Merge branch 'linus' into stackprotector
[pandora-kernel.git] / arch / x86 / kernel / process_64.c
index f73cfbc..749d5f8 100644 (file)
 #include <linux/kdebug.h>
 #include <linux/tick.h>
 #include <linux/prctl.h>
+#include <linux/uaccess.h>
+#include <linux/io.h>
 
-#include <asm/uaccess.h>
 #include <asm/pgtable.h>
 #include <asm/system.h>
-#include <asm/io.h>
 #include <asm/processor.h>
 #include <asm/i387.h>
 #include <asm/mmu_context.h>
 #include <asm/proto.h>
 #include <asm/ia32.h>
 #include <asm/idle.h>
+#include <asm/syscalls.h>
 
 asmlinkage extern void ret_from_fork(void);
 
 unsigned long kernel_thread_flags = CLONE_VM | CLONE_UNTRACED;
 
-unsigned long boot_option_idle_override = 0;
-EXPORT_SYMBOL(boot_option_idle_override);
-
-/*
- * Powermanagement idle function, if any..
- */
-void (*pm_idle)(void);
-EXPORT_SYMBOL(pm_idle);
-
 static ATOMIC_NOTIFIER_HEAD(idle_notifier);
 
 void idle_notifier_register(struct notifier_block *n)
@@ -95,48 +87,12 @@ void exit_idle(void)
        __exit_idle();
 }
 
-/*
- * We use this if we don't have any better
- * idle routine..
- */
-void default_idle(void)
-{
-       current_thread_info()->status &= ~TS_POLLING;
-       /*
-        * TS_POLLING-cleared state must be visible before we
-        * test NEED_RESCHED:
-        */
-       smp_mb();
-       if (!need_resched())
-               safe_halt();    /* enables interrupts racelessly */
-       else
-               local_irq_enable();
-       current_thread_info()->status |= TS_POLLING;
-}
-
-#ifdef CONFIG_HOTPLUG_CPU
-DECLARE_PER_CPU(int, cpu_state);
-
-#include <asm/nmi.h>
-/* We halt the CPU with physical CPU hotplug */
-static inline void play_dead(void)
-{
-       idle_task_exit();
-       wbinvd();
-       mb();
-       /* Ack it */
-       __get_cpu_var(cpu_state) = CPU_DEAD;
-
-       local_irq_disable();
-       while (1)
-               halt();
-}
-#else
+#ifndef CONFIG_SMP
 static inline void play_dead(void)
 {
        BUG();
 }
-#endif /* CONFIG_HOTPLUG_CPU */
+#endif
 
 /*
  * The idle thread. There's no useful work to be
@@ -160,14 +116,11 @@ void cpu_idle(void)
 
        /* endless idle loop with no priority at all */
        while (1) {
-               tick_nohz_stop_sched_tick();
+               tick_nohz_stop_sched_tick(1);
                while (!need_resched()) {
-                       void (*idle)(void);
 
                        rmb();
-                       idle = pm_idle;
-                       if (!idle)
-                               idle = default_idle;
+
                        if (cpu_is_offline(smp_processor_id()))
                                play_dead();
                        /*
@@ -177,7 +130,10 @@ void cpu_idle(void)
                         */
                        local_irq_disable();
                        enter_idle();
-                       idle();
+                       /* Don't trace irqs off for idle */
+                       stop_critical_timings();
+                       pm_idle();
+                       start_critical_timings();
                        /* In many cases the interrupt that ended idle
                           has already called exit_idle. But some idle
                           loops can be woken up without interrupt. */
@@ -192,7 +148,7 @@ void cpu_idle(void)
 }
 
