy_from_user(&set, &frame->uc.uc_sigmask, sizeof(set)))
 		goto badframe;
 	set_current_blocked(&set);
+	if (compat_restore_altstack(&frame->uc.uc_stack))
+		goto badframe;
 	if (restore_sigregs32(regs, &frame->uc.uc_mcontext))
 		goto badframe;
-	if (restore_sigregs_gprs_high(regs, frame->gprs_high))
+	if (restore_sigregs_ext32(regs, &frame->uc.uc_mcontext_ext))
 		goto badframe;
-	if (compat_restore_altstack(&frame->uc.uc_stack))
-		goto badframe; 
+	load_sigregs();
 	return regs->gprs[2];
 badframe:
 	force_sig(SIGSEGV, current);
@@ -324,37 +384,64 @@ static int setup_frame32(struct ksignal *ksig, sigset_t *set,
 			 struct pt_regs *regs)
 {
 	int sig = ksig->sig;
-	sigframe32 __user *frame = get_sigframe(&ksig->ka, regs, sizeof(sigframe32));
-
+	sigframe32 __user *frame;
+	struct sigcontext32 sc;
+	unsigned long restorer;
+	size_t frame_size;
+
+	/*
+	 * gprs_high are always present for 31-bit compat tasks.
+	 * The space for vector registers is only allocated if
+	 * the machine supports it
+	 */
+	frame_size = sizeof(*frame) - sizeof(frame->sregs_ext.__reserved);
+	if (!MACHINE_HAS_VX)
+		frame_size -= sizeof(frame->sregs_ext.vxrs_low) +
+			      sizeof(frame->sregs_ext.vxrs_high);
+	frame = get_sigframe(&ksig->ka, regs, frame_size);
 	if (frame == (void __user *) -1UL)
 		return -EFAULT;
 
-	if (__copy_to_user(&frame->sc.oldmask, &set->sig, _SIGMASK_COPY_SIZE32))
+	/* Set up backchain. */
+	if (__put_user(regs->gprs[15], (unsigned int __user *) frame))
+		return -EFAULT;
+
+	/* Create struct sigcontext32 on the signal stack */
+	memcpy(&sc.oldmask, &set->sig, _SIGMASK_COPY_SIZE32);
+	sc.sregs = (__u32)(unsigned long __force) &frame->sregs;
+	if (__copy_to_user(&frame->sc, &sc, sizeof(frame->sc)))
 		return -EFAULT;
 
+	/* Store registers needed to create the signal frame */
+	store_sigregs();
+
+	/* Create _sigregs32 on the signal stack */
 	if (save_sigregs32(regs, &frame->sregs))
 		return -EFAULT;
-	if (save_sigregs_gprs_high(regs, frame->gprs_high))
+
+	/* Place signal number on stack to allow backtrace from handler.  */
+	if (__put_user(regs->gprs[2], (int __force __user *) &frame->signo))
 		return -EFAULT;
-	if (__put_user((unsigned long) &frame->sregs, &frame->sc.sregs))
+
+	/* Create _sigregs_ext32 on the signal stack */
+	if (save_sigregs_ext32(regs, &frame->sregs_ext))
 		return -EFAULT;
 
 	/* Set up to return from userspace.  If provided, use a stub
 	   already in userspace.  */
 	if (ksig->ka.sa.sa_flags & SA_RESTORER) {
-		regs->gprs[14] = (__u64 __force) ksig->ka.sa.sa_restorer | PSW32_ADDR_AMODE;
+		restorer = (unsigned long __force)
+			ksig->ka.sa.sa_restorer | PSW32_ADDR_AMODE;
 	} else {
-		regs->gprs[14] = (__u64 __force) frame->retcode | PSW32_ADDR_AMODE;
-		if (__put_user(S390_SYSCALL_OPCODE | __NR_sigreturn,
-			       (u16 __force __user *)(frame->retcode)))
+		/* Signal frames without vectors registers are short ! */
+		__u16 __user *svc = (void *) frame + frame_size - 2;
+		if (__put_user(S390_SYSCALL_OPCODE | __NR_sigreturn, svc))
 			return -EFAULT;
+		restorer = (unsigned long __force) svc | PSW32_ADDR_AMODE;
         }
 
-	/* Set up backchain. */
-	if (__put_user(regs->gprs[15], (unsigned int __user *) frame))
-		return -EFAULT;
-
 	/* Set up registers for signal handler */
+	regs->gprs[14] = restorer;
 	regs->gprs[15] = (__force __u64) frame;
 	/* Force 31 bit amode and default user address space control. */
 	regs->psw.mask = PSW_MASK_BA |
@@ -375,50 +462,69 @@ static int setup_frame32(struct ksignal *ksig, sigset_t *set,
 		regs->gprs[6] = task_thread_info(current)->last_break;
 	}
 
-	/* Place signal number on stack to allow backtrace from handler.  */
-	if (__put_user(regs->gprs[2], (int __force __user *) &frame->signo))
-		return -EFAULT;
 	return 0;
 }
 
 static int setup_rt_frame32(struct ksignal *ksig, sigset_t *set,
 			    struct pt_regs *regs)
 {
-	int err = 0;
-	rt_sigframe32 __user *frame = get_sigframe(&ksig->ka, regs, sizeof(rt_sigframe32));
-
+	rt_sigframe32 __user *frame;
+	unsigned long restorer;
+	size_t frame_size;
+	u32 uc_flags;
+
+	frame_size = sizeof(*frame) -
+		     sizeof(frame->uc.uc_mcontext_ext.__reserved);
+	/*
+	 * gprs_high are always present for 31-bit compat tasks.
+	 * The space for vector registers is only allocated if
+	 * the machine supports it
+	 */
+	uc_flags = UC_GPRS_HIGH;
+	if (MACHINE_HAS_VX) {
+		if (current->thread.vxrs)
+			uc_flags |= UC_VXRS;
+	} else
+		frame_size -= sizeof(frame->uc.uc_mcontext_ext.vxrs_low) +
+			      sizeof(frame->uc.uc_mcontext_ext.vxrs_high);
+	frame = get_sigframe(&ksig->ka, regs, frame_size);
 	if (frame == (void __user *) -1UL)
 		return -EFAULT;
 
