2 * Copyright (C) 1995 Linus Torvalds
4 * Pentium III FXSR, SSE support
5 * Gareth Hughes <gareth@valinux.com>, May 2000
9 * This file handles the architecture-dependent parts of process handling..
14 #include <linux/cpu.h>
15 #include <linux/errno.h>
16 #include <linux/sched.h>
18 #include <linux/kernel.h>
20 #include <linux/elfcore.h>
21 #include <linux/smp.h>
22 #include <linux/stddef.h>
23 #include <linux/slab.h>
24 #include <linux/vmalloc.h>
25 #include <linux/user.h>
26 #include <linux/interrupt.h>
27 #include <linux/utsname.h>
28 #include <linux/delay.h>
29 #include <linux/reboot.h>
30 #include <linux/init.h>
31 #include <linux/mc146818rtc.h>
32 #include <linux/module.h>
33 #include <linux/kallsyms.h>
34 #include <linux/ptrace.h>
35 #include <linux/random.h>
36 #include <linux/personality.h>
37 #include <linux/tick.h>
38 #include <linux/percpu.h>
40 #include <asm/uaccess.h>
41 #include <asm/pgtable.h>
42 #include <asm/system.h>
45 #include <asm/processor.h>
49 #ifdef CONFIG_MATH_EMULATION
50 #include <asm/math_emu.h>
53 #include <linux/err.h>
55 #include <asm/tlbflush.h>
57 #include <asm/kdebug.h>
59 asmlinkage
void ret_from_fork(void) __asm__("ret_from_fork");
61 static int hlt_counter
;
63 unsigned long boot_option_idle_override
= 0;
64 EXPORT_SYMBOL(boot_option_idle_override
);
66 DEFINE_PER_CPU(struct task_struct
*, current_task
) = &init_task
;
67 EXPORT_PER_CPU_SYMBOL(current_task
);
69 DEFINE_PER_CPU(int, cpu_number
);
70 EXPORT_PER_CPU_SYMBOL(cpu_number
);
73 * Return saved PC of a blocked thread.
75 unsigned long thread_saved_pc(struct task_struct
*tsk
)
77 return ((unsigned long *)tsk
->thread
.sp
)[3];
81 * Powermanagement idle function, if any..
83 void (*pm_idle
)(void);
84 EXPORT_SYMBOL(pm_idle
);
86 void disable_hlt(void)
91 EXPORT_SYMBOL(disable_hlt
);
98 EXPORT_SYMBOL(enable_hlt
);
101 * We use this if we don't have any better
104 void default_idle(void)
106 if (!hlt_counter
&& boot_cpu_data
.hlt_works_ok
) {
107 current_thread_info()->status
&= ~TS_POLLING
;
109 * TS_POLLING-cleared state must be visible before we
115 if (!need_resched()) {
120 t0n
= ktime_to_ns(t0
);
121 safe_halt(); /* enables interrupts racelessly */
124 t1n
= ktime_to_ns(t1
);
125 sched_clock_idle_wakeup_event(t1n
- t0n
);
128 current_thread_info()->status
|= TS_POLLING
;
130 /* loop is done by the caller */
134 #ifdef CONFIG_APM_MODULE
135 EXPORT_SYMBOL(default_idle
);
139 * On SMP it's slightly faster (but much more power-consuming!)
140 * to poll the ->work.need_resched flag instead of waiting for the
141 * cross-CPU IPI to arrive. Use this option with caution.