 /* Prints also some state that isn't saved in the pt_regs */
-void __show_regs(struct pt_regs * regs)
+void __show_regs(struct pt_regs *regs, int all)
 {
        unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L, fs, gs, shadowgs;
        unsigned long d0, d1, d2, d3, d6, d7;
@@ -201,60 +157,65 @@ void __show_regs(struct pt_regs * regs)
 
        printk("\n");
        print_modules();
-       printk("Pid: %d, comm: %.20s %s %s %.*s\n",
+       printk(KERN_INFO "Pid: %d, comm: %.20s %s %s %.*s\n",
                current->pid, current->comm, print_tainted(),
                init_utsname()->release,
                (int)strcspn(init_utsname()->version, " "),
                init_utsname()->version);
-       printk("RIP: %04lx:[<%016lx>] ", regs->cs & 0xffff, regs->ip);
+       printk(KERN_INFO "RIP: %04lx:[<%016lx>] ", regs->cs & 0xffff, regs->ip);
        printk_address(regs->ip, 1);
-       printk("RSP: %04lx:%016lx  EFLAGS: %08lx\n", regs->ss, regs->sp,
-               regs->flags);
-       printk("RAX: %016lx RBX: %016lx RCX: %016lx\n",
+       printk(KERN_INFO "RSP: %04lx:%016lx  EFLAGS: %08lx\n", regs->ss,
+                       regs->sp, regs->flags);
+       printk(KERN_INFO "RAX: %016lx RBX: %016lx RCX: %016lx\n",
               regs->ax, regs->bx, regs->cx);
-       printk("RDX: %016lx RSI: %016lx RDI: %016lx\n",
+       printk(KERN_INFO "RDX: %016lx RSI: %016lx RDI: %016lx\n",
               regs->dx, regs->si, regs->di);
-       printk("RBP: %016lx R08: %016lx R09: %016lx\n",
+       printk(KERN_INFO "RBP: %016lx R08: %016lx R09: %016lx\n",
               regs->bp, regs->r8, regs->r9);
-       printk("R10: %016lx R11: %016lx R12: %016lx\n",
-              regs->r10, regs->r11, regs->r12); 
-       printk("R13: %016lx R14: %016lx R15: %016lx\n",
-              regs->r13, regs->r14, regs->r15); 
-
-       asm("movl %%ds,%0" : "=r" (ds)); 
-       asm("movl %%cs,%0" : "=r" (cs)); 
-       asm("movl %%es,%0" : "=r" (es)); 
+       printk(KERN_INFO "R10: %016lx R11: %016lx R12: %016lx\n",
+              regs->r10, regs->r11, regs->r12);
+       printk(KERN_INFO "R13: %016lx R14: %016lx R15: %016lx\n",
+              regs->r13, regs->r14, regs->r15);
+
+       asm("movl %%ds,%0" : "=r" (ds));
+       asm("movl %%cs,%0" : "=r" (cs));
+       asm("movl %%es,%0" : "=r" (es));
        asm("movl %%fs,%0" : "=r" (fsindex));
        asm("movl %%gs,%0" : "=r" (gsindex));
 
        rdmsrl(MSR_FS_BASE, fs);
-       rdmsrl(MSR_GS_BASE, gs); 
-       rdmsrl(MSR_KERNEL_GS_BASE, shadowgs); 
+       rdmsrl(MSR_GS_BASE, gs);
+       rdmsrl(MSR_KERNEL_GS_BASE, shadowgs);
+
+       if (!all)
+               return;
 
        cr0 = read_cr0();
        cr2 = read_cr2();
        cr3 = read_cr3();
        cr4 = read_cr4();
 
-       printk("FS:  %016lx(%04x) GS:%016lx(%04x) knlGS:%016lx\n", 
-              fs,fsindex,gs,gsindex,shadowgs); 
-       printk("CS:  %04x DS: %04x ES: %04x CR0: %016lx\n", cs, ds, es, cr0); 
-       printk("CR2: %016lx CR3: %016lx CR4: %016lx\n", cr2, cr3, cr4);
+       printk(KERN_INFO "FS:  %016lx(%04x) GS:%016lx(%04x) knlGS:%016lx\n",
+              fs, fsindex, gs, gsindex, shadowgs);
+       printk(KERN_INFO "CS:  %04x DS: %04x ES: %04x CR0: %016lx\n", cs, ds,
+                       es, cr0);
+       printk(KERN_INFO "CR2: %016lx CR3: %016lx CR4: %016lx\n", cr2, cr3,
+                       cr4);
 
