15 files changed, 246 insertions, 80 deletions

diff --git a/arch/sparc/kernel/entry.h b/arch/sparc/kernel/entry.h
index 0c218e4c0881..cc3c5cb47cda 100644
--- a/arch/sparc/kernel/entry.h
+++ b/arch/sparc/kernel/entry.h
@@ -59,6 +59,13 @@ struct popc_6insn_patch_entry {
 extern struct popc_6insn_patch_entry __popc_6insn_patch,
         __popc_6insn_patch_end;
 
+struct pause_patch_entry {
+        unsigned int    addr;
+        unsigned int    insns[3];
+};
+extern struct pause_patch_entry __pause_3insn_patch,
+        __pause_3insn_patch_end;
+
 extern void __init per_cpu_patch(void);
 extern void sun4v_patch_1insn_range(struct sun4v_1insn_patch_entry *,
                                     struct sun4v_1insn_patch_entry *);
diff --git a/arch/sparc/kernel/leon_kernel.c b/arch/sparc/kernel/leon_kernel.c
index f8b6eee40bde..87f60ee65433 100644
--- a/arch/sparc/kernel/leon_kernel.c
+++ b/arch/sparc/kernel/leon_kernel.c
@@ -56,11 +56,13 @@ static inline unsigned int leon_eirq_get(int cpu)
 static void leon_handle_ext_irq(unsigned int irq, struct irq_desc *desc)
 {
         unsigned int eirq;
+        struct irq_bucket *p;
         int cpu = sparc_leon3_cpuid();
 
         eirq = leon_eirq_get(cpu);
-        if ((eirq & 0x10) && irq_map[eirq]->irq) /* bit4 tells if IRQ happened */
-                generic_handle_irq(irq_map[eirq]->irq);
+        p = irq_map[eirq];
+        if ((eirq & 0x10) && p && p->irq) /* bit4 tells if IRQ happened */
+                generic_handle_irq(p->irq);
 }
 
 /* The extended IRQ controller has been found, this function registers it */
diff --git a/arch/sparc/kernel/perf_event.c b/arch/sparc/kernel/perf_event.c
index e48651dace1b..b5c38faa4ead 100644
--- a/arch/sparc/kernel/perf_event.c
+++ b/arch/sparc/kernel/perf_event.c
@@ -817,15 +817,17 @@ static u64 nop_for_index(int idx)
 
 static inline void sparc_pmu_enable_event(struct cpu_hw_events *cpuc, struct hw_perf_event *hwc, int idx)
 {
-        u64 val, mask = mask_for_index(idx);
+        u64 enc, val, mask = mask_for_index(idx);
         int pcr_index = 0;
 
         if (sparc_pmu->num_pcrs > 1)
                 pcr_index = idx;
 
+        enc = perf_event_get_enc(cpuc->events[idx]);
+
         val = cpuc->pcr[pcr_index];
         val &= ~mask;
-        val |= hwc->config;
+        val |= event_encoding(enc, idx);
         cpuc->pcr[pcr_index] = val;
 
         pcr_ops->write_pcr(pcr_index, cpuc->pcr[pcr_index]);
@@ -1738,8 +1740,6 @@ static void perf_callchain_user_64(struct perf_callchain_entry *entry,
 {
         unsigned long ufp;
 
-        perf_callchain_store(entry, regs->tpc);
-
         ufp = regs->u_regs[UREG_I6] + STACK_BIAS;
         do {
                 struct sparc_stackf *usf, sf;
@@ -1760,19 +1760,27 @@ static void perf_callchain_user_32(struct perf_callchain_entry *entry,
 {
         unsigned long ufp;
 
-        perf_callchain_store(entry, regs->tpc);
-
         ufp = regs->u_regs[UREG_I6] & 0xffffffffUL;
         do {
-                struct sparc_stackf32 *usf, sf;
                 unsigned long pc;
 
-                usf = (struct sparc_stackf32 *) ufp;
-                if (__copy_from_user_inatomic(&sf, usf, sizeof(sf)))
-                        break;
+                if (thread32_stack_is_64bit(ufp)) {
+                        struct sparc_stackf *usf, sf;
 
