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>
42 #include <asm/uaccess.h>
43 #include <asm/pgtable.h>
44 #include <asm/system.h>
47 #include <asm/processor.h>
51 #ifdef CONFIG_MATH_EMULATION
52 #include <asm/math_emu.h>
55 #include <linux/err.h>
57 #include <asm/tlbflush.h>
60 asmlinkage void ret_from_fork(void) __asm__("ret_from_fork");
62 static int hlt_counter;
64 unsigned long boot_option_idle_override = 0;
65 EXPORT_SYMBOL(boot_option_idle_override);
68 * Return saved PC of a blocked thread.
70 unsigned long thread_saved_pc(struct task_struct *tsk)
72 return ((unsigned long *)tsk->thread.esp)[3];
76 * Powermanagement idle function, if any..
78 void (*pm_idle)(void);
79 EXPORT_SYMBOL(pm_idle);
80 static DEFINE_PER_CPU(unsigned int, cpu_idle_state);
82 void disable_hlt(void)
87 EXPORT_SYMBOL(disable_hlt);
94 EXPORT_SYMBOL(enable_hlt);
97 * We use this if we don't have any better
100 void default_idle(void)
104 if (!hlt_counter && boot_cpu_data.hlt_works_ok) {
105 current_thread_info()->status &= ~TS_POLLING;
106 smp_mb__after_clear_bit();
107 while (!need_resched()) {
114 current_thread_info()->status |= TS_POLLING;
116 while (!need_resched())
120 #ifdef CONFIG_APM_MODULE
121 EXPORT_SYMBOL(default_idle);
125 * On SMP it's slightly faster (but much more power-consuming!)
126 * to poll the ->work.need_resched flag instead of waiting for the
127 * cross-CPU IPI to arrive. Use this option with caution.
129 static void poll_idle (void)
138 : : "i"(_TIF_NEED_RESCHED), "m" (current_thread_info()->flags));
141 #ifdef CONFIG_HOTPLUG_CPU
143 /* We don't actually take CPU down, just spin without interrupts. */
144 static inline void play_dead(void)
146 /* This must be done before dead CPU ack */
151 __get_cpu_var(cpu_state) = CPU_DEAD;
154 * With physical CPU hotplug, we should halt the cpu
161 static inline void play_dead(void)
165 #endif /* CONFIG_HOTPLUG_CPU */
168 * The idle thread. There's no useful work to be
169 * done, so just try to conserve power and have a
170 * low exit latency (ie sit in a loop waiting for
171 * somebody to say that they'd like to reschedule)
175 int cpu = smp_processor_id();
177 current_thread_info()->status |= TS_POLLING;
179 /* endless idle loop with no priority at all */
181 while (!need_resched()) {
184 if (__get_cpu_var(cpu_idle_state))
185 __get_cpu_var(cpu_idle_state) = 0;
193 if (cpu_is_offline(cpu))
196 __get_cpu_var(irq_stat).idle_timestamp = jiffies;
199 preempt_enable_no_resched();
205 void cpu_idle_wait(void)
207 unsigned int cpu, this_cpu = get_cpu();
210 set_cpus_allowed(current, cpumask_of_cpu(this_cpu));
214 for_each_online_cpu(cpu) {
215 per_cpu(cpu_idle_state, cpu) = 1;
219 __get_cpu_var(cpu_idle_state) = 0;
224 for_each_online_cpu(cpu) {
225 if (cpu_isset(cpu, map) && !per_cpu(cpu_idle_state, cpu))
228 cpus_and(map, map, cpu_online_map);
229 } while (!cpus_empty(map));
231 EXPORT_SYMBOL_GPL(cpu_idle_wait);
234 * This uses new MONITOR/MWAIT instructions on P4 processors with PNI,
235 * which can obviate IPI to trigger checking of need_resched.
236 * We execute MONITOR against need_resched and enter optimized wait state
237 * through MWAIT. Whenever someone changes need_resched, we would be woken
238 * up from MWAIT (without an IPI).
240 static void mwait_idle(void)
244 while (!need_resched()) {
245 __monitor((void *)¤t_thread_info()->flags, 0, 0);
253 void __devinit select_idle_routine(const struct cpuinfo_x86 *c)
255 if (cpu_has(c, X86_FEATURE_MWAIT)) {
256 printk("monitor/mwait feature present.\n");
258 * Skip, if setup has overridden idle.