-	if (copy_siginfo_to_user32(&frame->info, &ksig->info))
-		return -EFAULT;
-
-	/* Create the ucontext.  */
-	err |= __put_user(UC_EXTENDED, &frame->uc.uc_flags);
-	err |= __put_user(0, &frame->uc.uc_link);
-	err |= __compat_save_altstack(&frame->uc.uc_stack, regs->gprs[15]);
-	err |= save_sigregs32(regs, &frame->uc.uc_mcontext);
-	err |= save_sigregs_gprs_high(regs, frame->gprs_high);
-	err |= __copy_to_user(&frame->uc.uc_sigmask, set, sizeof(*set));
-	if (err)
+	/* Set up backchain. */
+	if (__put_user(regs->gprs[15], (unsigned int __force __user *) frame))
 		return -EFAULT;
 
 	/* Set up to return from userspace.  If provided, use a stub
 	   already in userspace.  */
 	if (ksig->ka.sa.sa_flags & SA_RESTORER) {
-		regs->gprs[14] = (__u64 __force) ksig->ka.sa.sa_restorer | PSW32_ADDR_AMODE;
+		restorer = (unsigned long __force)
+			ksig->ka.sa.sa_restorer | PSW32_ADDR_AMODE;
 	} else {
-		regs->gprs[14] = (__u64 __force) frame->retcode | PSW32_ADDR_AMODE;
-		if (__put_user(S390_SYSCALL_OPCODE | __NR_rt_sigreturn,
-			       (u16 __force __user *)(frame->retcode)))
+		__u16 __user *svc = &frame->svc_insn;
+		if (__put_user(S390_SYSCALL_OPCODE | __NR_rt_sigreturn, svc))
 			return -EFAULT;
+		restorer = (unsigned long __force) svc | PSW32_ADDR_AMODE;
 	}
 
-	/* Set up backchain. */
-	if (__put_user(regs->gprs[15], (unsigned int __force __user *) frame))
+	/* Create siginfo on the signal stack */
+	if (copy_siginfo_to_user32(&frame->info, &ksig->info))
+		return -EFAULT;
+
+	/* Store registers needed to create the signal frame */
+	store_sigregs();
+
+	/* Create ucontext on the signal stack. */
+	if (__put_user(uc_flags, &frame->uc.uc_flags) ||
+	    __put_user(0, &frame->uc.uc_link) ||
+	    __compat_save_altstack(&frame->uc.uc_stack, regs->gprs[15]) ||
+	    save_sigregs32(regs, &frame->uc.uc_mcontext) ||
+	    __copy_to_user(&frame->uc.uc_sigmask, set, sizeof(*set)) ||
+	    save_sigregs_ext32(regs, &frame->uc.uc_mcontext_ext))
 		return -EFAULT;
 
 	/* Set up registers for signal handler */
+	regs->gprs[14] = restorer;
 	regs->gprs[15] = (__force __u64) frame;
 	/* Force 31 bit amode and default user address space control. */
 	regs->psw.mask = PSW_MASK_BA |

arch/s390/kernel/early.c

@@ -392,6 +392,8 @@ static __init void detect_machine_facilities(void)
 		S390_lowcore.machine_flags |= MACHINE_FLAG_TE;
 	if (test_facility(51))
 		S390_lowcore.machine_flags |= MACHINE_FLAG_TLB_LC;
+	if (test_facility(129))
+		S390_lowcore.machine_flags |= MACHINE_FLAG_VX;
 #endif
 }
 

arch/s390/kernel/entry.h

@@ -21,6 +21,8 @@ void psw_idle(struct s390_idle_data *, unsigned long);
 asmlinkage long do_syscall_trace_enter(struct pt_regs *regs);
 asmlinkage void do_syscall_trace_exit(struct pt_regs *regs);
 
+int alloc_vector_registers(struct task_struct *tsk);
+
 void do_protection_exception(struct pt_regs *regs);
 void do_dat_exception(struct pt_regs *regs);
 
@@ -43,6 +45,7 @@ void special_op_exception(struct pt_regs *regs);
 void specification_exception(struct pt_regs *regs);
 void transaction_exception(struct pt_regs *regs);
 void translation_exception(struct pt_regs *regs);
+void vector_exception(struct pt_regs *regs);
 
 void do_per_trap(struct pt_regs *regs);
 void do_report_trap(struct pt_regs *regs, int si_signo, int si_code, char *str);

arch/s390/kernel/nmi.c

@@ -20,6 +20,7 @@
 #include <asm/cputime.h>
 #include <asm/nmi.h>
 #include <asm/crw.h>
+#include <asm/switch_to.h>
 
 struct mcck_struct {
 	int kill_task;
@@ -163,6 +164,21 @@ static int notrace s390_revalidate_registers(struct mci *mci)
 			"	ld	15,120(%0)\n"
 			: : "a" (fpt_save_area));
 	}
+
+#ifdef CONFIG_64BIT
+	/* Revalidate vector registers */
+	if (MACHINE_HAS_VX && current->thread.vxrs) {
+		if (!mci->vr) {
+			/*
+			 * Vector registers can't be restored and therefore
+			 * the process needs to be terminated.
+			 */
+			kill_task = 1;
+		}
+		restore_vx_regs((__vector128 *)
+				S390_lowcore.vector_save_area_addr);
+	}
+#endif
 	/* Revalidate access registers */
 	asm volatile(
 		"	lam	0,15,0(%0)"

arch/s390/kernel/pgm_check.S

@@ -49,7 +49,7 @@ PGM_CHECK_DEFAULT			/* 17 */
 PGM_CHECK_64BIT(transaction_exception)	/* 18 */
 PGM_CHECK_DEFAULT			/* 19 */
 PGM_CHECK_DEFAULT			/* 1a */
-PGM_CHECK_DEFAULT			/* 1b */
+PGM_CHECK_64BIT(vector_exception)	/* 1b */
 PGM_CHECK(space_switch_exception)	/* 1c */
 PGM_CHECK(hfp_sqrt_exception)		/* 1d */
 PGM_CHECK_DEFAULT			/* 1e */

arch/s390/kernel/processor.c

@@ -39,7 +39,7 @@ static int show_cpuinfo(struct seq_file *m, void *v)
 {
 	static const char *hwcap_str[] = {
 		"esan3", "zarch", "stfle", "msa", "ldisp", "eimm", "dfp",
-		"edat", "etf3eh", "highgprs", "te"
+		"edat", "etf3eh", "highgprs", "te", "vx"
 	};
 	unsigned long n = (unsigned long) v - 1;
 	int i;

arch/s390/kernel/ptrace.c

@@ -38,15 +38,6 @@
 #define CREATE_TRACE_POINTS
 #include <trace/events/syscalls.h>
 