-                pc = sf.callers_pc;
-                ufp = (unsigned long)sf.fp;
+                        ufp += STACK_BIAS;
+                        usf = (struct sparc_stackf *) ufp;
+                        if (__copy_from_user_inatomic(&sf, usf, sizeof(sf)))
+                                break;
+                        pc = sf.callers_pc & 0xffffffff;
+                        ufp = ((unsigned long) sf.fp) & 0xffffffff;
+                } else {
+                        struct sparc_stackf32 *usf, sf;
+                        usf = (struct sparc_stackf32 *) ufp;
+                        if (__copy_from_user_inatomic(&sf, usf, sizeof(sf)))
+                                break;
+                        pc = sf.callers_pc;
+                        ufp = (unsigned long)sf.fp;
+                }
                 perf_callchain_store(entry, pc);
         } while (entry->nr < PERF_MAX_STACK_DEPTH);
 }
@@ -1780,6 +1788,11 @@ static void perf_callchain_user_32(struct perf_callchain_entry *entry,
 void
 perf_callchain_user(struct perf_callchain_entry *entry, struct pt_regs *regs)
 {
+        perf_callchain_store(entry, regs->tpc);
+
+        if (!current->mm)
+                return;
+
         flushw_user();
         if (test_thread_flag(TIF_32BIT))
                 perf_callchain_user_32(entry, regs);
diff --git a/arch/sparc/kernel/process_64.c b/arch/sparc/kernel/process_64.c
index fcaa59421126..c6e0c2910043 100644
--- a/arch/sparc/kernel/process_64.c
+++ b/arch/sparc/kernel/process_64.c
@@ -27,6 +27,7 @@
 #include <linux/tick.h>
 #include <linux/init.h>
 #include <linux/cpu.h>
+#include <linux/perf_event.h>
 #include <linux/elfcore.h>
 #include <linux/sysrq.h>
 #include <linux/nmi.h>
@@ -47,6 +48,7 @@
 #include <asm/syscalls.h>
 #include <asm/irq_regs.h>
 #include <asm/smp.h>
+#include <asm/pcr.h>
 
 #include "kstack.h"
 
@@ -204,18 +206,22 @@ void show_regs(struct pt_regs *regs)
         show_stack(current, (unsigned long *) regs->u_regs[UREG_FP]);
 }
 
-struct global_reg_snapshot global_reg_snapshot[NR_CPUS];
-static DEFINE_SPINLOCK(global_reg_snapshot_lock);
+union global_cpu_snapshot global_cpu_snapshot[NR_CPUS];
+static DEFINE_SPINLOCK(global_cpu_snapshot_lock);
 
 static void __global_reg_self(struct thread_info *tp, struct pt_regs *regs,
                               int this_cpu)
 {
+        struct global_reg_snapshot *rp;
+
         flushw_all();
 
-        global_reg_snapshot[this_cpu].tstate = regs->tstate;
-        global_reg_snapshot[this_cpu].tpc = regs->tpc;
-        global_reg_snapshot[this_cpu].tnpc = regs->tnpc;
-        global_reg_snapshot[this_cpu].o7 = regs->u_regs[UREG_I7];
+        rp = &global_cpu_snapshot[this_cpu].reg;
+
+        rp->tstate = regs->tstate;
+        rp->tpc = regs->tpc;
+        rp->tnpc = regs->tnpc;
+        rp->o7 = regs->u_regs[UREG_I7];
 
         if (regs->tstate & TSTATE_PRIV) {
                 struct reg_window *rw;
@@ -223,17 +229,17 @@ static void __global_reg_self(struct thread_info *tp, struct pt_regs *regs,
                 rw = (struct reg_window *)
                         (regs->u_regs[UREG_FP] + STACK_BIAS);
                 if (kstack_valid(tp, (unsigned long) rw)) {
-                        global_reg_snapshot[this_cpu].i7 = rw->ins[7];
+                        rp->i7 = rw->ins[7];
                         rw = (struct reg_window *)
                                 (rw->ins[6] + STACK_BIAS);
                         if (kstack_valid(tp, (unsigned long) rw))
-                                global_reg_snapshot[this_cpu].rpc = rw->ins[7];
+                                rp->rpc = rw->ins[7];
                 }
         } else {
-                global_reg_snapshot[this_cpu].i7 = 0;
-                global_reg_snapshot[this_cpu].rpc = 0;
+                rp->i7 = 0;
+                rp->rpc = 0;
         }
-        global_reg_snapshot[this_cpu].thread = tp;
+        rp->thread = tp;
 }
 
