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>
39 #include <linux/prctl.h>
41 #include <asm/uaccess.h>
42 #include <asm/pgtable.h>
43 #include <asm/system.h>
46 #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 safe_halt(); /* enables interrupts racelessly */
118 current_thread_info()->status |= TS_POLLING;
121 /* loop is done by the caller */
125 #ifdef CONFIG_APM_MODULE
126 EXPORT_SYMBOL(default_idle);
129 #ifdef CONFIG_HOTPLUG_CPU
131 /* We don't actually take CPU down, just spin without interrupts. */
132 static inline void play_dead(void)
134 /* This must be done before dead CPU ack */
139 __get_cpu_var(cpu_state) = CPU_DEAD;
142 * With physical CPU hotplug, we should halt the cpu
149 static inline void play_dead(void)
153 #endif /* CONFIG_HOTPLUG_CPU */
156 * The idle thread. There's no useful work to be
157 * done, so just try to conserve power and have a
158 * low exit latency (ie sit in a loop waiting for
159 * somebody to say that they'd like to reschedule)
163 int cpu = smp_processor_id();
165 current_thread_info()->status |= TS_POLLING;
167 /* endless idle loop with no priority at all */
169 tick_nohz_stop_sched_tick();
170 while (!need_resched()) {
177 if (rcu_pending(cpu))
178 rcu_check_callbacks(cpu, 0);
183 if (cpu_is_offline(cpu))
187 __get_cpu_var(irq_stat).idle_timestamp = jiffies;
188 /* Don't trace irqs off for idle */
189 stop_critical_timings();
191 start_critical_timings();
193 tick_nohz_restart_sched_tick();
194 preempt_enable_no_resched();
200 void __show_registers(struct pt_regs *regs, int all)
202 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L;
203 unsigned long d0, d1, d2, d3, d6, d7;
205 unsigned short ss, gs;
207 if (user_mode_vm(regs)) {
209 ss = regs->ss & 0xffff;
212 sp = (unsigned long) (®s->sp);
218 printk("Pid: %d, comm: %s %s (%s %.*s)\n",
219 task_pid_nr(current), current->comm,
220 print_tainted(), init_utsname()->release,
221 (int)strcspn(init_utsname()->version, " "),
222 init_utsname()->version);
224 printk("EIP: %04x:[<%08lx>] EFLAGS: %08lx CPU: %d\n",
225 (u16)regs->cs, regs->ip, regs->flags,
227 print_symbol("EIP is at %s\n", regs->ip);
229 printk("EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
230 regs->ax, regs->bx, regs->cx, regs->dx);
231 printk("ESI: %08lx EDI: %08lx EBP: %08lx ESP: %08lx\n",
232 regs->si, regs->di, regs->bp, sp);
233 printk(" DS: %04x ES: %04x FS: %04x GS: %04x SS: %04x\n",
234 (u16)regs->ds, (u16)regs->es, (u16)regs->fs, gs, ss);
242 cr4 = read_cr4_safe();
243 printk("CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n",
250 printk("DR0: %08lx DR1: %08lx DR2: %08lx DR3: %08lx\n",
255 printk("DR6: %08lx DR7: %08lx\n",
259 void show_regs(struct pt_regs *regs)
261 __show_registers(regs, 1);
262 show_trace(NULL, regs, ®s->sp, regs->bp);
266 * This gets run with %bx containing the
267 * function to call, and %dx containing
270 extern void kernel_thread_helper(void);
273 * Create a kernel thread
275 int kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
279 memset(®s, 0, sizeof(regs));
281 regs.bx = (unsigned long) fn;
282 regs.dx = (unsigned long) arg;
286 regs.fs = __KERNEL_PERCPU;
288 regs.ip = (unsigned long) kernel_thread_helper;
289 regs.cs = __KERNEL_CS | get_kernel_rpl();
290 regs.flags = X86_EFLAGS_IF | X86_EFLAGS_SF | X86_EFLAGS_PF | 0x2;
292 /* Ok, create the new process.. */
293 return do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0, ®s, 0, NULL, NULL);
295 EXPORT_SYMBOL(kernel_thread);
298 * Free current thread data structures etc..
