2 * linux/arch/i386/kernel/process.c
4 * Copyright (C) 1995 Linus Torvalds
6 * Pentium III FXSR, SSE support
7 * Gareth Hughes <gareth@valinux.com>, May 2000
11 * This file handles the architecture-dependent parts of process handling..
16 #include <linux/cpu.h>
17 #include <linux/errno.h>
18 #include <linux/sched.h>
20 #include <linux/kernel.h>
22 #include <linux/elfcore.h>
23 #include <linux/smp.h>
24 #include <linux/smp_lock.h>
25 #include <linux/stddef.h>
26 #include <linux/slab.h>
27 #include <linux/vmalloc.h>
28 #include <linux/user.h>
29 #include <linux/a.out.h>
30 #include <linux/interrupt.h>
31 #include <linux/config.h>
32 #include <linux/utsname.h>
33 #include <linux/delay.h>
34 #include <linux/reboot.h>
35 #include <linux/init.h>
36 #include <linux/mc146818rtc.h>
37 #include <linux/module.h>
38 #include <linux/kallsyms.h>
39 #include <linux/ptrace.h>
40 #include <linux/random.h>
41 #include <linux/kprobes.h>
43 #include <asm/uaccess.h>
44 #include <asm/pgtable.h>
45 #include <asm/system.h>
48 #include <asm/processor.h>
52 #ifdef CONFIG_MATH_EMULATION
53 #include <asm/math_emu.h>
56 #include <linux/err.h>
58 #include <asm/tlbflush.h>
61 asmlinkage void ret_from_fork(void) __asm__("ret_from_fork");
63 static int hlt_counter;
65 unsigned long boot_option_idle_override = 0;
66 EXPORT_SYMBOL(boot_option_idle_override);
69 * Return saved PC of a blocked thread.
71 unsigned long thread_saved_pc(struct task_struct *tsk)
73 return ((unsigned long *)tsk->thread.esp)[3];
77 * Powermanagement idle function, if any..
79 void (*pm_idle)(void);
80 EXPORT_SYMBOL(pm_idle);
81 static DEFINE_PER_CPU(unsigned int, cpu_idle_state);
83 void disable_hlt(void)
88 EXPORT_SYMBOL(disable_hlt);
95 EXPORT_SYMBOL(enable_hlt);
98 * We use this if we don't have any better
101 void default_idle(void)
105 if (!hlt_counter && boot_cpu_data.hlt_works_ok) {
106 clear_thread_flag(TIF_POLLING_NRFLAG);
107 smp_mb__after_clear_bit();
108 while (!need_resched()) {
115 set_thread_flag(TIF_POLLING_NRFLAG);
117 while (!need_resched())
121 #ifdef CONFIG_APM_MODULE
122 EXPORT_SYMBOL(default_idle);
126 * On SMP it's slightly faster (but much more power-consuming!)
127 * to poll the ->work.need_resched flag instead of waiting for the
128 * cross-CPU IPI to arrive. Use this option with caution.
130 static void poll_idle (void)
139 : : "i"(_TIF_NEED_RESCHED), "m" (current_thread_info()->flags));
142 #ifdef CONFIG_HOTPLUG_CPU
144 /* We don't actually take CPU down, just spin without interrupts. */
145 static inline void play_dead(void)
147 /* This must be done before dead CPU ack */
152 __get_cpu_var(cpu_state) = CPU_DEAD;
155 * With physical CPU hotplug, we should halt the cpu
162 static inline void play_dead(void)
166 #endif /* CONFIG_HOTPLUG_CPU */
169 * The idle thread. There's no useful work to be
170 * done, so just try to conserve power and have a
171 * low exit latency (ie sit in a loop waiting for
172 * somebody to say that they'd like to reschedule)
176 int cpu = smp_processor_id();
178 set_thread_flag(TIF_POLLING_NRFLAG);
180 /* endless idle loop with no priority at all */
182 while (!need_resched()) {
185 if (__get_cpu_var(cpu_idle_state))
186 __get_cpu_var(cpu_idle_state) = 0;
194 if (cpu_is_offline(cpu))
197 __get_cpu_var(irq_stat).idle_timestamp = jiffies;
200 preempt_enable_no_resched();
206 void cpu_idle_wait(void)
208 unsigned int cpu, this_cpu = get_cpu();
211 set_cpus_allowed(current, cpumask_of_cpu(this_cpu));
215 for_each_online_cpu(cpu) {
216 per_cpu(cpu_idle_state, cpu) = 1;
220 __get_cpu_var(cpu_idle_state) = 0;
225 for_each_online_cpu(cpu) {
226 if (cpu_isset(cpu, map) && !per_cpu(cpu_idle_state, cpu))
229 cpus_and(map, map, cpu_online_map);
230 } while (!cpus_empty(map));
232 EXPORT_SYMBOL_GPL(cpu_idle_wait);
235 * This uses new MONITOR/MWAIT instructions on P4 processors with PNI,
236 * which can obviate IPI to trigger checking of need_resched.