-enum s390_regset {
-	REGSET_GENERAL,
-	REGSET_FP,
-	REGSET_LAST_BREAK,
-	REGSET_TDB,
-	REGSET_SYSTEM_CALL,
-	REGSET_GENERAL_EXTENDED,
-};
-
 void update_cr_regs(struct task_struct *task)
 {
 	struct pt_regs *regs = task_pt_regs(task);
@@ -55,27 +46,39 @@ void update_cr_regs(struct task_struct *task)
 
 #ifdef CONFIG_64BIT
 	/* Take care of the enable/disable of transactional execution. */
-	if (MACHINE_HAS_TE) {
+	if (MACHINE_HAS_TE || MACHINE_HAS_VX) {
 		unsigned long cr, cr_new;
 
 		__ctl_store(cr, 0, 0);
-		/* Set or clear transaction execution TXC bit 8. */
-		cr_new = cr | (1UL << 55);
-		if (task->thread.per_flags & PER_FLAG_NO_TE)
-			cr_new &= ~(1UL << 55);
+		cr_new = cr;
+		if (MACHINE_HAS_TE) {
+			/* Set or clear transaction execution TXC bit 8. */
+			cr_new |= (1UL << 55);
+			if (task->thread.per_flags & PER_FLAG_NO_TE)
+				cr_new &= ~(1UL << 55);
+		}
+		if (MACHINE_HAS_VX) {
+			/* Enable/disable of vector extension */
+			cr_new &= ~(1UL << 17);
+			if (task->thread.vxrs)
+				cr_new |= (1UL << 17);
+		}
 		if (cr_new != cr)
 			__ctl_load(cr_new, 0, 0);
-		/* Set or clear transaction execution TDC bits 62 and 63. */
-		__ctl_store(cr, 2, 2);
-		cr_new = cr & ~3UL;
-		if (task->thread.per_flags & PER_FLAG_TE_ABORT_RAND) {
-			if (task->thread.per_flags & PER_FLAG_TE_ABORT_RAND_TEND)
-				cr_new |= 1UL;
-			else
-				cr_new |= 2UL;
+		if (MACHINE_HAS_TE) {
+			/* Set/clear transaction execution TDC bits 62/63. */
+			__ctl_store(cr, 2, 2);
+			cr_new = cr & ~3UL;
+			if (task->thread.per_flags & PER_FLAG_TE_ABORT_RAND) {
+				if (task->thread.per_flags &
+				    PER_FLAG_TE_ABORT_RAND_TEND)
+					cr_new |= 1UL;
+				else
+					cr_new |= 2UL;
+			}
+			if (cr_new != cr)
+				__ctl_load(cr_new, 2, 2);
 		}
-		if (cr_new != cr)
-			__ctl_load(cr_new, 2, 2);
 	}
 #endif
 	/* Copy user specified PER registers */
@@ -926,7 +929,15 @@ static int s390_fpregs_get(struct task_struct *target,
 		save_fp_ctl(&target->thread.fp_regs.fpc);
 		save_fp_regs(target->thread.fp_regs.fprs);
 	}
+#ifdef CONFIG_64BIT
+	else if (target->thread.vxrs) {
+		int i;
 
+		for (i = 0; i < __NUM_VXRS_LOW; i++)
+			target->thread.fp_regs.fprs[i] =
+				*(freg_t *)(target->thread.vxrs + i);
+	}
+#endif
 	return user_regset_copyout(&pos, &count, &kbuf, &ubuf,
 				   &target->thread.fp_regs, 0, -1);
 }
@@ -960,9 +971,20 @@ static int s390_fpregs_set(struct task_struct *target,
 					target->thread.fp_regs.fprs,
 					offsetof(s390_fp_regs, fprs), -1);
 
-	if (rc == 0 && target == current) {
-		restore_fp_ctl(&target->thread.fp_regs.fpc);
-		restore_fp_regs(target->thread.fp_regs.fprs);
+	if (rc == 0) {
+		if (target == current) {
+			restore_fp_ctl(&target->thread.fp_regs.fpc);
+			restore_fp_regs(target->thread.fp_regs.fprs);
+		}
+#ifdef CONFIG_64BIT
+		else if (target->thread.vxrs) {
+			int i;
+
+			for (i = 0; i < __NUM_VXRS_LOW; i++)
+				*(freg_t *)(target->thread.vxrs + i) =
+					target->thread.fp_regs.fprs[i];
+		}
+#endif
 	}
 
 	return rc;
@@ -1018,6 +1040,95 @@ static int s390_tdb_set(struct task_struct *target,
 	return 0;
 }
 
+static int s390_vxrs_active(struct task_struct *target,
+			      const struct user_regset *regset)
+{
+	return !!target->thread.vxrs;
+}
+
+static int s390_vxrs_low_get(struct task_struct *target,
+			     const struct user_regset *regset,
+			     unsigned int pos, unsigned int count,
+			     void *kbuf, void __user *ubuf)
+{
+	__u64 vxrs[__NUM_VXRS_LOW];
+	int i;
+
+	if (target->thread.vxrs) {
+		if (target == current)
+			save_vx_regs(target->thread.vxrs);
+		for (i = 0; i < __NUM_VXRS_LOW; i++)
+			vxrs[i] = *((__u64 *)(target->thread.vxrs + i) + 1);
+	} else
+		memset(vxrs, 0, sizeof(vxrs));
+	return user_regset_copyout(&pos, &count, &kbuf, &ubuf, vxrs, 0, -1);
+}
+
+static int s390_vxrs_low_set(struct task_struct *target,
+			     const struct user_regset *regset,
+			     unsigned int pos, unsigned int count,
+			     const void *kbuf, const void __user *ubuf)
+{
+	__u64 vxrs[__NUM_VXRS_LOW];
+	int i, rc;
+
+	if (!target->thread.vxrs) {
+		rc = alloc_vector_registers(target);
+		if (rc)
+			return rc;
+	} else if (target == current)
+		save_vx_regs(target->thread.vxrs);
+
+	rc = user_regset_copyin(&pos, &count, &kbuf, &ubuf, vxrs, 0, -1);
+	if (rc == 0) {
+		for (i = 0; i < __NUM_VXRS_LOW; i++)
+			*((__u64 *)(target->thread.vxrs + i) + 1) = vxrs[i];
+		if (target == current)
+			restore_vx_regs(target->thread.vxrs);
+	}
+
+	return rc;
+}
+
+static int s390_vxrs_high_get(struct task_struct *target,
+			      const struct user_regset *regset,
+			      unsigned int pos, unsigned int count,
+			      void *kbuf, void __user *ubuf)
+{
+	__vector128 vxrs[__NUM_VXRS_HIGH];
+
+	if (target->thread.vxrs) {
+		if (target == current)
+			save_vx_regs(target->thread.vxrs);
+		memcpy(vxrs, target->thread.vxrs + __NUM_VXRS_LOW,
+		       sizeof(vxrs));
+	} else
+		memset(vxrs, 0, sizeof(vxrs));
+	return user_regset_copyout(&pos, &count, &kbuf, &ubuf, vxrs, 0, -1);
+}
+
+static int s390_vxrs_high_set(struct task_struct *target,
+			      const struct user_regset *regset,
+			      unsigned int pos, unsigned int count,
+			      const void *kbuf, const void __user *ubuf)
+{
+	int rc;
+
+	if (!target->thread.vxrs) {
+		rc = alloc_vector_registers(target);
+		if (rc)
+			return rc;
+	} else if (target == current)
+		save_vx_regs(target->thread.vxrs);
+
+	rc = user_regset_copyin(&pos, &count, &kbuf, &ubuf,
+				target->thread.vxrs + __NUM_VXRS_LOW, 0, -1);
+	if (rc == 0 && target == current)
+		restore_vx_regs(target->thread.vxrs);
+
+	return rc;
+}
+
 #endif
 