143 static void poll_idle(void)
148 #ifdef CONFIG_HOTPLUG_CPU
150 /* We don't actually take CPU down, just spin without interrupts. */
151 static inline void play_dead(void)
153 /* This must be done before dead CPU ack */
158 __get_cpu_var(cpu_state
) = CPU_DEAD
;
161 * With physical CPU hotplug, we should halt the cpu
168 static inline void play_dead(void)
172 #endif /* CONFIG_HOTPLUG_CPU */
175 * The idle thread. There's no useful work to be
176 * done, so just try to conserve power and have a
177 * low exit latency (ie sit in a loop waiting for
178 * somebody to say that they'd like to reschedule)
182 int cpu
= smp_processor_id();
184 current_thread_info()->status
|= TS_POLLING
;
186 /* endless idle loop with no priority at all */
188 tick_nohz_stop_sched_tick();
189 while (!need_resched()) {
196 if (rcu_pending(cpu
))
197 rcu_check_callbacks(cpu
, 0);
202 if (cpu_is_offline(cpu
))
205 __get_cpu_var(irq_stat
).idle_timestamp
= jiffies
;
208 tick_nohz_restart_sched_tick();
209 preempt_enable_no_resched();
215 static void do_nothing(void *unused
)
220 * cpu_idle_wait - Used to ensure that all the CPUs discard old value of
221 * pm_idle and update to new pm_idle value. Required while changing pm_idle
222 * handler on SMP systems.
224 * Caller must have changed pm_idle to the new value before the call. Old
225 * pm_idle value will not be used by any CPU after the return of this function.
227 void cpu_idle_wait(void)
230 /* kick all the CPUs so that they exit out of pm_idle */
231 smp_call_function(do_nothing
, NULL
, 0, 1);
233 EXPORT_SYMBOL_GPL(cpu_idle_wait
);
236 * This uses new MONITOR/MWAIT instructions on P4 processors with PNI,
237 * which can obviate IPI to trigger checking of need_resched.
238 * We execute MONITOR against need_resched and enter optimized wait state
239 * through MWAIT. Whenever someone changes need_resched, we would be woken
240 * up from MWAIT (without an IPI).
242 * New with Core Duo processors, MWAIT can take some hints based on CPU
245 void mwait_idle_with_hints(unsigned long ax
, unsigned long cx
)
247 if (!need_resched()) {
248 __monitor((void *)¤t_thread_info()->flags
, 0, 0);
255 /* Default MONITOR/MWAIT with no hints, used for default C1 state */
256 static void mwait_idle(void)
259 mwait_idle_with_hints(0, 0);
262 static int __cpuinit
mwait_usable(const struct cpuinfo_x86
*c
)
266 /* Any C1 states supported? */
267 return c
->cpuid_level
>= 5 && ((cpuid_edx(5) >> 4) & 0xf) > 0;
270 void __cpuinit
select_idle_routine(const struct cpuinfo_x86
*c
)
276 #ifdef CONFIG_X86_SMP
277 if (pm_idle
== poll_idle
&& smp_num_siblings
> 1) {
278 printk(KERN_WARNING
"WARNING: polling idle and HT enabled,"
279 " performance may degrade.\n");
282 if (cpu_has(c
, X86_FEATURE_MWAIT
) && mwait_usable(c
)) {
284 * Skip, if setup has overridden idle.
285 * One CPU supports mwait => All CPUs supports mwait
288 printk(KERN_INFO
"using mwait in idle threads.\n");
289 pm_idle
= mwait_idle
;
295 static int __init
idle_setup(char *str
)
297 if (!strcmp(str
, "poll")) {
298 printk("using polling idle threads.\n");
300 } else if (!strcmp(str
, "mwait"))
305 boot_option_idle_override
= 1;
308 early_param("idle", idle_setup
);
310 void __show_registers(struct pt_regs
*regs
, int all
)
312 unsigned long cr0
= 0L, cr2
= 0L, cr3