237 * We execute MONITOR against need_resched and enter optimized wait state
238 * through MWAIT. Whenever someone changes need_resched, we would be woken
239 * up from MWAIT (without an IPI).
241 static void mwait_idle(void)
245 while (!need_resched()) {
246 __monitor((void *)¤t_thread_info()->flags, 0, 0);
254 void __devinit select_idle_routine(const struct cpuinfo_x86 *c)
256 if (cpu_has(c, X86_FEATURE_MWAIT)) {
257 printk("monitor/mwait feature present.\n");
259 * Skip, if setup has overridden idle.
260 * One CPU supports mwait => All CPUs supports mwait
263 printk("using mwait in idle threads.\n");
264 pm_idle = mwait_idle;
269 static int __init idle_setup (char *str)
271 if (!strncmp(str, "poll", 4)) {
272 printk("using polling idle threads.\n");
274 #ifdef CONFIG_X86_SMP
275 if (smp_num_siblings > 1)
276 printk("WARNING: polling idle and HT enabled, performance may degrade.\n");
278 } else if (!strncmp(str, "halt", 4)) {
279 printk("using halt in idle threads.\n");
280 pm_idle = default_idle;
283 boot_option_idle_override = 1;
287 __setup("idle=", idle_setup);
289 void show_regs(struct pt_regs * regs)
291 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L;
294 printk("Pid: %d, comm: %20s\n", current->pid, current->comm);
295 printk("EIP: %04x:[<%08lx>] CPU: %d\n",0xffff & regs->xcs,regs->eip, smp_processor_id());
296 print_symbol("EIP is at %s\n", regs->eip);
299 printk(" ESP: %04x:%08lx",0xffff & regs->xss,regs->esp);
300 printk(" EFLAGS: %08lx %s (%s)\n",
301 regs->eflags, print_tainted(), system_utsname.release);
302 printk("EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
303 regs->eax,regs->ebx,regs->ecx,regs->edx);
304 printk("ESI: %08lx EDI: %08lx EBP: %08lx",
305 regs->esi, regs->edi, regs->ebp);
306 printk(" DS: %04x ES: %04x\n",
307 0xffff & regs->xds,0xffff & regs->xes);
312 cr4 = read_cr4_safe();
313 printk("CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n", cr0, cr2, cr3, cr4);
314 show_trace(NULL, ®s->esp);
318 * This gets run with %ebx containing the
319 * function to call, and %edx containing
322 extern void kernel_thread_helper(void);
323 __asm__(".section .text\n"
325 "kernel_thread_helper:\n\t"
334 * Create a kernel thread
336 int kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
340 memset(®s, 0, sizeof(regs));
342 regs.ebx = (unsigned long) fn;
343 regs.edx = (unsigned long) arg;
345 regs.xds = __USER_DS;
346 regs.xes = __USER_DS;
348 regs.eip = (unsigned long) kernel_thread_helper;
349 regs.xcs = __KERNEL_CS;
350 regs.eflags = X86_EFLAGS_IF | X86_EFLAGS_SF | X86_EFLAGS_PF | 0x2;
352 /* Ok, create the new process.. */
353 return do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0, ®s, 0, NULL, NULL);
355 EXPORT_SYMBOL(kernel_thread);
358 * Free current thread data structures etc..