300 void exit_thread(void)
302 /* The process may have allocated an io port bitmap... nuke it. */
303 if (unlikely(test_thread_flag(TIF_IO_BITMAP))) {
304 struct task_struct *tsk = current;
305 struct thread_struct *t = &tsk->thread;
307 struct tss_struct *tss = &per_cpu(init_tss, cpu);
309 kfree(t->io_bitmap_ptr);
310 t->io_bitmap_ptr = NULL;
311 clear_thread_flag(TIF_IO_BITMAP);
313 * Careful, clear this in the TSS too:
315 memset(tss->io_bitmap, 0xff, tss->io_bitmap_max);
316 t->io_bitmap_max = 0;
317 tss->io_bitmap_owner = NULL;
318 tss->io_bitmap_max = 0;
319 tss->x86_tss.io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
324 void flush_thread(void)
326 struct task_struct *tsk = current;
328 tsk->thread.debugreg0 = 0;
329 tsk->thread.debugreg1 = 0;
330 tsk->thread.debugreg2 = 0;
331 tsk->thread.debugreg3 = 0;
332 tsk->thread.debugreg6 = 0;
333 tsk->thread.debugreg7 = 0;
334 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
335 clear_tsk_thread_flag(tsk, TIF_DEBUG);
337 * Forget coprocessor state..
343 void release_thread(struct task_struct *dead_task)
345 BUG_ON(dead_task->mm);
346 release_vm86_irqs(dead_task);
350 * This gets called before we allocate a new thread and copy
351 * the current task into it.
353 void prepare_to_copy(struct task_struct *tsk)
358 int copy_thread(int nr, unsigned long clone_flags, unsigned long sp,
359 unsigned long unused,
360 struct task_struct * p, struct pt_regs * regs)
362 struct pt_regs * childregs;
363 struct task_struct *tsk;
366 childregs = task_pt_regs(p);
371 p->thread.sp = (unsigned long) childregs;
372 p->thread.sp0 = (unsigned long) (childregs+1);
374 p->thread.ip = (unsigned long) ret_from_fork;
376 savesegment(gs, p->thread.gs);
379 if (unlikely(test_tsk_thread_flag(tsk, TIF_IO_BITMAP))) {
380 p->thread.io_bitmap_ptr = kmemdup(tsk->thread.io_bitmap_ptr,
381 IO_BITMAP_BYTES, GFP_KERNEL);
382 if (!p->thread.io_bitmap_ptr) {
383 p->thread.io_bitmap_max = 0;
386 set_tsk_thread_flag(p, TIF_IO_BITMAP);
392 * Set a new TLS for the child thread?
394 if (clone_flags & CLONE_SETTLS)
395 err = do_set_thread_area(p, -1,
396 (struct user_desc __user *)childregs->si, 0);
398 if (err && p->thread.io_bitmap_ptr) {
399 kfree(p->thread.io_bitmap_ptr);
400 p->thread.io_bitmap_max = 0;
406 start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp)
408 __asm__("movl %0, %%gs" :: "r"(0));
411 regs->ds = __USER_DS;
412 regs->es = __USER_DS;
413 regs->ss = __USER_DS;
414 regs->cs = __USER_CS;
418 * Free the old FP and other extended state
420 free_thread_xstate(current);
422 EXPORT_SYMBOL_GPL(start_thread);
424 static void hard_disable_TSC(void)
426 write_cr4(read_cr4() | X86_CR4_TSD);
429 void disable_TSC(void)
432 if (!test_and_set_thread_flag(TIF_NOTSC))
434 * Must flip the CPU state synchronously with
435 * TIF_NOTSC in the current running context.
441 static void hard_enable_TSC(void)
443 write_cr4(read_cr4() & ~X86_CR4_TSD);
446 static void enable_TSC(void)
449 if (test_and_clear_thread_flag(TIF_NOTSC))
451 * Must flip the CPU state synchronously with
452 * TIF_NOTSC in the current running context.