 static int s390_system_call_get(struct task_struct *target,
@@ -1041,7 +1152,7 @@ static int s390_system_call_set(struct task_struct *target,
 }
 
 static const struct user_regset s390_regsets[] = {
-	[REGSET_GENERAL] = {
+	{
 		.core_note_type = NT_PRSTATUS,
 		.n = sizeof(s390_regs) / sizeof(long),
 		.size = sizeof(long),
@@ -1049,7 +1160,7 @@ static const struct user_regset s390_regsets[] = {
 		.get = s390_regs_get,
 		.set = s390_regs_set,
 	},
-	[REGSET_FP] = {
+	{
 		.core_note_type = NT_PRFPREG,
 		.n = sizeof(s390_fp_regs) / sizeof(long),
 		.size = sizeof(long),
@@ -1057,8 +1168,16 @@ static const struct user_regset s390_regsets[] = {
 		.get = s390_fpregs_get,
 		.set = s390_fpregs_set,
 	},
+	{
+		.core_note_type = NT_S390_SYSTEM_CALL,
+		.n = 1,
+		.size = sizeof(unsigned int),
+		.align = sizeof(unsigned int),
+		.get = s390_system_call_get,
+		.set = s390_system_call_set,
+	},
 #ifdef CONFIG_64BIT
-	[REGSET_LAST_BREAK] = {
+	{
 		.core_note_type = NT_S390_LAST_BREAK,
 		.n = 1,
 		.size = sizeof(long),
@@ -1066,7 +1185,7 @@ static const struct user_regset s390_regsets[] = {
 		.get = s390_last_break_get,
 		.set = s390_last_break_set,
 	},
-	[REGSET_TDB] = {
+	{
 		.core_note_type = NT_S390_TDB,
 		.n = 1,
 		.size = 256,
@@ -1074,15 +1193,25 @@ static const struct user_regset s390_regsets[] = {
 		.get = s390_tdb_get,
 		.set = s390_tdb_set,
 	},
-#endif
-	[REGSET_SYSTEM_CALL] = {
-		.core_note_type = NT_S390_SYSTEM_CALL,
-		.n = 1,
-		.size = sizeof(unsigned int),
-		.align = sizeof(unsigned int),
-		.get = s390_system_call_get,
-		.set = s390_system_call_set,
+	{
+		.core_note_type = NT_S390_VXRS_LOW,
+		.n = __NUM_VXRS_LOW,
+		.size = sizeof(__u64),
+		.align = sizeof(__u64),
+		.active = s390_vxrs_active,
+		.get = s390_vxrs_low_get,
+		.set = s390_vxrs_low_set,
+	},
+	{
+		.core_note_type = NT_S390_VXRS_HIGH,
+		.n = __NUM_VXRS_HIGH,
+		.size = sizeof(__vector128),
+		.align = sizeof(__vector128),
+		.active = s390_vxrs_active,
+		.get = s390_vxrs_high_get,
+		.set = s390_vxrs_high_set,
 	},
+#endif
 };
 
 static const struct user_regset_view user_s390_view = {
@@ -1247,7 +1376,7 @@ static int s390_compat_last_break_set(struct task_struct *target,
 }
 
 static const struct user_regset s390_compat_regsets[] = {
-	[REGSET_GENERAL] = {
+	{
 		.core_note_type = NT_PRSTATUS,
 		.n = sizeof(s390_compat_regs) / sizeof(compat_long_t),
 		.size = sizeof(compat_long_t),
@@ -1255,7 +1384,7 @@ static const struct user_regset s390_compat_regsets[] = {
 		.get = s390_compat_regs_get,
 		.set = s390_compat_regs_set,
 	},
-	[REGSET_FP] = {
+	{
 		.core_note_type = NT_PRFPREG,
 		.n = sizeof(s390_fp_regs) / sizeof(compat_long_t),
 		.size = sizeof(compat_long_t),
@@ -1263,7 +1392,15 @@ static const struct user_regset s390_compat_regsets[] = {
 		.get = s390_fpregs_get,
 		.set = s390_fpregs_set,
 	},
-	[REGSET_LAST_BREAK] = {
+	{
+		.core_note_type = NT_S390_SYSTEM_CALL,
+		.n = 1,
+		.size = sizeof(compat_uint_t),
+		.align = sizeof(compat_uint_t),
+		.get = s390_system_call_get,
+		.set = s390_system_call_set,
+	},
+	{
 		.core_note_type = NT_S390_LAST_BREAK,
 		.n = 1,
 		.size = sizeof(long),
@@ -1271,7 +1408,7 @@ static const struct user_regset s390_compat_regsets[] = {
 		.get = s390_compat_last_break_get,
 		.set = s390_compat_last_break_set,
 	},
-	[REGSET_TDB] = {
+	{
 		.core_note_type = NT_S390_TDB,
 		.n = 1,
 		.size = 256,
@@ -1279,15 +1416,25 @@ static const struct user_regset s390_compat_regsets[] = {
 		.get = s390_tdb_get,
 		.set = s390_tdb_set,
 	},
-	[REGSET_SYSTEM_CALL] = {
-		.core_note_type = NT_S390_SYSTEM_CALL,
-		.n = 1,
-		.size = sizeof(compat_uint_t),
-		.align = sizeof(compat_uint_t),
-		.get = s390_system_call_get,
-		.set = s390_system_call_set,
+	{
+		.core_note_type = NT_S390_VXRS_LOW,
+		.n = __NUM_VXRS_LOW,
+		.size = sizeof(__u64),
+		.align = sizeof(__u64),
+		.active = s390_vxrs_active,
+		.get = s390_vxrs_low_get,
+		.set = s390_vxrs_low_set,
+	},
+	{
+		.core_note_type = NT_S390_VXRS_HIGH,
+		.n = __NUM_VXRS_HIGH,
+		.size = sizeof(__vector128),
+		.align = sizeof(__vector128),
+		.active = s390_vxrs_active,
+		.get = s390_vxrs_high_get,
+		.set = s390_vxrs_high_set,
 	},
-	[REGSET_GENERAL_EXTENDED] = {
+	{
 		.core_note_type = NT_S390_HIGH_GPRS,
 		.n = sizeof(s390_compat_regs_high) / sizeof(compat_long_t),
 		.size = sizeof(compat_long_t),