        get_debugreg(d0, 0);
        get_debugreg(d1, 1);
        get_debugreg(d2, 2);
-       printk("DR0: %016lx DR1: %016lx DR2: %016lx\n", d0, d1, d2);
+       printk(KERN_INFO "DR0: %016lx DR1: %016lx DR2: %016lx\n", d0, d1, d2);
        get_debugreg(d3, 3);
        get_debugreg(d6, 6);
        get_debugreg(d7, 7);
-       printk("DR3: %016lx DR6: %016lx DR7: %016lx\n", d3, d6, d7);
+       printk(KERN_INFO "DR3: %016lx DR6: %016lx DR7: %016lx\n", d3, d6, d7);
 }
 
 void show_regs(struct pt_regs *regs)
 {
-       printk("CPU %d:", smp_processor_id());
-       __show_regs(regs);
+       printk(KERN_INFO "CPU %d:", smp_processor_id());
+       __show_regs(regs, 1);
        show_trace(NULL, regs, (void *)(regs + 1), regs->bp);
 }
 
@@ -279,6 +240,14 @@ void exit_thread(void)
                t->io_bitmap_max = 0;
                put_cpu();
        }
+#ifdef CONFIG_X86_DS
+       /* Free any DS contexts that have not been properly released. */
+       if (unlikely(t->ds_ctx)) {
+               /* we clear debugctl to make sure DS is not used. */
+               update_debugctlmsr(0);
+               ds_free(t->ds_ctx);
+       }
+#endif /* CONFIG_X86_DS */
 }
 
 void flush_thread(void)
@@ -354,10 +323,10 @@ void prepare_to_copy(struct task_struct *tsk)
 
 int copy_thread(int nr, unsigned long clone_flags, unsigned long sp,
                unsigned long unused,
-       struct task_struct * p, struct pt_regs * regs)
+       struct task_struct *p, struct pt_regs *regs)
 {
        int err;
-       struct pt_regs * childregs;
+       struct pt_regs *childregs;
        struct task_struct *me = current;
 
        childregs = ((struct pt_regs *)
@@ -378,10 +347,10 @@ int copy_thread(int nr, unsigned long clone_flags, unsigned long sp,
        p->thread.fs = me->thread.fs;
        p->thread.gs = me->thread.gs;
 
-       asm("mov %%gs,%0" : "=m" (p->thread.gsindex));
-       asm("mov %%fs,%0" : "=m" (p->thread.fsindex));
-       asm("mov %%es,%0" : "=m" (p->thread.es));
-       asm("mov %%ds,%0" : "=m" (p->thread.ds));
+       savesegment(gs, p->thread.gsindex);
+       savesegment(fs, p->thread.fsindex);
+       savesegment(es, p->thread.es);
+       savesegment(ds, p->thread.ds);
 
        if (unlikely(test_tsk_thread_flag(me, TIF_IO_BITMAP))) {
                p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL);
@@ -402,10 +371,10 @@ int copy_thread(int nr, unsigned long clone_flags, unsigned long sp,
                if (test_thread_flag(TIF_IA32))
                        err = do_set_thread_area(p, -1,
                                (struct user_desc __user *)childregs->si, 0);
-               else                    
-#endif  
-                       err = do_arch_prctl(p, ARCH_SET_FS, childregs->r8); 
-               if (err) 
+               else
+#endif
+                       err = do_arch_prctl(p, ARCH_SET_FS, childregs->r8);
+               if (err)
                        goto out;
        }
        err = 0;
@@ -420,7 +389,9 @@ out:
 void
 start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp)
 {
-       asm volatile("movl %0, %%fs; movl %0, %%es; movl %0, %%ds" :: "r"(0));
+       loadsegment(fs, 0);
+       loadsegment(es, 0);
+       loadsegment(ds, 0);
        load_gs_index(0);
        regs->ip                = new_ip;
        regs->sp                = new_sp;
@@ -510,13 +481,27 @@ static inline void __switch_to_xtra(struct task_struct *prev_p,
        next = &next_p->thread;
 