 /* In order to avoid hangs we do not try to synchronize with the
@@ -261,9 +267,9 @@ void arch_trigger_all_cpu_backtrace(void)
         if (!regs)
                 regs = tp->kregs;
 
-        spin_lock_irqsave(&global_reg_snapshot_lock, flags);
+        spin_lock_irqsave(&global_cpu_snapshot_lock, flags);
 
-        memset(global_reg_snapshot, 0, sizeof(global_reg_snapshot));
+        memset(global_cpu_snapshot, 0, sizeof(global_cpu_snapshot));
 
         this_cpu = raw_smp_processor_id();
 
@@ -272,7 +278,7 @@ void arch_trigger_all_cpu_backtrace(void)
         smp_fetch_global_regs();
 
         for_each_online_cpu(cpu) {
-                struct global_reg_snapshot *gp = &global_reg_snapshot[cpu];
+                struct global_reg_snapshot *gp = &global_cpu_snapshot[cpu].reg;
 
                 __global_reg_poll(gp);
 
@@ -295,9 +301,9 @@ void arch_trigger_all_cpu_backtrace(void)
                 }
         }
 
-        memset(global_reg_snapshot, 0, sizeof(global_reg_snapshot));
+        memset(global_cpu_snapshot, 0, sizeof(global_cpu_snapshot));
 
-        spin_unlock_irqrestore(&global_reg_snapshot_lock, flags);
+        spin_unlock_irqrestore(&global_cpu_snapshot_lock, flags);
 }
 
 #ifdef CONFIG_MAGIC_SYSRQ
@@ -309,16 +315,90 @@ static void sysrq_handle_globreg(int key)
 
 static struct sysrq_key_op sparc_globalreg_op = {
         .handler        = sysrq_handle_globreg,
-        .help_msg       = "Globalregs",
+        .help_msg       = "global-regs(Y)",
         .action_msg     = "Show Global CPU Regs",
 };
 
-static int __init sparc_globreg_init(void)
+static void __global_pmu_self(int this_cpu)
 {
-        return register_sysrq_key('y', &sparc_globalreg_op);
+        struct global_pmu_snapshot *pp;
+        int i, num;
+
+        pp = &global_cpu_snapshot[this_cpu].pmu;
+
+        num = 1;
+        if (tlb_type == hypervisor &&
+            sun4v_chip_type >= SUN4V_CHIP_NIAGARA4)
+                num = 4;
+
+        for (i = 0; i < num; i++) {
+                pp->pcr[i] = pcr_ops->read_pcr(i);
+                pp->pic[i] = pcr_ops->read_pic(i);
+        }
 }
 
-core_initcall(sparc_globreg_init);
+static void __global_pmu_poll(struct global_pmu_snapshot *pp)
+{
+        int limit = 0;
+
+        while (!pp->pcr[0] && ++limit < 100) {
+                barrier();
+                udelay(1);
+        }
+}
+
+static void pmu_snapshot_all_cpus(void)
+{
+        unsigned long flags;
+        int this_cpu, cpu;
+
+        spin_lock_irqsave(&global_cpu_snapshot_lock, flags);
+
+        memset(global_cpu_snapshot, 0, sizeof(global_cpu_snapshot));
+
+        this_cpu = raw_smp_processor_id();
+
+        __global_pmu_self(this_cpu);
+
+        smp_fetch_global_pmu();
+
+        for_each_online_cpu(cpu) {
+                struct global_pmu_snapshot *pp = &global_cpu_snapshot[cpu].pmu;
+
+                __global_pmu_poll(pp);
+
+                printk("%c CPU[%3d]: PCR[%08lx:%08lx:%08lx:%08lx] PIC[%08lx:%08lx:%08lx:%08lx]\n",
+                       (cpu == this_cpu ? '*' : ' '), cpu,
+                       pp->pcr[0], pp->pcr[1], pp->pcr[2], pp->pcr[3],
+                       pp->pic[0], pp->pic[1], pp->pic[2], pp->pic[3]);
+        }
+
+        memset(global_cpu_snapshot, 0, sizeof(global_cpu_snapshot));
+
+        spin_unlock_irqrestore(&global_cpu_snapshot_lock, flags);
+}
+
+static void sysrq_handle_globpmu(int key)
+{
+        pmu_snapshot_all_cpus();
+}
+
+static struct sysrq_key_op sparc_globalpmu_op = {
+        .handler        = sysrq_handle_globpmu,
+        .help_msg       = "global-pmu(X)",
+        .action_msg     = "Show Global PMU Regs",
+};
+
+static int __init sparc_sysrq_init(void)
+{
+        int ret = register_sysrq_key('y', &sparc_globalreg_op);
+
+        if (!ret)
+                ret = register_sysrq_key('x', &sparc_globalpmu_op);
+        return ret;
+}
+
+core_initcall(sparc_sysrq_init);
 