arch/s390/kernel/setup.c

@@ -343,6 +343,9 @@ static void __init setup_lowcore(void)
 		__ctl_set_bit(14, 29);
 	}
 #else
+	if (MACHINE_HAS_VX)
+		lc->vector_save_area_addr =
+			(unsigned long) &lc->vector_save_area;
 	lc->vdso_per_cpu_data = (unsigned long) &lc->paste[0];
 #endif
 	lc->sync_enter_timer = S390_lowcore.sync_enter_timer;
@@ -765,6 +768,12 @@ static void __init setup_hwcaps(void)
 	 */
 	if (test_facility(50) && test_facility(73))
 		elf_hwcap |= HWCAP_S390_TE;
+
+	/*
+	 * Vector extension HWCAP_S390_VXRS is bit 11.
+	 */
+	if (test_facility(129))
+		elf_hwcap |= HWCAP_S390_VXRS;
 #endif
 
 	get_cpu_id(&cpu_id);

arch/s390/kernel/signal.c

@@ -31,30 +31,117 @@
 #include <asm/switch_to.h>
 #include "entry.h"
 
-typedef struct 
+/*
+ * Layout of an old-style signal-frame:
+ *	-----------------------------------------
+ *	| save area (_SIGNAL_FRAMESIZE)		|
+ *	-----------------------------------------
+ *	| struct sigcontext			|
+ *	|	oldmask				|
+ *	|	_sigregs *			|
+ *	-----------------------------------------
+ *	| _sigregs with				|
+ *	|	_s390_regs_common		|
+ *	|	_s390_fp_regs			|
+ *	-----------------------------------------
+ *	| int signo				|
+ *	-----------------------------------------
+ *	| _sigregs_ext with			|
+ *	|	gprs_high 64 byte (opt)		|
+ *	|	vxrs_low 128 byte (opt)		|
+ *	|	vxrs_high 256 byte (opt)	|
+ *	|	reserved 128 byte (opt)		|
+ *	-----------------------------------------
+ *	| __u16 svc_insn			|
+ *	-----------------------------------------
+ * The svc_insn entry with the sigreturn system call opcode does not
+ * have a fixed position and moves if gprs_high or vxrs exist.
+ * Future extensions will be added to _sigregs_ext.
+ */
+struct sigframe
 {
 	__u8 callee_used_stack[__SIGNAL_FRAMESIZE];
 	struct sigcontext sc;
 	_sigregs sregs;
 	int signo;
-	__u8 retcode[S390_SYSCALL_SIZE];
-} sigframe;
+	_sigregs_ext sregs_ext;
+	__u16 svc_insn;		/* Offset of svc_insn is NOT fixed! */
+};
 
-typedef struct 
+/*
+ * Layout of an rt signal-frame:
+ *	-----------------------------------------
+ *	| save area (_SIGNAL_FRAMESIZE)		|
+ *	-----------------------------------------
+ *	| svc __NR_rt_sigreturn 2 byte		|
+ *	-----------------------------------------
+ *	| struct siginfo			|
+ *	-----------------------------------------
+ *	| struct ucontext_extended with		|
+ *	|	unsigned long uc_flags		|
+ *	|	struct ucontext *uc_link	|
+ *	|	stack_t uc_stack		|
+ *	|	_sigregs uc_mcontext with	|
+ *	|		_s390_regs_common	|
+ *	|		_s390_fp_regs		|
+ *	|	sigset_t uc_sigmask		|
+ *	|	_sigregs_ext uc_mcontext_ext	|
+ *	|		gprs_high 64 byte (opt)	|
+ *	|		vxrs_low 128 byte (opt)	|
+ *	|		vxrs_high 256 byte (opt)|
+ *	|		reserved 128 byte (opt)	|
+ *	-----------------------------------------
+ * Future extensions will be added to _sigregs_ext.
+ */
+struct rt_sigframe
 {
 	__u8 callee_used_stack[__SIGNAL_FRAMESIZE];
-	__u8 retcode[S390_SYSCALL_SIZE];
+	__u16 svc_insn;
 	struct siginfo info;
-	struct ucontext uc;
-} rt_sigframe;
+	struct ucontext_extended uc;
+};
+
+/* Store registers needed to create the signal frame */
+static void store_sigregs(void)
+{
+	save_access_regs(current->thread.acrs);
+	save_fp_ctl(&current->thread.fp_regs.fpc);
+#ifdef CONFIG_64BIT
+	if (current->thread.vxrs) {
+		int i;
+
+		save_vx_regs(current->thread.vxrs);
+		for (i = 0; i < __NUM_FPRS; i++)
+			current->thread.fp_regs.fprs[i] =
+				*(freg_t *)(current->thread.vxrs + i);
+	} else
+#endif
+		save_fp_regs(current->thread.fp_regs.fprs);
+}
+
+/* Load registers after signal return */
+static void load_sigregs(void)
+{
+	restore_access_regs(current->thread.acrs);
+	/* restore_fp_ctl is done in restore_sigregs */
+#ifdef CONFIG_64BIT
+	if (current->thread.vxrs) {
+		int i;
+
+		for (i = 0; i < __NUM_FPRS; i++)
+			*(freg_t *)(current->thread.vxrs + i) =
+				current->thread.fp_regs.fprs[i];
+		restore_vx_regs(current->thread.vxrs);
+	} else
+#endif
+		restore_fp_regs(current->thread.fp_regs.fprs);
+}
 
