Merge branch 'perf/urgent' of git://git.kernel.org/pub/scm/linux/kernel/git/frederic...
[safe/jmp/linux-2.6] / arch / x86 / kernel / process.c
index cf8ee00..e7e3521 100644 (file)
@@ -9,16 +9,18 @@
 #include <linux/pm.h>
 #include <linux/clockchips.h>
 #include <linux/random.h>
+#include <linux/user-return-notifier.h>
+#include <linux/dmi.h>
+#include <linux/utsname.h>
 #include <trace/events/power.h>
+#include <linux/hw_breakpoint.h>
 #include <asm/system.h>
 #include <asm/apic.h>
 #include <asm/syscalls.h>
 #include <asm/idle.h>
 #include <asm/uaccess.h>
 #include <asm/i387.h>
-#include <asm/ds.h>
 #include <asm/debugreg.h>
-#include <asm/hw_breakpoint.h>
 
 unsigned long idle_halt;
 EXPORT_SYMBOL(idle_halt);
@@ -29,28 +31,22 @@ struct kmem_cache *task_xstate_cachep;
 
 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
 {
+       int ret;
+
        *dst = *src;
-       if (src->thread.xstate) {
-               dst->thread.xstate = kmem_cache_alloc(task_xstate_cachep,
-                                                     GFP_KERNEL);
-               if (!dst->thread.xstate)
-                       return -ENOMEM;
-               WARN_ON((unsigned long)dst->thread.xstate & 15);
-               memcpy(dst->thread.xstate, src->thread.xstate, xstate_size);
+       if (fpu_allocated(&src->thread.fpu)) {
+               memset(&dst->thread.fpu, 0, sizeof(dst->thread.fpu));
+               ret = fpu_alloc(&dst->thread.fpu);
+               if (ret)
+                       return ret;
+               fpu_copy(&dst->thread.fpu, &src->thread.fpu);
        }
        return 0;
 }
 
 void free_thread_xstate(struct task_struct *tsk)
 {
-       if (tsk->thread.xstate) {
-               kmem_cache_free(task_xstate_cachep, tsk->thread.xstate);
-               tsk->thread.xstate = NULL;
-       }
-       if (unlikely(test_tsk_thread_flag(tsk, TIF_DEBUG)))
-               flush_thread_hw_breakpoint(tsk);
-
-       WARN(tsk->thread.ds_ctx, "leaking DS context\n");
+       fpu_free(&tsk->thread.fpu);
 }
 
 void free_thread_info(struct thread_info *ti)
@@ -91,24 +87,37 @@ void exit_thread(void)
        }
 }
 
+void show_regs(struct pt_regs *regs)
+{
+       show_registers(regs);
+       show_trace(NULL, regs, (unsigned long *)kernel_stack_pointer(regs),
+                  regs->bp);
+}
+
+void show_regs_common(void)
+{
+       const char *board, *product;
+
+       board = dmi_get_system_info(DMI_BOARD_NAME);
+       if (!board)
+               board = "";
+       product = dmi_get_system_info(DMI_PRODUCT_NAME);
+       if (!product)
+               product = "";
+
+       printk(KERN_CONT "\n");
+       printk(KERN_DEFAULT "Pid: %d, comm: %.20s %s %s %.*s %s/%s\n",
+               current->pid, current->comm, print_tainted(),
+               init_utsname()->release,
+               (int)strcspn(init_utsname()->version, " "),
+               init_utsname()->version, board, product);
+}
+
 void flush_thread(void)
 {
        struct task_struct *tsk = current;
 
-#ifdef CONFIG_X86_64
-       if (test_tsk_thread_flag(tsk, TIF_ABI_PENDING)) {
-               clear_tsk_thread_flag(tsk, TIF_ABI_PENDING);
-               if (test_tsk_thread_flag(tsk, TIF_IA32)) {
-                       clear_tsk_thread_flag(tsk, TIF_IA32);
-               } else {
-                       set_tsk_thread_flag(tsk, TIF_IA32);
-                       current_thread_info()->status |= TS_COMPAT;
-               }
-       }
-#endif
-
-       if (unlikely(test_tsk_thread_flag(tsk, TIF_DEBUG)))
-               flush_thread_hw_breakpoint(tsk);
+       flush_ptrace_hw_breakpoint(tsk);
        memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
        /*
         * Forget coprocessor state..
@@ -184,11 +193,16 @@ void __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p,
        prev = &prev_p->thread;
        next = &next_p->thread;
 