= 0L, cr4
= 0L;
313 unsigned long d0
, d1
, d2
, d3
, d6
, d7
;
315 unsigned short ss
, gs
;
317 if (user_mode_vm(regs
)) {
319 ss
= regs
->ss
& 0xffff;
322 sp
= (unsigned long) (®s
->sp
);
328 printk("Pid: %d, comm: %s %s (%s %.*s)\n",
329 task_pid_nr(current
), current
->comm
,
330 print_tainted(), init_utsname()->release
,
331 (int)strcspn(init_utsname()->version
, " "),
332 init_utsname()->version
);
334 printk("EIP: %04x:[<%08lx>] EFLAGS: %08lx CPU: %d\n",
335 0xffff & regs
->cs
, regs
->ip
, regs
->flags
,
337 print_symbol("EIP is at %s\n", regs
->ip
);
339 printk("EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
340 regs
->ax
, regs
->bx
, regs
->cx
, regs
->dx
);
341 printk("ESI: %08lx EDI: %08lx EBP: %08lx ESP: %08lx\n",
342 regs
->si
, regs
->di
, regs
->bp
, sp
);
343 printk(" DS: %04x ES: %04x FS: %04x GS: %04x SS: %04x\n",
344 regs
->ds
& 0xffff, regs
->es
& 0xffff,
345 regs
->fs
& 0xffff, gs
, ss
);
353 cr4
= read_cr4_safe();
354 printk("CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n",
361 printk("DR0: %08lx DR1: %08lx DR2: %08lx DR3: %08lx\n",
366 printk("DR6: %08lx DR7: %08lx\n",
370 void show_regs(struct pt_regs
*regs
)
372 __show_registers(regs
, 1);
373 show_trace(NULL
, regs
, ®s
->sp
, regs
->bp
);
377 * This gets run with %bx containing the
378 * function to call, and %dx containing
381 extern void kernel_thread_helper(void);
384 * Create a kernel thread
386 int kernel_thread(int (*fn
)(void *), void * arg
, unsigned long flags
)
390 memset(®s
, 0, sizeof(regs
));
392 regs
.bx
= (unsigned long) fn
;
393 regs
.dx
= (unsigned long) arg
;
397 regs
.fs
= __KERNEL_PERCPU
;
399 regs
.ip
= (unsigned long) kernel_thread_helper
;
400 regs
.cs
= __KERNEL_CS
| get_kernel_rpl();
401 regs
.flags
= X86_EFLAGS_IF
| X86_EFLAGS_SF
| X86_EFLAGS_PF
| 0x2;
403 /* Ok, create the new process.. */
404 return do_fork(flags
| CLONE_VM
| CLONE_UNTRACED
, 0, ®s
, 0, NULL
, NULL
);
406 EXPORT_SYMBOL(kernel_thread
);
409 * Free current thread data structures etc..
411 void exit_thread(void)
413 /* The process may have allocated an io port bitmap... nuke it. */
414 if (unlikely(test_thread_flag(TIF_IO_BITMAP
))) {
415 struct task_struct
*tsk
= current
;
416 struct thread_struct
*t
= &tsk
->thread
;
418 struct tss_struct
*tss
= &per_cpu(init_tss
, cpu
);
420 kfree(t
->io_bitmap_ptr
);
421 t
->io_bitmap_ptr
= NULL
;
422 clear_thread_flag(TIF_IO_BITMAP
);
424 * Careful, clear this in the TSS too:
426 memset(tss
->io_bitmap
, 0xff, tss
->io_bitmap_max
);
427 t
->io_bitmap_max
= 0;
428 tss
->io_bitmap_owner
= NULL
;
429 tss
->io_bitmap_max
= 0;
430 tss
->x86_tss
.io_bitmap_base
= INVALID_IO_BITMAP_OFFSET
;
435 void flush_thread(void)
437 struct task_struct
*tsk
= current
;
439 tsk
->thread
.debugreg0
= 0;
440 tsk
->thread
.debugreg1
= 0;
441 tsk
->thread
.debugreg2
= 0;
442 tsk
->thread
.debugreg3
= 0;
443 tsk
->thread
.debugreg6
= 0;
444 tsk
->thread
.debugreg7
= 0;
445 memset(tsk
->thread
.tls_array
, 0, sizeof(tsk
->thread
.tls_array
));
446 clear_tsk_thread_flag(tsk
, TIF_DEBUG
);
448 * Forget coprocessor state..
454 void release_thread(struct task_struct
*dead_task
)
456 BUG_ON(dead_task
->mm
);
457 release_vm86_irqs(dead_task
);
461 * This gets called before we allocate a new thread and copy
462 * the current task into it.