        debugctl = prev->debugctlmsr;
-       if (next->ds_area_msr != prev->ds_area_msr) {
-               /* we clear debugctl to make sure DS
-                * is not in use when we change it */
-               debugctl = 0;
-               update_debugctlmsr(0);
-               wrmsrl(MSR_IA32_DS_AREA, next->ds_area_msr);
+
+#ifdef CONFIG_X86_DS
+       {
+               unsigned long ds_prev = 0, ds_next = 0;
+
+               if (prev->ds_ctx)
+                       ds_prev = (unsigned long)prev->ds_ctx->ds;
+               if (next->ds_ctx)
+                       ds_next = (unsigned long)next->ds_ctx->ds;
+
+               if (ds_next != ds_prev) {
+                       /*
+                        * We clear debugctl to make sure DS
+                        * is not in use when we change it:
+                        */
+                       debugctl = 0;
+                       update_debugctlmsr(0);
+                       wrmsrl(MSR_IA32_DS_AREA, ds_next);
+               }
        }
+#endif /* CONFIG_X86_DS */
 
        if (next->debugctlmsr != debugctl)
                update_debugctlmsr(next->debugctlmsr);
@@ -554,13 +539,13 @@ static inline void __switch_to_xtra(struct task_struct *prev_p,
                memset(tss->io_bitmap, 0xff, prev->io_bitmap_max);
        }
 
-#ifdef X86_BTS
+#ifdef CONFIG_X86_PTRACE_BTS
        if (test_tsk_thread_flag(prev_p, TIF_BTS_TRACE_TS))
                ptrace_bts_take_timestamp(prev_p, BTS_TASK_DEPARTS);
 
        if (test_tsk_thread_flag(next_p, TIF_BTS_TRACE_TS))
                ptrace_bts_take_timestamp(next_p, BTS_TASK_ARRIVES);
-#endif
+#endif /* CONFIG_X86_PTRACE_BTS */
 }
 
 /*
@@ -575,13 +560,14 @@ static inline void __switch_to_xtra(struct task_struct *prev_p,
 struct task_struct *
 __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
 {
-       struct thread_struct *prev = &prev_p->thread,
-                                *next = &next_p->thread;
+       struct thread_struct *prev = &prev_p->thread;
+       struct thread_struct *next = &next_p->thread;
        int cpu = smp_processor_id();
        struct tss_struct *tss = &per_cpu(init_tss, cpu);
+       unsigned fsindex, gsindex;
 
        /* we're going to use this soon, after a few expensive things */
-       if (next_p->fpu_counter>5)
+       if (next_p->fpu_counter > 5)
                prefetch(next->xstate);
 
        /*
@@ -589,69 +575,82 @@ __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
         */
        load_sp0(tss, next);
 
-       /* 
+       /*
         * Switch DS and ES.
         * This won't pick up thread selector changes, but I guess that is ok.
         */
-       asm volatile("mov %%es,%0" : "=m" (prev->es));
+       savesegment(es, prev->es);
        if (unlikely(next->es | prev->es))
-               loadsegment(es, next->es); 
-       
-       asm volatile ("mov %%ds,%0" : "=m" (prev->ds));
+               loadsegment(es, next->es);
+
+       savesegment(ds, prev->ds);
        if (unlikely(next->ds | prev->ds))
                loadsegment(ds, next->ds);
 
+
+       /* We must save %fs and %gs before load_TLS() because
+        * %fs and %gs may be cleared by load_TLS().
+        *
+        * (e.g. xen_load_tls())
+        */
+       savesegment(fs, fsindex);
+       savesegment(gs, gsindex);
+
        load_TLS(next, cpu);
 