 /* Returns non-zero on fault. */
 static int save_sigregs(struct pt_regs *regs, _sigregs __user *sregs)
 {
 	_sigregs user_sregs;
 
-	save_access_regs(current->thread.acrs);
-
 	/* Copy a 'clean' PSW mask to the user to avoid leaking
 	   information about whether PER is currently on.  */
 	user_sregs.regs.psw.mask = PSW_USER_BITS |
@@ -63,12 +150,6 @@ static int save_sigregs(struct pt_regs *regs, _sigregs __user *sregs)
 	memcpy(&user_sregs.regs.gprs, &regs->gprs, sizeof(sregs->regs.gprs));
 	memcpy(&user_sregs.regs.acrs, current->thread.acrs,
 	       sizeof(user_sregs.regs.acrs));
-	/* 
-	 * We have to store the fp registers to current->thread.fp_regs
-	 * to merge them with the emulated registers.
-	 */
-	save_fp_ctl(&current->thread.fp_regs.fpc);
-	save_fp_regs(current->thread.fp_regs.fprs);
 	memcpy(&user_sregs.fpregs, &current->thread.fp_regs,
 	       sizeof(user_sregs.fpregs));
 	if (__copy_to_user(sregs, &user_sregs, sizeof(_sigregs)))
@@ -107,20 +188,64 @@ static int restore_sigregs(struct pt_regs *regs, _sigregs __user *sregs)
 	memcpy(&regs->gprs, &user_sregs.regs.gprs, sizeof(sregs->regs.gprs));
 	memcpy(&current->thread.acrs, &user_sregs.regs.acrs,
 	       sizeof(current->thread.acrs));
-	restore_access_regs(current->thread.acrs);
 
 	memcpy(&current->thread.fp_regs, &user_sregs.fpregs,
 	       sizeof(current->thread.fp_regs));
 
-	restore_fp_regs(current->thread.fp_regs.fprs);
 	clear_pt_regs_flag(regs, PIF_SYSCALL); /* No longer in a system call */
 	return 0;
 }
 
+/* Returns non-zero on fault. */
+static int save_sigregs_ext(struct pt_regs *regs,
+			    _sigregs_ext __user *sregs_ext)
+{
+#ifdef CONFIG_64BIT
+	__u64 vxrs[__NUM_VXRS_LOW];
+	int i;
+
+	/* Save vector registers to signal stack */
+	if (current->thread.vxrs) {
+		for (i = 0; i < __NUM_VXRS_LOW; i++)
+			vxrs[i] = *((__u64 *)(current->thread.vxrs + i) + 1);
+		if (__copy_to_user(&sregs_ext->vxrs_low, vxrs,
+				   sizeof(sregs_ext->vxrs_low)) ||
+		    __copy_to_user(&sregs_ext->vxrs_high,
+				   current->thread.vxrs + __NUM_VXRS_LOW,
+				   sizeof(sregs_ext->vxrs_high)))
+			return -EFAULT;
+	}
+#endif
+	return 0;
+}
+
+static int restore_sigregs_ext(struct pt_regs *regs,
+			       _sigregs_ext __user *sregs_ext)
+{
+#ifdef CONFIG_64BIT
+	__u64 vxrs[__NUM_VXRS_LOW];
+	int i;
+
+	/* Restore vector registers from signal stack */
+	if (current->thread.vxrs) {
+		if (__copy_from_user(vxrs, &sregs_ext->vxrs_low,
+				     sizeof(sregs_ext->vxrs_low)) ||
+		    __copy_from_user(current->thread.vxrs + __NUM_VXRS_LOW,
+				     &sregs_ext->vxrs_high,
+				     sizeof(sregs_ext->vxrs_high)))
+			return -EFAULT;
+		for (i = 0; i < __NUM_VXRS_LOW; i++)
+			*((__u64 *)(current->thread.vxrs + i) + 1) = vxrs[i];
+	}
+#endif
+	return 0;
+}
+
 SYSCALL_DEFINE0(sigreturn)
 {
 	struct pt_regs *regs = task_pt_regs(current);
-	sigframe __user *frame = (sigframe __user *)regs->gprs[15];
+	struct sigframe __user *frame =
+		(struct sigframe __user *) regs->gprs[15];
 	sigset_t set;
 
 	if (__copy_from_user(&set.sig, &frame->sc.oldmask, _SIGMASK_COPY_SIZE))
@@ -128,6 +253,9 @@ SYSCALL_DEFINE0(sigreturn)
 	set_current_blocked(&set);
 	if (restore_sigregs(regs, &frame->sregs))
 		goto badframe;
+	if (restore_sigregs_ext(regs, &frame->sregs_ext))
+		goto badframe;
+	load_sigregs();
 	return regs->gprs[2];
 badframe:
 	force_sig(SIGSEGV, current);
@@ -137,27 +265,26 @@ SYSCALL_DEFINE0(sigreturn)
 SYSCALL_DEFINE0(rt_sigreturn)
 {
 	struct pt_regs *regs = task_pt_regs(current);
-	rt_sigframe __user *frame = (rt_sigframe __user *)regs->gprs[15];
+	struct rt_sigframe __user *frame =
+		(struct rt_sigframe __user *)regs->gprs[15];
 	sigset_t set;
 
 	if (__copy_from_user(&set.sig, &frame->uc.uc_sigmask, sizeof(set)))
 		goto badframe;
 	set_current_blocked(&set);
+	if (restore_altstack(&frame->uc.uc_stack))
+		goto badframe;
 	if (restore_sigregs(regs, &frame->uc.uc_mcontext))
 		goto badframe;
-	if (restore_altstack(&frame->uc.uc_stack))
+	if (restore_sigregs_ext(regs, &frame->uc.uc_mcontext_ext))
 		goto badframe;
+	load_sigregs();
 	return regs->gprs[2];
 badframe:
 	force_sig(SIGSEGV, current);
 	return 0;
 }
 