464 void prepare_to_copy(struct task_struct
*tsk
)
469 int copy_thread(int nr
, unsigned long clone_flags
, unsigned long sp
,
470 unsigned long unused
,
471 struct task_struct
* p
, struct pt_regs
* regs
)
473 struct pt_regs
* childregs
;
474 struct task_struct
*tsk
;
477 childregs
= task_pt_regs(p
);
482 p
->thread
.sp
= (unsigned long) childregs
;
483 p
->thread
.sp0
= (unsigned long) (childregs
+1);
485 p
->thread
.ip
= (unsigned long) ret_from_fork
;
487 savesegment(gs
, p
->thread
.gs
);
490 if (unlikely(test_tsk_thread_flag(tsk
, TIF_IO_BITMAP
))) {
491 p
->thread
.io_bitmap_ptr
= kmemdup(tsk
->thread
.io_bitmap_ptr
,
492 IO_BITMAP_BYTES
, GFP_KERNEL
);
493 if (!p
->thread
.io_bitmap_ptr
) {
494 p
->thread
.io_bitmap_max
= 0;
497 set_tsk_thread_flag(p
, TIF_IO_BITMAP
);
503 * Set a new TLS for the child thread?
505 if (clone_flags
& CLONE_SETTLS
)
506 err
= do_set_thread_area(p
, -1,
507 (struct user_desc __user
*)childregs
->si
, 0);
509 if (err
&& p
->thread
.io_bitmap_ptr
) {
510 kfree(p
->thread
.io_bitmap_ptr
);
511 p
->thread
.io_bitmap_max
= 0;
516 #ifdef CONFIG_SECCOMP
517 static void hard_disable_TSC(void)
519 write_cr4(read_cr4() | X86_CR4_TSD
);
521 void disable_TSC(void)
524 if (!test_and_set_thread_flag(TIF_NOTSC
))
526 * Must flip the CPU state synchronously with
527 * TIF_NOTSC in the current running context.
532 static void hard_enable_TSC(void)
534 write_cr4(read_cr4() & ~X86_CR4_TSD
);
536 #endif /* CONFIG_SECCOMP */
539 __switch_to_xtra(struct task_struct
*prev_p
, struct task_struct
*next_p
,
540 struct tss_struct
*tss
)
542 struct thread_struct
*prev
, *next
;
543 unsigned long debugctl
;
545 prev
= &prev_p
->thread
;
546 next
= &next_p
->thread
;
548 debugctl
= prev
->debugctlmsr
;
549 if (next
->ds_area_msr
!= prev
->ds_area_msr
) {
550 /* we clear debugctl to make sure DS
551 * is not in use when we change it */
553 wrmsrl(MSR_IA32_DEBUGCTLMSR
, 0);
554 wrmsr(MSR_IA32_DS_AREA
, next
->ds_area_msr
, 0);
557 if (next
->debugctlmsr
!= debugctl
)
558 wrmsr(MSR_IA32_DEBUGCTLMSR
, next
->debugctlmsr
, 0);
560 if (test_tsk_thread_flag(next_p
, TIF_DEBUG
)) {
561 set_debugreg(next
->debugreg0
, 0);
562 set_debugreg(next
->debugreg1
, 1);
563 set_debugreg(next
->debugreg2
, 2);
564 set_debugreg(next
->debugreg3
, 3);
566 set_debugreg(next
->debugreg6
, 6);
567 set_debugreg(next
->debugreg7
, 7);
570 #ifdef CONFIG_SECCOMP
571 if (test_tsk_thread_flag(prev_p
, TIF_NOTSC
) ^
572 test_tsk_thread_flag(next_p
, TIF_NOTSC
)) {
573 /* prev and next are different */
574 if (test_tsk_thread_flag(next_p
, TIF_NOTSC
))
582 if (test_tsk_thread_flag(prev_p
, TIF_BTS_TRACE_TS
))
583 ptrace_bts_take_timestamp(prev_p
, BTS_TASK_DEPARTS
);
585 if (test_tsk_thread_flag(next_p
, TIF_BTS_TRACE_TS
))
586 ptrace_bts_take_timestamp(next_p
, BTS_TASK_ARRIVES
);
590 if (!test_tsk_thread_flag(next_p
, TIF_IO_BITMAP
)) {
592 * Disable the bitmap via an invalid offset. We still cache
593 * the previous bitmap owner and the IO bitmap contents:
595 tss
->x86_tss
.io_bitmap_base
= INVALID_IO_BITMAP_OFFSET
;
599 if (likely(next
== tss
->io_bitmap_owner
)) {
601 * Previous owner of the bitmap (hence the bitmap content)
602 * matches the next task, we dont have to do anything but
603 * to set a valid offset in the TSS:
605 tss
->x86_tss
.io_bitmap_base
= IO_BITMAP_OFFSET
;
609 * Lazy TSS's I/O bitmap copy. We set an invalid offset here
610 * and we let the task to get a GPF in case an I/O instruction
611 * is performed. The handler of the GPF will verify that the
612 * faulting task has a valid I/O bitmap and, it true, does the
613 * real copy and restart the instruction. This will save us
614 * redundant copies when the currently switched task does not
615 * perform any I/O during its timeslice.