-       /* 
+       /*
+        * Leave lazy mode, flushing any hypercalls made here.
+        * This must be done before restoring TLS segments so
+        * the GDT and LDT are properly updated, and must be
+        * done before math_state_restore, so the TS bit is up
+        * to date.
+        */
+       arch_leave_lazy_cpu_mode();
+
+       /*
         * Switch FS and GS.
+        *
+        * Segment register != 0 always requires a reload.  Also
+        * reload when it has changed.  When prev process used 64bit
+        * base always reload to avoid an information leak.
         */
-       { 
-               unsigned fsindex;
-               asm volatile("movl %%fs,%0" : "=r" (fsindex)); 
-               /* segment register != 0 always requires a reload. 
-                  also reload when it has changed. 
-                  when prev process used 64bit base always reload
-                  to avoid an information leak. */
-               if (unlikely(fsindex | next->fsindex | prev->fs)) {
-                       loadsegment(fs, next->fsindex);
-                       /* check if the user used a selector != 0
-                        * if yes clear 64bit base, since overloaded base
-                         * is always mapped to the Null selector
-                         */
-                       if (fsindex)
-                       prev->fs = 0;                           
-               }
-               /* when next process has a 64bit base use it */
-               if (next->fs) 
-                       wrmsrl(MSR_FS_BASE, next->fs); 
-               prev->fsindex = fsindex;
+       if (unlikely(fsindex | next->fsindex | prev->fs)) {
+               loadsegment(fs, next->fsindex);
+               /*
+                * Check if the user used a selector != 0; if yes
+                *  clear 64bit base, since overloaded base is always
+                *  mapped to the Null selector
+                */
+               if (fsindex)
+                       prev->fs = 0;
        }
-       { 
-               unsigned gsindex;
-               asm volatile("movl %%gs,%0" : "=r" (gsindex)); 
-               if (unlikely(gsindex | next->gsindex | prev->gs)) {
-                       load_gs_index(next->gsindex);
-                       if (gsindex)
-                       prev->gs = 0;                           
-               }
-               if (next->gs)
-                       wrmsrl(MSR_KERNEL_GS_BASE, next->gs); 
-               prev->gsindex = gsindex;
+       /* when next process has a 64bit base use it */
+       if (next->fs)
+               wrmsrl(MSR_FS_BASE, next->fs);
+       prev->fsindex = fsindex;
+
+       if (unlikely(gsindex | next->gsindex | prev->gs)) {
+               load_gs_index(next->gsindex);
+               if (gsindex)
+                       prev->gs = 0;
        }
+       if (next->gs)
+               wrmsrl(MSR_KERNEL_GS_BASE, next->gs);
+       prev->gsindex = gsindex;
 
        /* Must be after DS reload */
        unlazy_fpu(prev_p);
 
-       /* 
+       /*
         * Switch the PDA and FPU contexts.
         */
        prev->usersp = read_pda(oldrsp);
        write_pda(oldrsp, next->usersp);
-       write_pda(pcurrent, next_p); 
+       write_pda(pcurrent, next_p);
 
        write_pda(kernelstack,
-       (unsigned long)task_stack_page(next_p) + THREAD_SIZE - PDA_STACKOFFSET);
+                 (unsigned long)task_stack_page(next_p) +
+                 THREAD_SIZE - PDA_STACKOFFSET);
 #ifdef CONFIG_CC_STACKPROTECTOR
        /*
         * Build time only check to make sure the stack_canary is at
@@ -687,7 +686,7 @@ long sys_execve(char __user *name, char __user * __user *argv,
                char __user * __user *envp, struct pt_regs *regs)
 {
        long error;
-       char * filename;
+       char *filename;
 
        filename = getname(name);
        error = PTR_ERR(filename);
@@ -745,55 +744,55 @@ asmlinkage long sys_vfork(struct pt_regs *regs)
 unsigned long get_wchan(struct task_struct *p)
 {
        unsigned long stack;
-       u64 fp,ip;
+       u64 fp, ip;
        int count = 0;
 