-/*
- * Set up a signal frame.
- */
-
-
 /*
  * Determine which stack to use..
  */
@@ -195,39 +322,63 @@ static inline int map_signal(int sig)
 static int setup_frame(int sig, struct k_sigaction *ka,
 		       sigset_t *set, struct pt_regs * regs)
 {
-	sigframe __user *frame;
-
-	frame = get_sigframe(ka, regs, sizeof(sigframe));
+	struct sigframe __user *frame;
+	struct sigcontext sc;
+	unsigned long restorer;
+	size_t frame_size;
 
+	/*
+	 * gprs_high are only present for a 31-bit task running on
+	 * a 64-bit kernel (see compat_signal.c) but the space for
+	 * gprs_high need to be allocated if vector registers are
+	 * included in the signal frame on a 31-bit system.
+	 */
+	frame_size = sizeof(*frame) - sizeof(frame->sregs_ext);
+	if (MACHINE_HAS_VX)
+		frame_size += sizeof(frame->sregs_ext);
+	frame = get_sigframe(ka, regs, frame_size);
 	if (frame == (void __user *) -1UL)
 		return -EFAULT;
 
-	if (__copy_to_user(&frame->sc.oldmask, &set->sig, _SIGMASK_COPY_SIZE))
+	/* Set up backchain. */
+	if (__put_user(regs->gprs[15], (addr_t __user *) frame))
 		return -EFAULT;
 
+	/* Create struct sigcontext on the signal stack */
+	memcpy(&sc.oldmask, &set->sig, _SIGMASK_COPY_SIZE);
+	sc.sregs = (_sigregs __user __force *) &frame->sregs;
+	if (__copy_to_user(&frame->sc, &sc, sizeof(frame->sc)))
+		return -EFAULT;
+
+	/* Store registers needed to create the signal frame */
+	store_sigregs();
+
+	/* Create _sigregs on the signal stack */
 	if (save_sigregs(regs, &frame->sregs))
 		return -EFAULT;
-	if (__put_user(&frame->sregs, &frame->sc.sregs))
+
+	/* Place signal number on stack to allow backtrace from handler.  */
+	if (__put_user(regs->gprs[2], (int __user *) &frame->signo))
+		return -EFAULT;
+
+	/* Create _sigregs_ext on the signal stack */
+	if (save_sigregs_ext(regs, &frame->sregs_ext))
 		return -EFAULT;
 
 	/* Set up to return from userspace.  If provided, use a stub
 	   already in userspace.  */
 	if (ka->sa.sa_flags & SA_RESTORER) {
-                regs->gprs[14] = (unsigned long)
-			ka->sa.sa_restorer | PSW_ADDR_AMODE;
+		restorer = (unsigned long) ka->sa.sa_restorer | PSW_ADDR_AMODE;
 	} else {
-                regs->gprs[14] = (unsigned long)
-			frame->retcode | PSW_ADDR_AMODE;
-		if (__put_user(S390_SYSCALL_OPCODE | __NR_sigreturn,
-	                       (u16 __user *)(frame->retcode)))
+		/* Signal frame without vector registers are short ! */
+		__u16 __user *svc = (void *) frame + frame_size - 2;
+		if (__put_user(S390_SYSCALL_OPCODE | __NR_sigreturn, svc))
 			return -EFAULT;
+		restorer = (unsigned long) svc | PSW_ADDR_AMODE;
 	}
 
-	/* Set up backchain. */
-	if (__put_user(regs->gprs[15], (addr_t __user *) frame))
-		return -EFAULT;
-
 	/* Set up registers for signal handler */
+	regs->gprs[14] = restorer;
 	regs->gprs[15] = (unsigned long) frame;
 	/* Force default amode and default user address space control. */
 	regs->psw.mask = PSW_MASK_EA | PSW_MASK_BA |
@@ -247,54 +398,69 @@ static int setup_frame(int sig, struct k_sigaction *ka,
 		regs->gprs[5] = regs->int_parm_long;
 		regs->gprs[6] = task_thread_info(current)->last_break;
 	}
-
-	/* Place signal number on stack to allow backtrace from handler.  */
-	if (__put_user(regs->gprs[2], (int __user *) &frame->signo))
-		return -EFAULT;
 	return 0;
 }
 
 static int setup_rt_frame(struct ksignal *ksig, sigset_t *set,
 			  struct pt_regs *regs)
 {
-	int err = 0;
-	rt_sigframe __user *frame;
-
-	frame = get_sigframe(&ksig->ka, regs, sizeof(rt_sigframe));
+	struct rt_sigframe __user *frame;
+	unsigned long uc_flags, restorer;
+	size_t frame_size;
 
+	frame_size = sizeof(struct rt_sigframe) - sizeof(_sigregs_ext);
+	/*
+	 * gprs_high are only present for a 31-bit task running on
+	 * a 64-bit kernel (see compat_signal.c) but the space for
+	 * gprs_high need to be allocated if vector registers are
+	 * included in the signal frame on a 31-bit system.
+	 */
+	uc_flags = 0;
+#ifdef CONFIG_64BIT
+	if (MACHINE_HAS_VX) {
+		frame_size += sizeof(_sigregs_ext);
+		if (current->thread.vxrs)
+			uc_flags |= UC_VXRS;
+	}
+#endif
+	frame = get_sigframe(&ksig->ka, regs, frame_size);
 	if (frame == (void __user *) -1UL)
 		return -EFAULT;
 
-	if (copy_siginfo_to_user(&frame->info, &ksig->info))
-		return -EFAULT;
-
-	/* Create the ucontext.  */
-	err |= __put_user(0, &frame->uc.uc_flags);
-	err |= __put_user(NULL, &frame->uc.uc_link);
-	err |= __save_altstack(&frame->uc.uc_stack, regs->gprs[15]);
-	err |= save_sigregs(regs, &frame->uc.uc_mcontext);
-	err |= __copy_to_user(&frame->uc.uc_sigmask, set, sizeof(*set));
-	if (err)
+	/* Set up backchain. */
+	if (__put_user(regs->gprs[15], (addr_t __user *) frame))
 		return -EFAULT;
 
 	/* Set up to return from userspace.  If provided, use a stub
 	   already in userspace.  */
 	if (ksig->ka.sa.sa_flags & SA_RESTORER) {
-                regs->gprs[14] = (unsigned long)
+		restorer = (unsigned long)
 			ksig->ka.sa.sa_restorer | PSW_ADDR_AMODE;
 	} else {
-                regs->gprs[14] = (unsigned long)
-			frame->retcode | PSW_ADDR_AMODE;
-		if (__put_user(S390_SYSCALL_OPCODE | __NR_rt_sigreturn,
-			       (u16 __user *)(frame->retcode)))
+		__u16 __user *svc = &frame->svc_insn;
+		if (__put_user(S390_SYSCALL_OPCODE | __NR_rt_sigreturn, svc))
 			return -EFAULT;
+		restorer = (unsigned long) svc | PSW_ADDR_AMODE;
 	}
 