259 * One CPU supports mwait => All CPUs supports mwait
262 printk("using mwait in idle threads.\n");
263 pm_idle = mwait_idle;
268 static int __init idle_setup (char *str)
270 if (!strncmp(str, "poll", 4)) {
271 printk("using polling idle threads.\n");
273 #ifdef CONFIG_X86_SMP
274 if (smp_num_siblings > 1)
275 printk("WARNING: polling idle and HT enabled, performance may degrade.\n");
277 } else if (!strncmp(str, "halt", 4)) {
278 printk("using halt in idle threads.\n");
279 pm_idle = default_idle;
282 boot_option_idle_override = 1;
286 __setup("idle=", idle_setup);
288 void show_regs(struct pt_regs * regs)
290 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L;
293 printk("Pid: %d, comm: %20s\n", current->pid, current->comm);
294 printk("EIP: %04x:[<%08lx>] CPU: %d\n",0xffff & regs->xcs,regs->eip, smp_processor_id());
295 print_symbol("EIP is at %s\n", regs->eip);
297 if (user_mode_vm(regs))
298 printk(" ESP: %04x:%08lx",0xffff & regs->xss,regs->esp);
299 printk(" EFLAGS: %08lx %s (%s %.*s)\n",
300 regs->eflags, print_tainted(), system_utsname.release,
301 (int)strcspn(system_utsname.version, " "),
302 system_utsname.version);
303 printk("EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
304 regs->eax,regs->ebx,regs->ecx,regs->edx);
305 printk("ESI: %08lx EDI: %08lx EBP: %08lx",
306 regs->esi, regs->edi, regs->ebp);
307 printk(" DS: %04x ES: %04x\n",
308 0xffff & regs->xds,0xffff & regs->xes);
313 cr4 = read_cr4_safe();
314 printk("CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n", cr0, cr2, cr3, cr4);
315 show_trace(NULL, regs, ®s->esp);
319 * This gets run with %ebx containing the
320 * function to call, and %edx containing
323 extern void kernel_thread_helper(void);
324 __asm__(".section .text\n"
326 "kernel_thread_helper:\n\t"
335 * Create a kernel thread
337 int kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
341 memset(®s, 0, sizeof(regs));
343 regs.ebx = (unsigned long) fn;
344 regs.edx = (unsigned long) arg;
346 regs.xds = __USER_DS;
347 regs.xes = __USER_DS;
349 regs.eip = (unsigned long) kernel_thread_helper;
350 regs.xcs = __KERNEL_CS;
351 regs.eflags = X86_EFLAGS_IF | X86_EFLAGS_SF | X86_EFLAGS_PF | 0x2;
353 /* Ok, create the new process.. */
354 return do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0, ®s, 0, NULL, NULL);
356 EXPORT_SYMBOL(kernel_thread);
359 * Free current thread data structures etc..
361 void exit_thread(void)
363 struct task_struct *tsk = current;
364 struct thread_struct *t = &tsk->thread;
366 /* The process may have allocated an io port bitmap... nuke it. */
367 if (unlikely(NULL != t->io_bitmap_ptr)) {
369 struct tss_struct *tss = &per_cpu(init_tss, cpu);
371 kfree(t->io_bitmap_ptr);
372 t->io_bitmap_ptr = NULL;
374 * Careful, clear this in the TSS too:
376 memset(tss->io_bitmap, 0xff, tss->io_bitmap_max);
377 t->io_bitmap_max = 0;
378 tss->io_bitmap_owner = NULL;
379 tss->io_bitmap_max = 0;
380 tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
385 void flush_thread(void)
387 struct task_struct *tsk = current;
389 memset(tsk->thread.debugreg, 0, sizeof(unsigned long)*8);
390 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
392 * Forget coprocessor state..
398 void release_thread(struct task_struct *dead_task)
400 BUG_ON(dead_task->mm);
401 release_vm86_irqs(dead_task);
405 * This gets called before we allocate a new thread and copy
406 * the current task into it.
408 void prepare_to_copy(struct task_struct *tsk)
413 int copy_thread(int nr, unsigned long clone_flags, unsigned long esp,
414 unsigned long unused,
415 struct task_struct * p, struct pt_regs * regs)
417 struct pt_regs * childregs;
418 struct task_struct *tsk;
421 childregs = task_pt_regs(p);
424 childregs->esp = esp;
426 p->thread.esp = (unsigned long) childregs;
427 p->thread.esp0 = (unsigned long) (childregs+1);
429 p->thread.eip = (unsigned long) ret_from_fork;
431 savesegment(fs,p->thread.fs);
432 savesegment(gs,p->thread.gs);
435 if (unlikely(NULL != tsk->thread.io_bitmap_ptr)) {
436 p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL);
437 if (!p->thread.io_bitmap_ptr) {
438 p->thread.io_bitmap_max = 0;
441 memcpy(p->thread.io_bitmap_ptr, tsk->thread.io_bitmap_ptr,
446 * Set a new TLS for the child thread?