360 void exit_thread(void)
362 struct task_struct *tsk = current;
363 struct thread_struct *t = &tsk->thread;
366 * Remove function-return probe instances associated with this task
367 * and put them back on the free list. Do not insert an exit probe for
368 * this function, it will be disabled by kprobe_flush_task if you do.
370 kprobe_flush_task(tsk);
372 /* The process may have allocated an io port bitmap... nuke it. */
373 if (unlikely(NULL != t->io_bitmap_ptr)) {
375 struct tss_struct *tss = &per_cpu(init_tss, cpu);
377 kfree(t->io_bitmap_ptr);
378 t->io_bitmap_ptr = NULL;
380 * Careful, clear this in the TSS too:
382 memset(tss->io_bitmap, 0xff, tss->io_bitmap_max);
383 t->io_bitmap_max = 0;
384 tss->io_bitmap_owner = NULL;
385 tss->io_bitmap_max = 0;
386 tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
391 void flush_thread(void)
393 struct task_struct *tsk = current;
395 memset(tsk->thread.debugreg, 0, sizeof(unsigned long)*8);
396 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
398 * Forget coprocessor state..
404 void release_thread(struct task_struct *dead_task)
406 BUG_ON(dead_task->mm);
407 release_vm86_irqs(dead_task);
411 * This gets called before we allocate a new thread and copy
412 * the current task into it.
414 void prepare_to_copy(struct task_struct *tsk)
419 int copy_thread(int nr, unsigned long clone_flags, unsigned long esp,
420 unsigned long unused,
421 struct task_struct * p, struct pt_regs * regs)
423 struct pt_regs * childregs;
424 struct task_struct *tsk;
427 childregs = task_pt_regs(p);
430 childregs->esp = esp;
432 p->thread.esp = (unsigned long) childregs;
433 p->thread.esp0 = (unsigned long) (childregs+1);
435 p->thread.eip = (unsigned long) ret_from_fork;
437 savesegment(fs,p->thread.fs);
438 savesegment(gs,p->thread.gs);
441 if (unlikely(NULL != tsk->thread.io_bitmap_ptr)) {
442 p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL);
443 if (!p->thread.io_bitmap_ptr) {
444 p->thread.io_bitmap_max = 0;
447 memcpy(p->thread.io_bitmap_ptr, tsk->thread.io_bitmap_ptr,
452 * Set a new TLS for the child thread?
454 if (clone_flags & CLONE_SETTLS) {
455 struct desc_struct *desc;
456 struct user_desc info;
460 if (copy_from_user(&info, (void __user *)childregs->esi, sizeof(info)))
463 if (LDT_empty(&info))
466 idx = info.entry_number;
467 if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
470 desc = p->thread.tls_array + idx - GDT_ENTRY_TLS_MIN;
471 desc->a = LDT_entry_a(&info);
472 desc->b = LDT_entry_b(&info);
477 if (err && p->thread.io_bitmap_ptr) {
478 kfree(p->thread.io_bitmap_ptr);
479 p->thread.io_bitmap_max = 0;
485 * fill in the user structure for a core dump..