-	/* Set up backchain. */
-	if (__put_user(regs->gprs[15], (addr_t __user *) frame))
+	/* Create siginfo on the signal stack */
+	if (copy_siginfo_to_user(&frame->info, &ksig->info))
+		return -EFAULT;
+
+	/* Store registers needed to create the signal frame */
+	store_sigregs();
+
+	/* Create ucontext on the signal stack. */
+	if (__put_user(uc_flags, &frame->uc.uc_flags) ||
+	    __put_user(NULL, &frame->uc.uc_link) ||
+	    __save_altstack(&frame->uc.uc_stack, regs->gprs[15]) ||
+	    save_sigregs(regs, &frame->uc.uc_mcontext) ||
+	    __copy_to_user(&frame->uc.uc_sigmask, set, sizeof(*set)) ||
+	    save_sigregs_ext(regs, &frame->uc.uc_mcontext_ext))
 		return -EFAULT;
 
 	/* Set up registers for signal handler */
+	regs->gprs[14] = restorer;
 	regs->gprs[15] = (unsigned long) frame;
 	/* Force default amode and default user address space control. */
 	regs->psw.mask = PSW_MASK_EA | PSW_MASK_BA |

arch/s390/kernel/smp.c

@@ -179,6 +179,9 @@ static int pcpu_alloc_lowcore(struct pcpu *pcpu, int cpu)
 			goto out;
 	}
 #else
+	if (MACHINE_HAS_VX)
+		lc->vector_save_area_addr =
+			(unsigned long) &lc->vector_save_area;
 	if (vdso_alloc_per_cpu(lc))
 		goto out;
 #endif

arch/s390/kernel/traps.c

@@ -18,6 +18,8 @@
 #include <linux/ptrace.h>
 #include <linux/sched.h>
 #include <linux/mm.h>
+#include <linux/slab.h>
+#include <asm/switch_to.h>
 #include "entry.h"
 
 int show_unhandled_signals = 1;
@@ -303,6 +305,74 @@ DO_ERROR_INFO(specification_exception, SIGILL, ILL_ILLOPN,
 	      "specification exception");
 #endif
 
+#ifdef CONFIG_64BIT
+int alloc_vector_registers(struct task_struct *tsk)
+{
+	__vector128 *vxrs;
+	int i;
+
+	/* Allocate vector register save area. */
+	vxrs = kzalloc(sizeof(__vector128) * __NUM_VXRS,
+		       GFP_KERNEL|__GFP_REPEAT);
+	if (!vxrs)
+		return -ENOMEM;
+	preempt_disable();
+	if (tsk == current)
+		save_fp_regs(tsk->thread.fp_regs.fprs);
+	/* Copy the 16 floating point registers */
+	for (i = 0; i < 16; i++)
+		*(freg_t *) &vxrs[i] = tsk->thread.fp_regs.fprs[i];
+	tsk->thread.vxrs = vxrs;
+	if (tsk == current) {
+		__ctl_set_bit(0, 17);
+		restore_vx_regs(vxrs);
+	}
+	preempt_enable();
+	return 0;
+}
+
+void vector_exception(struct pt_regs *regs)
+{
+	int si_code, vic;
+
+	if (!MACHINE_HAS_VX) {
+		do_trap(regs, SIGILL, ILL_ILLOPN, "illegal operation");
+		return;
+	}
+
+	/* get vector interrupt code from fpc */
+	asm volatile("stfpc %0" : "=m" (current->thread.fp_regs.fpc));
+	vic = (current->thread.fp_regs.fpc & 0xf00) >> 8;
+	switch (vic) {
+	case 1: /* invalid vector operation */
+		si_code = FPE_FLTINV;
+		break;
+	case 2: /* division by zero */
+		si_code = FPE_FLTDIV;
+		break;
+	case 3: /* overflow */
+		si_code = FPE_FLTOVF;
+		break;
+	case 4: /* underflow */
+		si_code = FPE_FLTUND;
+		break;
+	case 5:	/* inexact */
+		si_code = FPE_FLTRES;
+		break;
+	default: /* unknown cause */
+		si_code = 0;
+	}
+	do_trap(regs, SIGFPE, si_code, "vector exception");
+}
+
+static int __init disable_vector_extension(char *str)
+{
+	S390_lowcore.machine_flags &= ~MACHINE_FLAG_VX;
+	return 1;
+}
+__setup("novx", disable_vector_extension);
+#endif
+
 void data_exception(struct pt_regs *regs)
 {
 	__u16 __user *location;
@@ -368,6 +438,18 @@ void data_exception(struct pt_regs *regs)
                 }
         }
 #endif 
+#ifdef CONFIG_64BIT
+	/* Check for vector register enablement */
+	if (MACHINE_HAS_VX && !current->thread.vxrs &&
+	    (current->thread.fp_regs.fpc & FPC_DXC_MASK) == 0xfe00) {
+		alloc_vector_registers(current);
+		/* Vector data exception is suppressing, rewind psw. */
+		regs->psw.addr = __rewind_psw(regs->psw, regs->int_code >> 16);
+		clear_pt_regs_flag(regs, PIF_PER_TRAP);
+		return;
+	}
+#endif
+
 	if (current->thread.fp_regs.fpc & FPC_DXC_MASK)
 		signal = SIGFPE;
 	else

include/uapi/linux/elf.h

@@ -391,6 +391,8 @@ typedef struct elf64_shdr {
 #define NT_S390_LAST_BREAK	0x306	/* s390 breaking event address */
 #define NT_S390_SYSTEM_CALL	0x307	/* s390 system call restart data */
 #define NT_S390_TDB	0x308		/* s390 transaction diagnostic block */
+#define NT_S390_VXRS_LOW	0x309	/* s390 vector registers 0-15 upper half */
+#define NT_S390_VXRS_HIGH	0x30a	/* s390 vector registers 16-31 */
 #define NT_ARM_VFP	0x400		/* ARM VFP/NEON registers */
 #define NT_ARM_TLS	0x401		/* ARM TLS register */
 #define NT_ARM_HW_BREAK	0x402		/* ARM hardware breakpoint registers */