 #endif
 
@@ -372,13 +452,16 @@ void flush_thread(void)
 /* It's a bit more tricky when 64-bit tasks are involved... */
 static unsigned long clone_stackframe(unsigned long csp, unsigned long psp)
 {
+        bool stack_64bit = test_thread_64bit_stack(psp);
         unsigned long fp, distance, rval;
 
-        if (!(test_thread_flag(TIF_32BIT))) {
+        if (stack_64bit) {
                 csp += STACK_BIAS;
                 psp += STACK_BIAS;
                 __get_user(fp, &(((struct reg_window __user *)psp)->ins[6]));
                 fp += STACK_BIAS;
+                if (test_thread_flag(TIF_32BIT))
+                        fp &= 0xffffffff;
         } else
                 __get_user(fp, &(((struct reg_window32 __user *)psp)->ins[6]));
 
@@ -392,7 +475,7 @@ static unsigned long clone_stackframe(unsigned long csp, unsigned long psp)
         rval = (csp - distance);
         if (copy_in_user((void __user *) rval, (void __user *) psp, distance))
                 rval = 0;
-        else if (test_thread_flag(TIF_32BIT)) {
+        else if (!stack_64bit) {
                 if (put_user(((u32)csp),
                              &(((struct reg_window32 __user *)rval)->ins[6])))
                         rval = 0;
@@ -427,18 +510,18 @@ void synchronize_user_stack(void)
 
         flush_user_windows();
         if ((window = get_thread_wsaved()) != 0) {
-                int winsize = sizeof(struct reg_window);
-                int bias = 0;
-
-                if (test_thread_flag(TIF_32BIT))
-                        winsize = sizeof(struct reg_window32);
-                else
-                        bias = STACK_BIAS;
-
                 window -= 1;
                 do {
-                        unsigned long sp = (t->rwbuf_stkptrs[window] + bias);
                         struct reg_window *rwin = &t->reg_window[window];
+                        int winsize = sizeof(struct reg_window);
+                        unsigned long sp;
+
+                        sp = t->rwbuf_stkptrs[window];
+
+                        if (test_thread_64bit_stack(sp))
+                                sp += STACK_BIAS;
+                        else
+                                winsize = sizeof(struct reg_window32);
 
                         if (!copy_to_user((char __user *)sp, rwin, winsize)) {
                                 shift_window_buffer(window, get_thread_wsaved() - 1, t);
@@ -464,13 +547,6 @@ void fault_in_user_windows(void)
 {
         struct thread_info *t = current_thread_info();
         unsigned long window;
-        int winsize = sizeof(struct reg_window);
-        int bias = 0;
-
-        if (test_thread_flag(TIF_32BIT))
-                winsize = sizeof(struct reg_window32);
-        else
-                bias = STACK_BIAS;
 
         flush_user_windows();
         window = get_thread_wsaved();
@@ -478,8 +554,16 @@ void fault_in_user_windows(void)
         if (likely(window != 0)) {
                 window -= 1;
                 do {
-                        unsigned long sp = (t->rwbuf_stkptrs[window] + bias);
                         struct reg_window *rwin = &t->reg_window[window];
+                        int winsize = sizeof(struct reg_window);
+                        unsigned long sp;
+
+                        sp = t->rwbuf_stkptrs[window];
+
+                        if (test_thread_64bit_stack(sp))
+                                sp += STACK_BIAS;
+                        else
+                                winsize = sizeof(struct reg_window32);
 