617 tss
->x86_tss
.io_bitmap_base
= INVALID_IO_BITMAP_OFFSET_LAZY
;
621 * switch_to(x,yn) should switch tasks from x to y.
623 * We fsave/fwait so that an exception goes off at the right time
624 * (as a call from the fsave or fwait in effect) rather than to
625 * the wrong process. Lazy FP saving no longer makes any sense
626 * with modern CPU's, and this simplifies a lot of things (SMP
627 * and UP become the same).
629 * NOTE! We used to use the x86 hardware context switching. The
630 * reason for not using it any more becomes apparent when you
631 * try to recover gracefully from saved state that is no longer
632 * valid (stale segment register values in particular). With the
633 * hardware task-switch, there is no way to fix up bad state in
634 * a reasonable manner.
636 * The fact that Intel documents the hardware task-switching to
637 * be slow is a fairly red herring - this code is not noticeably
638 * faster. However, there _is_ some room for improvement here,
639 * so the performance issues may eventually be a valid point.
640 * More important, however, is the fact that this allows us much
643 * The return value (in %ax) will be the "prev" task after
644 * the task-switch, and shows up in ret_from_fork in entry.S,
647 struct task_struct
* __switch_to(struct task_struct
*prev_p
, struct task_struct
*next_p
)
649 struct thread_struct
*prev
= &prev_p
->thread
,
650 *next
= &next_p
->thread
;
651 int cpu
= smp_processor_id();
652 struct tss_struct
*tss
= &per_cpu(init_tss
, cpu
);
654 /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
656 __unlazy_fpu(prev_p
);
659 /* we're going to use this soon, after a few expensive things */
660 if (next_p
->fpu_counter
> 5)
661 prefetch(&next
->i387
.fxsave
);
669 * Save away %gs. No need to save %fs, as it was saved on the
670 * stack on entry. No need to save %es and %ds, as those are
671 * always kernel segments while inside the kernel. Doing this
672 * before setting the new TLS descriptors avoids the situation
673 * where we temporarily have non-reloadable segments in %fs
674 * and %gs. This could be an issue if the NMI handler ever
675 * used %fs or %gs (it does not today), or if the kernel is
676 * running inside of a hypervisor layer.
678 savesegment(gs
, prev
->gs
);
681 * Load the per-thread Thread-Local Storage descriptor.
686 * Restore IOPL if needed. In normal use, the flags restore
687 * in the switch assembly will handle this. But if the kernel
688 * is running virtualized at a non-zero CPL, the popf will
689 * not restore flags, so it must be done in a separate step.
691 if (get_kernel_rpl() && unlikely(prev
->iopl
!= next
->iopl
))
692 set_iopl_mask(next
->iopl
);
695 * Now maybe handle debug registers and/or IO bitmaps
697 if (unlikely(task_thread_info(prev_p
)->flags
& _TIF_WORK_CTXSW_PREV
||
698 task_thread_info(next_p
)->flags
& _TIF_WORK_CTXSW_NEXT
))
699 __switch_to_xtra(prev_p
, next_p
, tss
);
702 * Leave lazy mode, flushing any hypercalls made here.