458 int get_tsc_mode(unsigned long adr)
462 if (test_thread_flag(TIF_NOTSC))
463 val = PR_TSC_SIGSEGV;
467 return put_user(val, (unsigned int __user *)adr);
470 int set_tsc_mode(unsigned int val)
472 if (val == PR_TSC_SIGSEGV)
474 else if (val == PR_TSC_ENABLE)
483 __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p,
484 struct tss_struct *tss)
486 struct thread_struct *prev, *next;
487 unsigned long debugctl;
489 prev = &prev_p->thread;
490 next = &next_p->thread;
492 debugctl = prev->debugctlmsr;
493 if (next->ds_area_msr != prev->ds_area_msr) {
494 /* we clear debugctl to make sure DS
495 * is not in use when we change it */
497 update_debugctlmsr(0);
498 wrmsr(MSR_IA32_DS_AREA, next->ds_area_msr, 0);
501 if (next->debugctlmsr != debugctl)
502 update_debugctlmsr(next->debugctlmsr);
504 if (test_tsk_thread_flag(next_p, TIF_DEBUG)) {
505 set_debugreg(next->debugreg0, 0);
506 set_debugreg(next->debugreg1, 1);
507 set_debugreg(next->debugreg2, 2);
508 set_debugreg(next->debugreg3, 3);
510 set_debugreg(next->debugreg6, 6);
511 set_debugreg(next->debugreg7, 7);
514 if (test_tsk_thread_flag(prev_p, TIF_NOTSC) ^
515 test_tsk_thread_flag(next_p, TIF_NOTSC)) {
516 /* prev and next are different */
517 if (test_tsk_thread_flag(next_p, TIF_NOTSC))
524 if (test_tsk_thread_flag(prev_p, TIF_BTS_TRACE_TS))
525 ptrace_bts_take_timestamp(prev_p, BTS_TASK_DEPARTS);
527 if (test_tsk_thread_flag(next_p, TIF_BTS_TRACE_TS))
528 ptrace_bts_take_timestamp(next_p, BTS_TASK_ARRIVES);
532 if (!test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
534 * Disable the bitmap via an invalid offset. We still cache
535 * the previous bitmap owner and the IO bitmap contents:
537 tss->x86_tss.io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
541 if (likely(next == tss->io_bitmap_owner)) {
543 * Previous owner of the bitmap (hence the bitmap content)
544 * matches the next task, we dont have to do anything but
545 * to set a valid offset in the TSS:
547 tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET;
551 * Lazy TSS's I/O bitmap copy. We set an invalid offset here
552 * and we let the task to get a GPF in case an I/O instruction
553 * is performed. The handler of the GPF will verify that the
554 * faulting task has a valid I/O bitmap and, it true, does the
555 * real copy and restart the instruction. This will save us
556 * redundant copies when the currently switched task does not
557 * perform any I/O during its timeslice.
559 tss->x86_tss.io_bitmap_base = INVALID_IO_BITMAP_OFFSET_LAZY;
563 * switch_to(x,yn) should switch tasks from x to y.
565 * We fsave/fwait so that an exception goes off at the right time
566 * (as a call from the fsave or fwait in effect) rather than to
567 * the wrong process. Lazy FP saving no longer makes any sense
568 * with modern CPU's, and this simplifies a lot of things (SMP
569 * and UP become the same).
571 * NOTE! We used to use the x86 hardware context switching. The
572 * reason for not using it any more becomes apparent when you
573 * try to recover gracefully from saved state that is no longer
574 * valid (stale segment register values in particular). With the
575 * hardware task-switch, there is no way to fix up bad state in
576 * a reasonable manner.
578 * The fact that Intel documents the hardware task-switching to
579 * be slow is a fairly red herring - this code is not noticeably
580 * faster. However, there _is_ some room for improvement here,
581 * so the performance issues may eventually be a valid point.
582 * More important, however, is the fact that this allows us much
585 * The return value (in %ax) will be the "prev" task after
586 * the task-switch, and shows up in ret_from_fork in entry.S,
589 struct task_struct * __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
591 struct thread_struct *prev = &prev_p->thread,
592 *next = &next_p->thread;
593 int cpu = smp_processor_id();
594 struct tss_struct *tss = &per_cpu(init_tss, cpu);
596 /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
598 __unlazy_fpu(prev_p);
601 /* we're going to use this soon, after a few expensive things */
602 if (next_p->fpu_counter > 5)
603 prefetch(next->xstate);
611 * Save away %gs. No need to save %fs, as it was saved on the
612 * stack on entry. No need to save %es and %ds, as those are
613 * always kernel segments while inside the kernel. Doing this
614 * before setting the new TLS descriptors avoids the situation
615 * where we temporarily have non-reloadable segments in %fs
616 * and %gs. This could be an issue if the NMI handler ever
617 * used %fs or %gs (it does not today), or if the kernel is
618 * running inside of a hypervisor layer.