                         if (unlikely(sp & 0x7UL))
                                 stack_unaligned(sp);
diff --git a/arch/sparc/kernel/ptrace_64.c b/arch/sparc/kernel/ptrace_64.c
index 484dabac7045..7ff45e4ba681 100644
--- a/arch/sparc/kernel/ptrace_64.c
+++ b/arch/sparc/kernel/ptrace_64.c
@@ -151,7 +151,7 @@ static int regwindow64_get(struct task_struct *target,
 {
         unsigned long rw_addr = regs->u_regs[UREG_I6];
 
-        if (test_tsk_thread_flag(current, TIF_32BIT)) {
+        if (!test_thread_64bit_stack(rw_addr)) {
                 struct reg_window32 win32;
                 int i;
 
@@ -176,7 +176,7 @@ static int regwindow64_set(struct task_struct *target,
 {
         unsigned long rw_addr = regs->u_regs[UREG_I6];
 
-        if (test_tsk_thread_flag(current, TIF_32BIT)) {
+        if (!test_thread_64bit_stack(rw_addr)) {
                 struct reg_window32 win32;
                 int i;
 
diff --git a/arch/sparc/kernel/setup_64.c b/arch/sparc/kernel/setup_64.c
index 0800e71d8a88..0eaf0059aaef 100644
--- a/arch/sparc/kernel/setup_64.c
+++ b/arch/sparc/kernel/setup_64.c
@@ -316,6 +316,25 @@ static void __init popc_patch(void)
         }
 }
 
+static void __init pause_patch(void)
+{
+        struct pause_patch_entry *p;
+
+        p = &__pause_3insn_patch;
+        while (p < &__pause_3insn_patch_end) {
+                unsigned long i, addr = p->addr;
+
+                for (i = 0; i < 3; i++) {
+                        *(unsigned int *) (addr +  (i * 4)) = p->insns[i];
+                        wmb();
+                        __asm__ __volatile__("flush     %0"
+                                             : : "r" (addr +  (i * 4)));
+                }
+
+                p++;
+        }
+}
+
 #ifdef CONFIG_SMP
 void __init boot_cpu_id_too_large(int cpu)
 {
@@ -528,6 +547,8 @@ static void __init init_sparc64_elf_hwcap(void)
 
         if (sparc64_elf_hwcap & AV_SPARC_POPC)
                 popc_patch();
+        if (sparc64_elf_hwcap & AV_SPARC_PAUSE)
+                pause_patch();
 }
 
 void __init setup_arch(char **cmdline_p)
diff --git a/arch/sparc/kernel/signal_64.c b/arch/sparc/kernel/signal_64.c
index 867de2f8189c..689e1ba62809 100644
--- a/arch/sparc/kernel/signal_64.c
+++ b/arch/sparc/kernel/signal_64.c
@@ -295,9 +295,7 @@ void do_rt_sigreturn(struct pt_regs *regs)
                 err |= restore_fpu_state(regs, fpu_save);
 
         err |= __copy_from_user(&set, &sf->mask, sizeof(sigset_t));
-        err |= do_sigaltstack(&sf->stack, NULL, (unsigned long)sf);
-
-        if (err)
+        if (err || do_sigaltstack(&sf->stack, NULL, (unsigned long)sf) == -EFAULT)
                 goto segv;
 
         err |= __get_user(rwin_save, &sf->rwin_save);
diff --git a/arch/sparc/kernel/smp_64.c b/arch/sparc/kernel/smp_64.c
index 781bcb10b8bd..d94b878577b7 100644
--- a/arch/sparc/kernel/smp_64.c
+++ b/arch/sparc/kernel/smp_64.c
@@ -852,6 +852,8 @@ extern unsigned long xcall_flush_tlb_mm;
 extern unsigned long xcall_flush_tlb_pending;
 extern unsigned long xcall_flush_tlb_kernel_range;
 extern unsigned long xcall_fetch_glob_regs;
+extern unsigned long xcall_fetch_glob_pmu;
+extern unsigned long xcall_fetch_glob_pmu_n4;
 extern unsigned long xcall_receive_signal;
 extern unsigned long xcall_new_mmu_context_version;
 #ifdef CONFIG_KGDB
@@ -1000,6 +1002,15 @@ void smp_fetch_global_regs(void)
         smp_cross_call(&xcall_fetch_glob_regs, 0, 0, 0);
 }
 