703 * This must be done before restoring TLS segments so
704 * the GDT and LDT are properly updated, and must be
705 * done before math_state_restore, so the TS bit is up
708 arch_leave_lazy_cpu_mode();
710 /* If the task has used fpu the last 5 timeslices, just do a full
711 * restore of the math state immediately to avoid the trap; the
712 * chances of needing FPU soon are obviously high now
714 if (next_p
->fpu_counter
> 5)
715 math_state_restore();
718 * Restore %gs if needed (which is common)
720 if (prev
->gs
| next
->gs
)
721 loadsegment(gs
, next
->gs
);
723 x86_write_percpu(current_task
, next_p
);
728 asmlinkage
int sys_fork(struct pt_regs regs
)
730 return do_fork(SIGCHLD
, regs
.sp
, ®s
, 0, NULL
, NULL
);
733 asmlinkage
int sys_clone(struct pt_regs regs
)
735 unsigned long clone_flags
;
737 int __user
*parent_tidptr
, *child_tidptr
;
739 clone_flags
= regs
.bx
;
741 parent_tidptr
= (int __user
*)regs
.dx
;
742 child_tidptr
= (int __user
*)regs
.di
;
745 return do_fork(clone_flags
, newsp
, ®s
, 0, parent_tidptr
, child_tidptr
);
749 * This is trivial, and on the face of it looks like it
750 * could equally well be done in user mode.
752 * Not so, for quite unobvious reasons - register pressure.
753 * In user mode vfork() cannot have a stack frame, and if
754 * done by calling the "clone()" system call directly, you
755 * do not have enough call-clobbered registers to hold all
756 * the information you need.
758 asmlinkage
int sys_vfork(struct pt_regs regs
)
760 return do_fork(CLONE_VFORK
| CLONE_VM
| SIGCHLD
, regs
.sp
, ®s
, 0, NULL
, NULL
);
764 * sys_execve() executes a new program.
766 asmlinkage
int sys_execve(struct pt_regs regs
)
771 filename
= getname((char __user
*) regs
.bx
);
772 error
= PTR_ERR(filename
);
773 if (IS_ERR(filename
))
775 error
= do_execve(filename
,
776 (char __user
* __user
*) regs
.cx
,
777 (char __user
* __user
*) regs
.dx
,
780 /* Make sure we don't return using sysenter.. */
781 set_thread_flag(TIF_IRET
);
788 #define top_esp (THREAD_SIZE - sizeof(unsigned long))
789 #define top_ebp (THREAD_SIZE - 2*sizeof(unsigned long))
791 unsigned long get_wchan(struct task_struct
*p
)
793 unsigned long bp
, sp
, ip
;
794 unsigned long stack_page
;
796 if (!p
|| p
== current
|| p
->state
== TASK_RUNNING
)
798 stack_page
= (unsigned long)task_stack_page(p
);
800 if (!stack_page
|| sp
< stack_page
|| sp
> top_esp
+stack_page
)
802 /* include/asm-i386/system.h:switch_to() pushes bp last. */
803 bp
= *(unsigned long *) sp
;
805 if (bp
< stack_page
|| bp
> top_ebp
+stack_page
)
807 ip
= *(unsigned long *) (bp
+4);
808 if (!in_sched_functions(ip
))
810 bp
= *(unsigned long *) bp
;
811 } while (count
++ < 16);
815 unsigned long arch_align_stack(unsigned long sp
)
817 if (!(current
->personality
& ADDR_NO_RANDOMIZE
) && randomize_va_space
)
818 sp
-= get_random_int() % 8192;
822 unsigned long arch_randomize_brk(struct mm_struct
*mm
)
824 unsigned long range_end
= mm
->brk
+ 0x02000000;
825 return randomize_range(mm
->brk
, range_end
, 0) ? : mm
->brk
;