487 void dump_thread(struct pt_regs * regs, struct user * dump)
491 /* changed the size calculations - should hopefully work better. lbt */
492 dump->magic = CMAGIC;
493 dump->start_code = 0;
494 dump->start_stack = regs->esp & ~(PAGE_SIZE - 1);
495 dump->u_tsize = ((unsigned long) current->mm->end_code) >> PAGE_SHIFT;
496 dump->u_dsize = ((unsigned long) (current->mm->brk + (PAGE_SIZE-1))) >> PAGE_SHIFT;
497 dump->u_dsize -= dump->u_tsize;
499 for (i = 0; i < 8; i++)
500 dump->u_debugreg[i] = current->thread.debugreg[i];
502 if (dump->start_stack < TASK_SIZE)
503 dump->u_ssize = ((unsigned long) (TASK_SIZE - dump->start_stack)) >> PAGE_SHIFT;
505 dump->regs.ebx = regs->ebx;
506 dump->regs.ecx = regs->ecx;
507 dump->regs.edx = regs->edx;
508 dump->regs.esi = regs->esi;
509 dump->regs.edi = regs->edi;
510 dump->regs.ebp = regs->ebp;
511 dump->regs.eax = regs->eax;
512 dump->regs.ds = regs->xds;
513 dump->regs.es = regs->xes;
514 savesegment(fs,dump->regs.fs);
515 savesegment(gs,dump->regs.gs);
516 dump->regs.orig_eax = regs->orig_eax;
517 dump->regs.eip = regs->eip;
518 dump->regs.cs = regs->xcs;
519 dump->regs.eflags = regs->eflags;
520 dump->regs.esp = regs->esp;
521 dump->regs.ss = regs->xss;
523 dump->u_fpvalid = dump_fpu (regs, &dump->i387);
525 EXPORT_SYMBOL(dump_thread);
528 * Capture the user space registers if the task is not running (in user space)
530 int dump_task_regs(struct task_struct *tsk, elf_gregset_t *regs)
532 struct pt_regs ptregs = *task_pt_regs(tsk);
533 ptregs.xcs &= 0xffff;
534 ptregs.xds &= 0xffff;
535 ptregs.xes &= 0xffff;
536 ptregs.xss &= 0xffff;
538 elf_core_copy_regs(regs, &ptregs);
544 handle_io_bitmap(struct thread_struct *next, struct tss_struct *tss)
546 if (!next->io_bitmap_ptr) {
548 * Disable the bitmap via an invalid offset. We still cache
549 * the previous bitmap owner and the IO bitmap contents:
551 tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
554 if (likely(next == tss->io_bitmap_owner)) {
556 * Previous owner of the bitmap (hence the bitmap content)
557 * matches the next task, we dont have to do anything but
558 * to set a valid offset in the TSS:
560 tss->io_bitmap_base = IO_BITMAP_OFFSET;
564 * Lazy TSS's I/O bitmap copy. We set an invalid offset here
565 * and we let the task to get a GPF in case an I/O instruction
566 * is performed. The handler of the GPF will verify that the
567 * faulting task has a valid I/O bitmap and, it true, does the
568 * real copy and restart the instruction. This will save us
569 * redundant copies when the currently switched task does not
570 * perform any I/O during its timeslice.
572 tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET_LAZY;
576 * This function selects if the context switch from prev to next
577 * has to tweak the TSC disable bit in the cr4.
579 static inline void disable_tsc(struct task_struct *prev_p,
580 struct task_struct *next_p)
582 struct thread_info *prev, *next;
585 * gcc should eliminate the ->thread_info dereference if
586 * has_secure_computing returns 0 at compile time (SECCOMP=n).
588 prev = task_thread_info(prev_p);
589 next = task_thread_info(next_p);
591 if (has_secure_computing(prev) || has_secure_computing(next)) {
593 if (has_secure_computing(prev) &&
594 !has_secure_computing(next)) {
595 write_cr4(read_cr4() & ~X86_CR4_TSD);
596 } else if (!has_secure_computing(prev) &&
597 has_secure_computing(next))
598 write_cr4(read_cr4() | X86_CR4_TSD);
603 * switch_to(x,yn) should switch tasks from x to y.
605 * We fsave/fwait so that an exception goes off at the right time
606 * (as a call from the fsave or fwait in effect) rather than to
607 * the wrong process. Lazy FP saving no longer makes any sense
608 * with modern CPU's, and this simplifies a lot of things (SMP
609 * and UP become the same).
611 * NOTE! We used to use the x86 hardware context switching. The
612 * reason for not using it any more becomes apparent when you
613 * try to recover gracefully from saved state that is no longer
614 * valid (stale segment register values in particular). With the
615 * hardware task-switch, there is no way to fix up bad state in
616 * a reasonable manner.