+void smp_fetch_global_pmu(void)
+{
+        if (tlb_type == hypervisor &&
+            sun4v_chip_type >= SUN4V_CHIP_NIAGARA4)
+                smp_cross_call(&xcall_fetch_glob_pmu_n4, 0, 0, 0);
+        else
+                smp_cross_call(&xcall_fetch_glob_pmu, 0, 0, 0);
+}
+
 /* We know that the window frames of the user have been flushed
  * to the stack before we get here because all callers of us
  * are flush_tlb_*() routines, and these run after flush_cache_*()
diff --git a/arch/sparc/kernel/sys_sparc_64.c b/arch/sparc/kernel/sys_sparc_64.c
index 11c6c9603e71..878ef3d5fec5 100644
--- a/arch/sparc/kernel/sys_sparc_64.c
+++ b/arch/sparc/kernel/sys_sparc_64.c
@@ -751,3 +751,8 @@ int kernel_execve(const char *filename,
                       : "cc");
         return __res;
 }
+
+asmlinkage long sys_kern_features(void)
+{
+        return KERN_FEATURE_MIXED_MODE_STACK;
+}
diff --git a/arch/sparc/kernel/systbls_32.S b/arch/sparc/kernel/systbls_32.S
index 63402f9e9f51..5147f574f125 100644
--- a/arch/sparc/kernel/systbls_32.S
+++ b/arch/sparc/kernel/systbls_32.S
@@ -85,3 +85,4 @@ sys_call_table:
 /*325*/ .long sys_pwritev, sys_rt_tgsigqueueinfo, sys_perf_event_open, sys_recvmmsg, sys_fanotify_init
 /*330*/ .long sys_fanotify_mark, sys_prlimit64, sys_name_to_handle_at, sys_open_by_handle_at, sys_clock_adjtime
 /*335*/ .long sys_syncfs, sys_sendmmsg, sys_setns, sys_process_vm_readv, sys_process_vm_writev
+/*340*/ .long sys_ni_syscall, sys_kcmp
diff --git a/arch/sparc/kernel/systbls_64.S b/arch/sparc/kernel/systbls_64.S
index 3a58e0d66f51..1c9af9fa38e9 100644
--- a/arch/sparc/kernel/systbls_64.S
+++ b/arch/sparc/kernel/systbls_64.S
@@ -86,6 +86,7 @@ sys_call_table32:
         .word compat_sys_pwritev, compat_sys_rt_tgsigqueueinfo, sys_perf_event_open, compat_sys_recvmmsg, sys_fanotify_init
 /*330*/ .word sys32_fanotify_mark, sys_prlimit64, sys_name_to_handle_at, compat_sys_open_by_handle_at, compat_sys_clock_adjtime
         .word sys_syncfs, compat_sys_sendmmsg, sys_setns, compat_sys_process_vm_readv, compat_sys_process_vm_writev
+/*340*/ .word sys_kern_features, sys_kcmp
 
 #endif /* CONFIG_COMPAT */
 
@@ -163,3 +164,4 @@ sys_call_table:
         .word sys_pwritev, sys_rt_tgsigqueueinfo, sys_perf_event_open, sys_recvmmsg, sys_fanotify_init
 /*330*/ .word sys_fanotify_mark, sys_prlimit64, sys_name_to_handle_at, sys_open_by_handle_at, sys_clock_adjtime
         .word sys_syncfs, sys_sendmmsg, sys_setns, sys_process_vm_readv, sys_process_vm_writev
+/*340*/ .word sys_kern_features, sys_kcmp
diff --git a/arch/sparc/kernel/unaligned_64.c b/arch/sparc/kernel/unaligned_64.c
index f81d038f7340..8201c25e7669 100644
--- a/arch/sparc/kernel/unaligned_64.c
+++ b/arch/sparc/kernel/unaligned_64.c
@@ -113,21 +113,24 @@ static inline long sign_extend_imm13(long imm)
 
 static unsigned long fetch_reg(unsigned int reg, struct pt_regs *regs)
 {
-        unsigned long value;
+        unsigned long value, fp;
         