448 if (clone_flags & CLONE_SETTLS) {
449 struct desc_struct *desc;
450 struct user_desc info;
454 if (copy_from_user(&info, (void __user *)childregs->esi, sizeof(info)))
457 if (LDT_empty(&info))
460 idx = info.entry_number;
461 if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
464 desc = p->thread.tls_array + idx - GDT_ENTRY_TLS_MIN;
465 desc->a = LDT_entry_a(&info);
466 desc->b = LDT_entry_b(&info);
471 if (err && p->thread.io_bitmap_ptr) {
472 kfree(p->thread.io_bitmap_ptr);
473 p->thread.io_bitmap_max = 0;
479 * fill in the user structure for a core dump..
481 void dump_thread(struct pt_regs * regs, struct user * dump)
485 /* changed the size calculations - should hopefully work better. lbt */
486 dump->magic = CMAGIC;
487 dump->start_code = 0;
488 dump->start_stack = regs->esp & ~(PAGE_SIZE - 1);
489 dump->u_tsize = ((unsigned long) current->mm->end_code) >> PAGE_SHIFT;
490 dump->u_dsize = ((unsigned long) (current->mm->brk + (PAGE_SIZE-1))) >> PAGE_SHIFT;
491 dump->u_dsize -= dump->u_tsize;
493 for (i = 0; i < 8; i++)
494 dump->u_debugreg[i] = current->thread.debugreg[i];
496 if (dump->start_stack < TASK_SIZE)
497 dump->u_ssize = ((unsigned long) (TASK_SIZE - dump->start_stack)) >> PAGE_SHIFT;
499 dump->regs.ebx = regs->ebx;
500 dump->regs.ecx = regs->ecx;
501 dump->regs.edx = regs->edx;
502 dump->regs.esi = regs->esi;
503 dump->regs.edi = regs->edi;
504 dump->regs.ebp = regs->ebp;
505 dump->regs.eax = regs->eax;
506 dump->regs.ds = regs->xds;
507 dump->regs.es = regs->xes;
508 savesegment(fs,dump->regs.fs);
509 savesegment(gs,dump->regs.gs);
510 dump->regs.orig_eax = regs->orig_eax;
511 dump->regs.eip = regs->eip;
512 dump->regs.cs = regs->xcs;
513 dump->regs.eflags = regs->eflags;
514 dump->regs.esp = regs->esp;
515 dump->regs.ss = regs->xss;
517 dump->u_fpvalid = dump_fpu (regs, &dump->i387);
519 EXPORT_SYMBOL(dump_thread);
522 * Capture the user space registers if the task is not running (in user space)
524 int dump_task_regs(struct task_struct *tsk, elf_gregset_t *regs)
526 struct pt_regs ptregs = *task_pt_regs(tsk);
527 ptregs.xcs &= 0xffff;
528 ptregs.xds &= 0xffff;
529 ptregs.xes &= 0xffff;
530 ptregs.xss &= 0xffff;
532 elf_core_copy_regs(regs, &ptregs);
538 handle_io_bitmap(struct thread_struct *next, struct tss_struct *tss)
540 if (!next->io_bitmap_ptr) {
542 * Disable the bitmap via an invalid offset. We still cache
543 * the previous bitmap owner and the IO bitmap contents:
545 tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
548 if (likely(next == tss->io_bitmap_owner)) {
550 * Previous owner of the bitmap (hence the bitmap content)
551 * matches the next task, we dont have to do anything but
552 * to set a valid offset in the TSS:
554 tss->io_bitmap_base = IO_BITMAP_OFFSET;
558 * Lazy TSS's I/O bitmap copy. We set an invalid offset here
559 * and we let the task to get a GPF in case an I/O instruction
560 * is performed. The handler of the GPF will verify that the
561 * faulting task has a valid I/O bitmap and, it true, does the
562 * real copy and restart the instruction. This will save us
563 * redundant copies when the currently switched task does not
564 * perform any I/O during its timeslice.
566 tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET_LAZY;
570 * This function selects if the context switch from prev to next
571 * has to tweak the TSC disable bit in the cr4.
573 static inline void disable_tsc(struct task_struct *prev_p,
574 struct task_struct *next_p)
576 struct thread_info *prev, *next;
579 * gcc should eliminate the ->thread_info dereference if
580 * has_secure_computing returns 0 at compile time (SECCOMP=n).