618 * The fact that Intel documents the hardware task-switching to
619 * be slow is a fairly red herring - this code is not noticeably
620 * faster. However, there _is_ some room for improvement here,
621 * so the performance issues may eventually be a valid point.
622 * More important, however, is the fact that this allows us much
625 * The return value (in %eax) will be the "prev" task after
626 * the task-switch, and shows up in ret_from_fork in entry.S,
629 struct task_struct fastcall * __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
631 struct thread_struct *prev = &prev_p->thread,
632 *next = &next_p->thread;
633 int cpu = smp_processor_id();
634 struct tss_struct *tss = &per_cpu(init_tss, cpu);
636 /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
638 __unlazy_fpu(prev_p);
643 load_esp0(tss, next);
646 * Save away %fs and %gs. No need to save %es and %ds, as
647 * those are always kernel segments while inside the kernel.
648 * Doing this before setting the new TLS descriptors avoids
649 * the situation where we temporarily have non-reloadable
650 * segments in %fs and %gs. This could be an issue if the
651 * NMI handler ever used %fs or %gs (it does not today), or
652 * if the kernel is running inside of a hypervisor layer.
654 savesegment(fs, prev->fs);
655 savesegment(gs, prev->gs);
658 * Load the per-thread Thread-Local Storage descriptor.
663 * Restore %fs and %gs if needed.
665 * Glibc normally makes %fs be zero, and %gs is one of
668 if (unlikely(prev->fs | next->fs))
669 loadsegment(fs, next->fs);
671 if (prev->gs | next->gs)
672 loadsegment(gs, next->gs);
675 * Restore IOPL if needed.
677 if (unlikely(prev->iopl != next->iopl))
678 set_iopl_mask(next->iopl);
681 * Now maybe reload the debug registers
683 if (unlikely(next->debugreg[7])) {
684 set_debugreg(next->debugreg[0], 0);
685 set_debugreg(next->debugreg[1], 1);
686 set_debugreg(next->debugreg[2], 2);
687 set_debugreg(next->debugreg[3], 3);
689 set_debugreg(next->debugreg[6], 6);
690 set_debugreg(next->debugreg[7], 7);
693 if (unlikely(prev->io_bitmap_ptr || next->io_bitmap_ptr))
694 handle_io_bitmap(next, tss);
696 disable_tsc(prev_p, next_p);
701 asmlinkage int sys_fork(struct pt_regs regs)
703 return do_fork(SIGCHLD, regs.esp, ®s, 0, NULL, NULL);
706 asmlinkage int sys_clone(struct pt_regs regs)
708 unsigned long clone_flags;
710 int __user *parent_tidptr, *child_tidptr;
712 clone_flags = regs.ebx;
714 parent_tidptr = (int __user *)regs.edx;
715 child_tidptr = (int __user *)regs.edi;
718 return do_fork(clone_flags, newsp, ®s, 0, parent_tidptr, child_tidptr);
722 * This is trivial, and on the face of it looks like it
723 * could equally well be done in user mode.
725 * Not so, for quite unobvious reasons - register pressure.
726 * In user mode vfork() cannot have a stack frame, and if
727 * done by calling the "clone()" system call directly, you
728 * do not have enough call-clobbered registers to hold all
729 * the information you need.
731 asmlinkage int sys_vfork(struct pt_regs regs)
733 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs.esp, ®s, 0, NULL, NULL);
737 * sys_execve() executes a new program.