         if (reg < 16)
                 return (!reg ? 0 : regs->u_regs[reg]);
+
+        fp = regs->u_regs[UREG_FP];
+
         if (regs->tstate & TSTATE_PRIV) {
                 struct reg_window *win;
-                win = (struct reg_window *)(regs->u_regs[UREG_FP] + STACK_BIAS);
+                win = (struct reg_window *)(fp + STACK_BIAS);
                 value = win->locals[reg - 16];
-        } else if (test_thread_flag(TIF_32BIT)) {
+        } else if (!test_thread_64bit_stack(fp)) {
                 struct reg_window32 __user *win32;
-                win32 = (struct reg_window32 __user *)((unsigned long)((u32)regs->u_regs[UREG_FP]));
+                win32 = (struct reg_window32 __user *)((unsigned long)((u32)fp));
                 get_user(value, &win32->locals[reg - 16]);
         } else {
                 struct reg_window __user *win;
-                win = (struct reg_window __user *)(regs->u_regs[UREG_FP] + STACK_BIAS);
+                win = (struct reg_window __user *)(fp + STACK_BIAS);
                 get_user(value, &win->locals[reg - 16]);
         }
         return value;
@@ -135,19 +138,24 @@ static unsigned long fetch_reg(unsigned int reg, struct pt_regs *regs)
 
 static unsigned long *fetch_reg_addr(unsigned int reg, struct pt_regs *regs)
 {
+        unsigned long fp;
+
         if (reg < 16)
                 return &regs->u_regs[reg];
+
+        fp = regs->u_regs[UREG_FP];
+
         if (regs->tstate & TSTATE_PRIV) {
                 struct reg_window *win;
-                win = (struct reg_window *)(regs->u_regs[UREG_FP] + STACK_BIAS);
+                win = (struct reg_window *)(fp + STACK_BIAS);
                 return &win->locals[reg - 16];
-        } else if (test_thread_flag(TIF_32BIT)) {
+        } else if (!test_thread_64bit_stack(fp)) {
                 struct reg_window32 *win32;
-                win32 = (struct reg_window32 *)((unsigned long)((u32)regs->u_regs[UREG_FP]));
+                win32 = (struct reg_window32 *)((unsigned long)((u32)fp));
                 return (unsigned long *)&win32->locals[reg - 16];
         } else {
                 struct reg_window *win;
-                win = (struct reg_window *)(regs->u_regs[UREG_FP] + STACK_BIAS);
+                win = (struct reg_window *)(fp + STACK_BIAS);
                 return &win->locals[reg - 16];
         }
 }
@@ -392,13 +400,15 @@ int handle_popc(u32 insn, struct pt_regs *regs)
                 if (rd)
                         regs->u_regs[rd] = ret;
         } else {
-                if (test_thread_flag(TIF_32BIT)) {
+                unsigned long fp = regs->u_regs[UREG_FP];
+
+                if (!test_thread_64bit_stack(fp)) {
                         struct reg_window32 __user *win32;
-                        win32 = (struct reg_window32 __user *)((unsigned long)((u32)regs->u_regs[UREG_FP]));
+                        win32 = (struct reg_window32 __user *)((unsigned long)((u32)fp));
                         put_user(ret, &win32->locals[rd - 16]);
                 } else {
                         struct reg_window __user *win;
-                        win = (struct reg_window __user *)(regs->u_regs[UREG_FP] + STACK_BIAS);
+                        win = (struct reg_window __user *)(fp + STACK_BIAS);
                         put_user(ret, &win->locals[rd - 16]);
                 }
         }
@@ -554,7 +564,7 @@ void handle_ld_nf(u32 insn, struct pt_regs *regs)
                 reg[0] = 0;
                 if ((insn & 0x780000) == 0x180000)
                         reg[1] = 0;
-        } else if (test_thread_flag(TIF_32BIT)) {
+        } else if (!test_thread_64bit_stack(regs->u_regs[UREG_FP])) {
                 put_user(0, (int __user *) reg);
                 if ((insn & 0x780000) == 0x180000)
                         put_user(0, ((int __user *) reg) + 1);
diff --git a/arch/sparc/kernel/visemul.c b/arch/sparc/kernel/visemul.c
index 08e074b7eb6a..c096c624ac4d 100644
--- a/arch/sparc/kernel/visemul.c
+++ b/arch/sparc/kernel/visemul.c
@@ -149,21 +149,24 @@ static inline void maybe_flush_windows(unsigned int rs1, unsigned int rs2,
 
 static unsigned long fetch_reg(unsigned int reg, struct pt_regs *regs)
 {
-        unsigned long value;
+        unsigned long value, fp;
         