582 prev = task_thread_info(prev_p);
583 next = task_thread_info(next_p);
585 if (has_secure_computing(prev) || has_secure_computing(next)) {
587 if (has_secure_computing(prev) &&
588 !has_secure_computing(next)) {
589 write_cr4(read_cr4() & ~X86_CR4_TSD);
590 } else if (!has_secure_computing(prev) &&
591 has_secure_computing(next))
592 write_cr4(read_cr4() | X86_CR4_TSD);
597 * switch_to(x,yn) should switch tasks from x to y.
599 * We fsave/fwait so that an exception goes off at the right time
600 * (as a call from the fsave or fwait in effect) rather than to
601 * the wrong process. Lazy FP saving no longer makes any sense
602 * with modern CPU's, and this simplifies a lot of things (SMP
603 * and UP become the same).
605 * NOTE! We used to use the x86 hardware context switching. The
606 * reason for not using it any more becomes apparent when you
607 * try to recover gracefully from saved state that is no longer
608 * valid (stale segment register values in particular). With the
609 * hardware task-switch, there is no way to fix up bad state in
610 * a reasonable manner.
612 * The fact that Intel documents the hardware task-switching to
613 * be slow is a fairly red herring - this code is not noticeably
614 * faster. However, there _is_ some room for improvement here,
615 * so the performance issues may eventually be a valid point.
616 * More important, however, is the fact that this allows us much
619 * The return value (in %eax) will be the "prev" task after
620 * the task-switch, and shows up in ret_from_fork in entry.S,
623 struct task_struct fastcall * __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
625 struct thread_struct *prev = &prev_p->thread,
626 *next = &next_p->thread;
627 int cpu = smp_processor_id();
628 struct tss_struct *tss = &per_cpu(init_tss, cpu);
630 /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
632 __unlazy_fpu(prev_p);
637 load_esp0(tss, next);
640 * Save away %fs and %gs. No need to save %es and %ds, as
641 * those are always kernel segments while inside the kernel.
642 * Doing this before setting the new TLS descriptors avoids
643 * the situation where we temporarily have non-reloadable
644 * segments in %fs and %gs. This could be an issue if the
645 * NMI handler ever used %fs or %gs (it does not today), or
646 * if the kernel is running inside of a hypervisor layer.
648 savesegment(fs, prev->fs);
649 savesegment(gs, prev->gs);
652 * Load the per-thread Thread-Local Storage descriptor.
657 * Restore %fs and %gs if needed.
659 * Glibc normally makes %fs be zero, and %gs is one of
662 if (unlikely(prev->fs | next->fs))
663 loadsegment(fs, next->fs);
665 if (prev->gs | next->gs)
666 loadsegment(gs, next->gs);
669 * Restore IOPL if needed.
671 if (unlikely(prev->iopl != next->iopl))
672 set_iopl_mask(next->iopl);
675 * Now maybe reload the debug registers
677 if (unlikely(next->debugreg[7])) {
678 set_debugreg(next->debugreg[0], 0);
679 set_debugreg(next->debugreg[1], 1);
680 set_debugreg(next->debugreg[2], 2);
681 set_debugreg(next->debugreg[3], 3);
683 set_debugreg(next->debugreg[6], 6);
684 set_debugreg(next->debugreg[7], 7);
687 if (unlikely(prev->io_bitmap_ptr || next->io_bitmap_ptr))
688 handle_io_bitmap(next, tss);
690 disable_tsc(prev_p, next_p);
695 asmlinkage int sys_fork(struct pt_regs regs)
697 return do_fork(SIGCHLD, regs.esp, ®s, 0, NULL, NULL);
700 asmlinkage int sys_clone(struct pt_regs regs)
702 unsigned long clone_flags;
704 int __user *parent_tidptr, *child_tidptr;
706 clone_flags = regs.ebx;
708 parent_tidptr = (int __user *)regs.edx;
709 child_tidptr = (int __user *)regs.edi;
712 return do_fork(clone_flags, newsp, ®s, 0, parent_tidptr, child_tidptr);
716 * This is trivial, and on the face of it looks like it
717 * could equally well be done in user mode.
719 * Not so, for quite unobvious reasons - register pressure.
720 * In user mode vfork() cannot have a stack frame, and if
721 * done by calling the "clone()" system call directly, you
722 * do not have enough call-clobbered registers to hold all
723 * the information you need.
725 asmlinkage int sys_vfork(struct pt_regs regs)
727 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs.esp, ®s, 0, NULL, NULL);
731 * sys_execve() executes a new program.