739 asmlinkage int sys_execve(struct pt_regs regs)
744 filename = getname((char __user *) regs.ebx);
745 error = PTR_ERR(filename);
746 if (IS_ERR(filename))
748 error = do_execve(filename,
749 (char __user * __user *) regs.ecx,
750 (char __user * __user *) regs.edx,
754 current->ptrace &= ~PT_DTRACE;
755 task_unlock(current);
756 /* Make sure we don't return using sysenter.. */
757 set_thread_flag(TIF_IRET);
764 #define top_esp (THREAD_SIZE - sizeof(unsigned long))
765 #define top_ebp (THREAD_SIZE - 2*sizeof(unsigned long))
767 unsigned long get_wchan(struct task_struct *p)
769 unsigned long ebp, esp, eip;
770 unsigned long stack_page;
772 if (!p || p == current || p->state == TASK_RUNNING)
774 stack_page = (unsigned long)task_stack_page(p);
776 if (!stack_page || esp < stack_page || esp > top_esp+stack_page)
778 /* include/asm-i386/system.h:switch_to() pushes ebp last. */
779 ebp = *(unsigned long *) esp;
781 if (ebp < stack_page || ebp > top_ebp+stack_page)
783 eip = *(unsigned long *) (ebp+4);
784 if (!in_sched_functions(eip))
786 ebp = *(unsigned long *) ebp;
787 } while (count++ < 16);
790 EXPORT_SYMBOL(get_wchan);
793 * sys_alloc_thread_area: get a yet unused TLS descriptor index.
795 static int get_free_idx(void)
797 struct thread_struct *t = ¤t->thread;
800 for (idx = 0; idx < GDT_ENTRY_TLS_ENTRIES; idx++)
801 if (desc_empty(t->tls_array + idx))
802 return idx + GDT_ENTRY_TLS_MIN;
807 * Set a given TLS descriptor:
809 asmlinkage int sys_set_thread_area(struct user_desc __user *u_info)
811 struct thread_struct *t = ¤t->thread;
812 struct user_desc info;
813 struct desc_struct *desc;
816 if (copy_from_user(&info, u_info, sizeof(info)))
818 idx = info.entry_number;
821 * index -1 means the kernel should try to find and
822 * allocate an empty descriptor:
825 idx = get_free_idx();
828 if (put_user(idx, &u_info->entry_number))
832 if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
835 desc = t->tls_array + idx - GDT_ENTRY_TLS_MIN;
838 * We must not get preempted while modifying the TLS.
842 if (LDT_empty(&info)) {
846 desc->a = LDT_entry_a(&info);
847 desc->b = LDT_entry_b(&info);
857 * Get the current Thread-Local Storage area:
860 #define GET_BASE(desc) ( \
861 (((desc)->a >> 16) & 0x0000ffff) | \
862 (((desc)->b << 16) & 0x00ff0000) | \
863 ( (desc)->b & 0xff000000) )
865 #define GET_LIMIT(desc) ( \
866 ((desc)->a & 0x0ffff) | \
867 ((desc)->b & 0xf0000) )
869 #define GET_32BIT(desc) (((desc)->b >> 22) & 1)
870 #define GET_CONTENTS(desc) (((desc)->b >> 10) & 3)
871 #define GET_WRITABLE(desc) (((desc)->b >> 9) & 1)
872 #define GET_LIMIT_PAGES(desc) (((desc)->b >> 23) & 1)
873 #define GET_PRESENT(desc) (((desc)->b >> 15) & 1)
874 #define GET_USEABLE(desc) (((desc)->b >> 20) & 1)
876 asmlinkage int sys_get_thread_area(struct user_desc __user *u_info)
878 struct user_desc info;
879 struct desc_struct *desc;
882 if (get_user(idx, &u_info->entry_number))
884 if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
887 memset(&info, 0, sizeof(info));
889 desc = current->thread.tls_array + idx - GDT_ENTRY_TLS_MIN;
891 info.entry_number = idx;
892 info.base_addr = GET_BASE(desc);
893 info.limit = GET_LIMIT(desc);
894 info.seg_32bit = GET_32BIT(desc);
895 info.contents = GET_CONTENTS(desc);
896 info.read_exec_only = !GET_WRITABLE(desc);
897 info.limit_in_pages = GET_LIMIT_PAGES(desc);
898 info.seg_not_present = !GET_PRESENT(desc);
899 info.useable = GET_USEABLE(desc);
901 if (copy_to_user(u_info, &info, sizeof(info)))
906 unsigned long arch_align_stack(unsigned long sp)
908 if (randomize_va_space)
909 sp -= get_random_int() % 8192;