-       if (test_tsk_thread_flag(next_p, TIF_DS_AREA_MSR) ||
-           test_tsk_thread_flag(prev_p, TIF_DS_AREA_MSR))
-               ds_switch_to(prev_p, next_p);
-       else if (next->debugctlmsr != prev->debugctlmsr)
-               update_debugctlmsr(next->debugctlmsr);
+       if (test_tsk_thread_flag(prev_p, TIF_BLOCKSTEP) ^
+           test_tsk_thread_flag(next_p, TIF_BLOCKSTEP)) {
+               unsigned long debugctl = get_debugctlmsr();
+
+               debugctl &= ~DEBUGCTLMSR_BTF;
+               if (test_tsk_thread_flag(next_p, TIF_BLOCKSTEP))
+                       debugctl |= DEBUGCTLMSR_BTF;
+
+               update_debugctlmsr(debugctl);
+       }
 
        if (test_tsk_thread_flag(prev_p, TIF_NOTSC) ^
            test_tsk_thread_flag(next_p, TIF_NOTSC)) {
@@ -212,6 +226,7 @@ void __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p,
                 */
                memset(tss->io_bitmap, 0xff, prev->io_bitmap_max);
        }
+       propagate_user_return_notify(prev_p, next_p);
 }
 
 int sys_fork(struct pt_regs *regs)
@@ -235,6 +250,78 @@ int sys_vfork(struct pt_regs *regs)
                       NULL, NULL);
 }
 
+long
+sys_clone(unsigned long clone_flags, unsigned long newsp,
+         void __user *parent_tid, void __user *child_tid, struct pt_regs *regs)
+{
+       if (!newsp)
+               newsp = regs->sp;
+       return do_fork(clone_flags, newsp, regs, 0, parent_tid, child_tid);
+}
+
+/*
+ * This gets run with %si containing the
+ * function to call, and %di containing
+ * the "args".
+ */
+extern void kernel_thread_helper(void);
+
+/*
+ * Create a kernel thread
+ */
+int kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
+{
+       struct pt_regs regs;
+
+       memset(&regs, 0, sizeof(regs));
+
+       regs.si = (unsigned long) fn;
+       regs.di = (unsigned long) arg;
+
+#ifdef CONFIG_X86_32
+       regs.ds = __USER_DS;
+       regs.es = __USER_DS;
+       regs.fs = __KERNEL_PERCPU;
+       regs.gs = __KERNEL_STACK_CANARY;
+#else
+       regs.ss = __KERNEL_DS;
+#endif
+
+       regs.orig_ax = -1;
+       regs.ip = (unsigned long) kernel_thread_helper;
+       regs.cs = __KERNEL_CS | get_kernel_rpl();
+       regs.flags = X86_EFLAGS_IF | 0x2;
+
+       /* Ok, create the new process.. */
+       return do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0, &regs, 0, NULL, NULL);
+}
+EXPORT_SYMBOL(kernel_thread);
+
+/*
+ * sys_execve() executes a new program.
+ */
+long sys_execve(char __user *name, char __user * __user *argv,
+               char __user * __user *envp, struct pt_regs *regs)
+{
+       long error;
+       char *filename;
+
+       filename = getname(name);
+       error = PTR_ERR(filename);
+       if (IS_ERR(filename))
+               return error;
+       error = do_execve(filename, argv, envp, regs);
+
+#ifdef CONFIG_X86_32
+       if (error == 0) {
+               /* Make sure we don't return using sysenter.. */
+                set_thread_flag(TIF_IRET);
+        }
+#endif
+
+       putname(filename);
+       return error;
+}
 
 /*
  * Idle related variables and functions
@@ -439,21 +526,39 @@ static int __cpuinit mwait_usable(const struct cpuinfo_x86 *c)
 }
 
 /*
- * Check for AMD CPUs, which have potentially C1E support
+ * Check for AMD CPUs, where APIC timer interrupt does not wake up CPU from C1e.
+ * For more information see
+ * - Erratum #400 for NPT family 0xf and family 0x10 CPUs
+ * - Erratum #365 for family 0x11 (not affected because C1e not in use)
  */
 static int __cpuinit check_c1e_idle(const struct cpuinfo_x86 *c)
 {
+       u64 val;
        if (c->x86_vendor != X86_VENDOR_AMD)
-               return 0;
-
-       if (c->x86 < 0x0F)
-               return 0;
+               goto no_c1e_idle;
 
        /* Family 0x0f models < rev F do not have C1E */
-       if (c->x86 == 0x0f && c->x86_model < 0x40)
-               return 0;
+       if (c->x86 == 0x0F && c->x86_model >= 0x40)
+               return 1;
 
-       return 1;
+       if (c->x86 == 0x10) {
+               /*
+                * check OSVW bit for CPUs that are not affected
+                * by erratum #400
+                */
+               if (cpu_has(c, X86_FEATURE_OSVW)) {
+                       rdmsrl(MSR_AMD64_OSVW_ID_LENGTH, val);
+                       if (val >= 2) {
+                               rdmsrl(MSR_AMD64_OSVW_STATUS, val);
+                               if (!(val & BIT(1)))
+                                       goto no_c1e_idle;
+                       }
+               }
+               return 1;
+       }
+
+no_c1e_idle:
+       return 0;
 }
 
 static cpumask_var_t c1e_mask;
@@ -520,7 +625,7 @@ void __cpuinit select_idle_routine(const struct cpuinfo_x86 *c)
 {
 #ifdef CONFIG_SMP
        if (pm_idle == poll_idle && smp_num_siblings > 1) {
-               printk(KERN_WARNING "WARNING: polling idle and HT enabled,"
+               printk_once(KERN_WARNING "WARNING: polling idle and HT enabled,"
                        " performance may degrade.\n");
        }
 #endif