733 asmlinkage int sys_execve(struct pt_regs regs)
738 filename = getname((char __user *) regs.ebx);
739 error = PTR_ERR(filename);
740 if (IS_ERR(filename))
742 error = do_execve(filename,
743 (char __user * __user *) regs.ecx,
744 (char __user * __user *) regs.edx,
748 current->ptrace &= ~PT_DTRACE;
749 task_unlock(current);
750 /* Make sure we don't return using sysenter.. */
751 set_thread_flag(TIF_IRET);
758 #define top_esp (THREAD_SIZE - sizeof(unsigned long))
759 #define top_ebp (THREAD_SIZE - 2*sizeof(unsigned long))
761 unsigned long get_wchan(struct task_struct *p)
763 unsigned long ebp, esp, eip;
764 unsigned long stack_page;
766 if (!p || p == current || p->state == TASK_RUNNING)
768 stack_page = (unsigned long)task_stack_page(p);
770 if (!stack_page || esp < stack_page || esp > top_esp+stack_page)
772 /* include/asm-i386/system.h:switch_to() pushes ebp last. */
773 ebp = *(unsigned long *) esp;
775 if (ebp < stack_page || ebp > top_ebp+stack_page)
777 eip = *(unsigned long *) (ebp+4);
778 if (!in_sched_functions(eip))
780 ebp = *(unsigned long *) ebp;
781 } while (count++ < 16);
786 * sys_alloc_thread_area: get a yet unused TLS descriptor index.
788 static int get_free_idx(void)
790 struct thread_struct *t = ¤t->thread;
793 for (idx = 0; idx < GDT_ENTRY_TLS_ENTRIES; idx++)
794 if (desc_empty(t->tls_array + idx))
795 return idx + GDT_ENTRY_TLS_MIN;
800 * Set a given TLS descriptor:
802 asmlinkage int sys_set_thread_area(struct user_desc __user *u_info)
804 struct thread_struct *t = ¤t->thread;
805 struct user_desc info;
806 struct desc_struct *desc;
809 if (copy_from_user(&info, u_info, sizeof(info)))
811 idx = info.entry_number;
814 * index -1 means the kernel should try to find and
815 * allocate an empty descriptor:
818 idx = get_free_idx();
821 if (put_user(idx, &u_info->entry_number))
825 if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
828 desc = t->tls_array + idx - GDT_ENTRY_TLS_MIN;
831 * We must not get preempted while modifying the TLS.
835 if (LDT_empty(&info)) {
839 desc->a = LDT_entry_a(&info);
840 desc->b = LDT_entry_b(&info);
850 * Get the current Thread-Local Storage area:
853 #define GET_BASE(desc) ( \
854 (((desc)->a >> 16) & 0x0000ffff) | \
855 (((desc)->b << 16) & 0x00ff0000) | \
856 ( (desc)->b & 0xff000000) )
858 #define GET_LIMIT(desc) ( \
859 ((desc)->a & 0x0ffff) | \
860 ((desc)->b & 0xf0000) )
862 #define GET_32BIT(desc) (((desc)->b >> 22) & 1)
863 #define GET_CONTENTS(desc) (((desc)->b >> 10) & 3)
864 #define GET_WRITABLE(desc) (((desc)->b >> 9) & 1)
865 #define GET_LIMIT_PAGES(desc) (((desc)->b >> 23) & 1)
866 #define GET_PRESENT(desc) (((desc)->b >> 15) & 1)
867 #define GET_USEABLE(desc) (((desc)->b >> 20) & 1)
869 asmlinkage int sys_get_thread_area(struct user_desc __user *u_info)
871 struct user_desc info;
872 struct desc_struct *desc;
875 if (get_user(idx, &u_info->entry_number))
877 if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
880 memset(&info, 0, sizeof(info));
882 desc = current->thread.tls_array + idx - GDT_ENTRY_TLS_MIN;
884 info.entry_number = idx;
885 info.base_addr = GET_BASE(desc);
886 info.limit = GET_LIMIT(desc);
887 info.seg_32bit = GET_32BIT(desc);
888 info.contents = GET_CONTENTS(desc);
889 info.read_exec_only = !GET_WRITABLE(desc);
890 info.limit_in_pages = GET_LIMIT_PAGES(desc);
891 info.seg_not_present = !GET_PRESENT(desc);
892 info.useable = GET_USEABLE(desc);
894 if (copy_to_user(u_info, &info, sizeof(info)))
899 unsigned long arch_align_stack(unsigned long sp)
901 if (randomize_va_space)
902 sp -= get_random_int() % 8192;