-       if (!p || p == current || p->state==TASK_RUNNING)
-               return 0; 
+       if (!p || p == current || p->state == TASK_RUNNING)
+               return 0;
        stack = (unsigned long)task_stack_page(p);
-       if (p->thread.sp < stack || p->thread.sp > stack+THREAD_SIZE)
+       if (p->thread.sp < stack || p->thread.sp >= stack+THREAD_SIZE)
                return 0;
        fp = *(u64 *)(p->thread.sp);
-       do { 
+       do {
                if (fp < (unsigned long)stack ||
-                   fp > (unsigned long)stack+THREAD_SIZE)
-                       return 0; 
+                   fp >= (unsigned long)stack+THREAD_SIZE)
+                       return 0;
                ip = *(u64 *)(fp+8);
                if (!in_sched_functions(ip))
                        return ip;
-               fp = *(u64 *)fp; 
-       } while (count++ < 16); 
+               fp = *(u64 *)fp;
+       } while (count++ < 16);
        return 0;
 }
 
 long do_arch_prctl(struct task_struct *task, int code, unsigned long addr)
-{ 
-       int ret = 0; 
+{
+       int ret = 0;
        int doit = task == current;
        int cpu;
 
-       switch (code) { 
+       switch (code) {
        case ARCH_SET_GS:
                if (addr >= TASK_SIZE_OF(task))
-                       return -EPERM; 
+                       return -EPERM;
                cpu = get_cpu();
-               /* handle small bases via the GDT because that's faster to 
+               /* handle small bases via the GDT because that's faster to
                   switch. */
-               if (addr <= 0xffffffff) {  
-                       set_32bit_tls(task, GS_TLS, addr); 
-                       if (doit) { 
+               if (addr <= 0xffffffff) {
+                       set_32bit_tls(task, GS_TLS, addr);
+                       if (doit) {
                                load_TLS(&task->thread, cpu);
-                               load_gs_index(GS_TLS_SEL); 
+                               load_gs_index(GS_TLS_SEL);
                        }
-                       task->thread.gsindex = GS_TLS_SEL; 
+                       task->thread.gsindex = GS_TLS_SEL;
                        task->thread.gs = 0;
-               } else { 
+               } else {
                        task->thread.gsindex = 0;
                        task->thread.gs = addr;
                        if (doit) {
                                load_gs_index(0);
                                ret = checking_wrmsrl(MSR_KERNEL_GS_BASE, addr);
-                       } 
+                       }
                }
                put_cpu();
                break;
@@ -809,7 +808,7 @@ long do_arch_prctl(struct task_struct *task, int code, unsigned long addr)
                        set_32bit_tls(task, FS_TLS, addr);
                        if (doit) {
                                load_TLS(&task->thread, cpu);
-                               asm volatile("movl %0,%%fs" :: "r"(FS_TLS_SEL));
+                               loadsegment(fs, FS_TLS_SEL);
                        }
                        task->thread.fsindex = FS_TLS_SEL;
                        task->thread.fs = 0;
@@ -819,7 +818,7 @@ long do_arch_prctl(struct task_struct *task, int code, unsigned long addr)
                        if (doit) {
                                /* set the selector to 0 to not confuse
                                   __switch_to */
-                               asm volatile("movl %0,%%fs" :: "r" (0));
+                               loadsegment(fs, 0);
                                ret = checking_wrmsrl(MSR_FS_BASE, addr);
                        }
                }
@@ -842,13 +841,12 @@ long do_arch_prctl(struct task_struct *task, int code, unsigned long addr)
                if (task->thread.gsindex == GS_TLS_SEL)
                        base = read_32bit_tls(task, GS_TLS);
                else if (doit) {
-                       asm("movl %%gs,%0" : "=r" (gsindex));
+                       savesegment(gs, gsindex);
                        if (gsindex)
                                rdmsrl(MSR_KERNEL_GS_BASE, base);
                        else
                                base = task->thread.gs;
-               }
-               else
+               } else
                        base = task->thread.gs;
                ret = put_user(base, (unsigned long __user *)addr);
                break;