620 savesegment(gs, prev->gs);
623 * Load the per-thread Thread-Local Storage descriptor.
628 * Restore IOPL if needed. In normal use, the flags restore
629 * in the switch assembly will handle this. But if the kernel
630 * is running virtualized at a non-zero CPL, the popf will
631 * not restore flags, so it must be done in a separate step.
633 if (get_kernel_rpl() && unlikely(prev->iopl != next->iopl))
634 set_iopl_mask(next->iopl);
637 * Now maybe handle debug registers and/or IO bitmaps
639 if (unlikely(task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV ||
640 task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT))
641 __switch_to_xtra(prev_p, next_p, tss);
644 * Leave lazy mode, flushing any hypercalls made here.
645 * This must be done before restoring TLS segments so
646 * the GDT and LDT are properly updated, and must be
647 * done before math_state_restore, so the TS bit is up
650 arch_leave_lazy_cpu_mode();
652 /* If the task has used fpu the last 5 timeslices, just do a full
653 * restore of the math state immediately to avoid the trap; the
654 * chances of needing FPU soon are obviously high now
656 if (next_p->fpu_counter > 5)
657 math_state_restore();
660 * Restore %gs if needed (which is common)
662 if (prev->gs | next->gs)
663 loadsegment(gs, next->gs);
665 x86_write_percpu(current_task, next_p);
670 asmlinkage int sys_fork(struct pt_regs regs)
672 return do_fork(SIGCHLD, regs.sp, ®s, 0, NULL, NULL);
675 asmlinkage int sys_clone(struct pt_regs regs)
677 unsigned long clone_flags;
679 int __user *parent_tidptr, *child_tidptr;
681 clone_flags = regs.bx;
683 parent_tidptr = (int __user *)regs.dx;
684 child_tidptr = (int __user *)regs.di;
687 return do_fork(clone_flags, newsp, ®s, 0, parent_tidptr, child_tidptr);
691 * This is trivial, and on the face of it looks like it
692 * could equally well be done in user mode.
694 * Not so, for quite unobvious reasons - register pressure.
695 * In user mode vfork() cannot have a stack frame, and if
696 * done by calling the "clone()" system call directly, you
697 * do not have enough call-clobbered registers to hold all
698 * the information you need.
700 asmlinkage int sys_vfork(struct pt_regs regs)
702 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs.sp, ®s, 0, NULL, NULL);
706 * sys_execve() executes a new program.
708 asmlinkage int sys_execve(struct pt_regs regs)
713 filename = getname((char __user *) regs.bx);
714 error = PTR_ERR(filename);
715 if (IS_ERR(filename))
717 error = do_execve(filename,
718 (char __user * __user *) regs.cx,
719 (char __user * __user *) regs.dx,
722 /* Make sure we don't return using sysenter.. */
723 set_thread_flag(TIF_IRET);
730 #define top_esp (THREAD_SIZE - sizeof(unsigned long))
731 #define top_ebp (THREAD_SIZE - 2*sizeof(unsigned long))
733 unsigned long get_wchan(struct task_struct *p)
735 unsigned long bp, sp, ip;
736 unsigned long stack_page;
738 if (!p || p == current || p->state == TASK_RUNNING)
740 stack_page = (unsigned long)task_stack_page(p);
742 if (!stack_page || sp < stack_page || sp > top_esp+stack_page)
744 /* include/asm-i386/system.h:switch_to() pushes bp last. */
745 bp = *(unsigned long *) sp;
747 if (bp < stack_page || bp > top_ebp+stack_page)
749 ip = *(unsigned long *) (bp+4);
750 if (!in_sched_functions(ip))
752 bp = *(unsigned long *) bp;
753 } while (count++ < 16);
757 unsigned long arch_align_stack(unsigned long sp)
759 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
760 sp -= get_random_int() % 8192;
764 unsigned long arch_randomize_brk(struct mm_struct *mm)
766 unsigned long range_end = mm->brk + 0x02000000;
767 return randomize_range(mm->brk, range_end, 0) ? : mm->brk;