         if (reg < 16)
                 return (!reg ? 0 : regs->u_regs[reg]);
+
+        fp = regs->u_regs[UREG_FP];
+
         if (regs->tstate & TSTATE_PRIV) {
                 struct reg_window *win;
-                win = (struct reg_window *)(regs->u_regs[UREG_FP] + STACK_BIAS);
+                win = (struct reg_window *)(fp + STACK_BIAS);
                 value = win->locals[reg - 16];
-        } else if (test_thread_flag(TIF_32BIT)) {
+        } else if (!test_thread_64bit_stack(fp)) {
                 struct reg_window32 __user *win32;
-                win32 = (struct reg_window32 __user *)((unsigned long)((u32)regs->u_regs[UREG_FP]));
+                win32 = (struct reg_window32 __user *)((unsigned long)((u32)fp));
                 get_user(value, &win32->locals[reg - 16]);
         } else {
                 struct reg_window __user *win;
-                win = (struct reg_window __user *)(regs->u_regs[UREG_FP] + STACK_BIAS);
+                win = (struct reg_window __user *)(fp + STACK_BIAS);
                 get_user(value, &win->locals[reg - 16]);
         }
         return value;
@@ -172,16 +175,18 @@ static unsigned long fetch_reg(unsigned int reg, struct pt_regs *regs)
 static inline unsigned long __user *__fetch_reg_addr_user(unsigned int reg,
                                                           struct pt_regs *regs)
 {
+        unsigned long fp = regs->u_regs[UREG_FP];
+
         BUG_ON(reg < 16);
         BUG_ON(regs->tstate & TSTATE_PRIV);
 
-        if (test_thread_flag(TIF_32BIT)) {
+        if (!test_thread_64bit_stack(fp)) {
                 struct reg_window32 __user *win32;
-                win32 = (struct reg_window32 __user *)((unsigned long)((u32)regs->u_regs[UREG_FP]));
+                win32 = (struct reg_window32 __user *)((unsigned long)((u32)fp));
                 return (unsigned long __user *)&win32->locals[reg - 16];
         } else {
                 struct reg_window __user *win;
-                win = (struct reg_window __user *)(regs->u_regs[UREG_FP] + STACK_BIAS);
+                win = (struct reg_window __user *)(fp + STACK_BIAS);
                 return &win->locals[reg - 16];
         }
 }
@@ -204,7 +209,7 @@ static void store_reg(struct pt_regs *regs, unsigned long val, unsigned long rd)
         } else {
                 unsigned long __user *rd_user = __fetch_reg_addr_user(rd, regs);
 
-                if (test_thread_flag(TIF_32BIT))
+                if (!test_thread_64bit_stack(regs->u_regs[UREG_FP]))
                         __put_user((u32)val, (u32 __user *)rd_user);
                 else
                         __put_user(val, rd_user);
diff --git a/arch/sparc/kernel/vmlinux.lds.S b/arch/sparc/kernel/vmlinux.lds.S
index 89c2c29f154b..0bacceb19150 100644
--- a/arch/sparc/kernel/vmlinux.lds.S
+++ b/arch/sparc/kernel/vmlinux.lds.S
@@ -132,6 +132,11 @@ SECTIONS
                 *(.popc_6insn_patch)
                 __popc_6insn_patch_end = .;
         }
+        .pause_3insn_patch : {
+                __pause_3insn_patch = .;
+                *(.pause_3insn_patch)
+                __pause_3insn_patch_end = .;
+        }
         PERCPU_SECTION(SMP_CACHE_BYTES)
 
         . = ALIGN(PAGE_SIZE);
diff --git a/arch/sparc/kernel/winfixup.S b/arch/sparc/kernel/winfixup.S
index a6b0863c27df..1e67ce958369 100644
--- a/arch/sparc/kernel/winfixup.S
+++ b/arch/sparc/kernel/winfixup.S
@@ -43,6 +43,8 @@ spill_fixup_mna:
 spill_fixup_dax:
         TRAP_LOAD_THREAD_REG(%g6, %g1)
         ldx     [%g6 + TI_FLAGS], %g1
+        andcc   %sp, 0x1, %g0
+        movne   %icc, 0, %g1
         andcc   %g1, _TIF_32BIT, %g0
         ldub    [%g6 + TI_WSAVED], %g1
         sll     %g1, 3, %g3