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/errno.h>
17 #include <linux/sched.h>
19 #include <linux/kernel.h>
21 #include <linux/elfcore.h>
22 #include <linux/smp.h>
23 #include <linux/smp_lock.h>
24 #include <linux/stddef.h>
25 #include <linux/slab.h>
26 #include <linux/vmalloc.h>
27 #include <linux/user.h>
28 #include <linux/a.out.h>
29 #include <linux/interrupt.h>
30 #include <linux/config.h>
31 #include <linux/utsname.h>
32 #include <linux/delay.h>
33 #include <linux/reboot.h>
34 #include <linux/init.h>
35 #include <linux/mc146818rtc.h>
36 #include <linux/module.h>
37 #include <linux/kallsyms.h>
38 #include <linux/ptrace.h>
39 #include <linux/random.h>
41 #include <asm/uaccess.h>
42 #include <asm/pgtable.h>
43 #include <asm/system.h>
46 #include <asm/processor.h>
50 #ifdef CONFIG_MATH_EMULATION
51 #include <asm/math_emu.h>
54 #include <linux/irq.h>
55 #include <linux/err.h>
57 asmlinkage void ret_from_fork(void) __asm__("ret_from_fork");
59 static int hlt_counter;
61 unsigned long boot_option_idle_override = 0;
62 EXPORT_SYMBOL(boot_option_idle_override);
65 * Return saved PC of a blocked thread.
67 unsigned long thread_saved_pc(struct task_struct *tsk)
69 return ((unsigned long *)tsk->thread.esp)[3];
73 * Powermanagement idle function, if any..
75 void (*pm_idle)(void);
76 static DEFINE_PER_CPU(unsigned int, cpu_idle_state);
78 void disable_hlt(void)
83 EXPORT_SYMBOL(disable_hlt);
90 EXPORT_SYMBOL(enable_hlt);
93 * We use this if we don't have any better
96 void default_idle(void)
98 if (!hlt_counter && boot_cpu_data.hlt_works_ok) {
110 * On SMP it's slightly faster (but much more power-consuming!)
111 * to poll the ->work.need_resched flag instead of waiting for the
112 * cross-CPU IPI to arrive. Use this option with caution.
114 static void poll_idle (void)
121 * Deal with another CPU just having chosen a thread to
124 oldval = test_and_clear_thread_flag(TIF_NEED_RESCHED);
127 set_thread_flag(TIF_POLLING_NRFLAG);
133 : : "i"(_TIF_NEED_RESCHED), "m" (current_thread_info()->flags));
135 clear_thread_flag(TIF_POLLING_NRFLAG);
142 * The idle thread. There's no useful work to be
143 * done, so just try to conserve power and have a
144 * low exit latency (ie sit in a loop waiting for
145 * somebody to say that they'd like to reschedule)
149 /* endless idle loop with no priority at all */
151 while (!need_resched()) {
154 if (__get_cpu_var(cpu_idle_state))
155 __get_cpu_var(cpu_idle_state) = 0;
163 __get_cpu_var(irq_stat).idle_timestamp = jiffies;
170 void cpu_idle_wait(void)
172 unsigned int cpu, this_cpu = get_cpu();
175 set_cpus_allowed(current, cpumask_of_cpu(this_cpu));
179 for_each_online_cpu(cpu) {
180 per_cpu(cpu_idle_state, cpu) = 1;
184 __get_cpu_var(cpu_idle_state) = 0;
189 for_each_online_cpu(cpu) {
190 if (cpu_isset(cpu, map) && !per_cpu(cpu_idle_state, cpu))
193 cpus_and(map, map, cpu_online_map);
194 } while (!cpus_empty(map));
196 EXPORT_SYMBOL_GPL(cpu_idle_wait);
199 * This uses new MONITOR/MWAIT instructions on P4 processors with PNI,
200 * which can obviate IPI to trigger checking of need_resched.
201 * We execute MONITOR against need_resched and enter optimized wait state
202 * through MWAIT. Whenever someone changes need_resched, we would be woken
203 * up from MWAIT (without an IPI).
205 static void mwait_idle(void)
209 if (!need_resched()) {
210 set_thread_flag(TIF_POLLING_NRFLAG);
212 __monitor((void *)¤t_thread_info()->flags, 0, 0);
216 } while (!need_resched());
217 clear_thread_flag(TIF_POLLING_NRFLAG);
221 void __init select_idle_routine(const struct cpuinfo_x86 *c)
223 if (cpu_has(c, X86_FEATURE_MWAIT)) {
224 printk("monitor/mwait feature present.\n");
226 * Skip, if setup has overridden idle.
227 * One CPU supports mwait => All CPUs supports mwait
230 printk("using mwait in idle threads.\n");
231 pm_idle = mwait_idle;
236 static int __init idle_setup (char *str)
238 if (!strncmp(str, "poll", 4)) {
239 printk("using polling idle threads.\n");
241 #ifdef CONFIG_X86_SMP
242 if (smp_num_siblings > 1)
243 printk("WARNING: polling idle and HT enabled, performance may degrade.\n");
245 } else if (!strncmp(str, "halt", 4)) {
246 printk("using halt in idle threads.\n");
247 pm_idle = default_idle;
250 boot_option_idle_override = 1;
254 __setup("idle=", idle_setup);
256 void show_regs(struct pt_regs * regs)
258 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L;
261 printk("Pid: %d, comm: %20s\n", current->pid, current->comm);
262 printk("EIP: %04x:[<%08lx>] CPU: %d\n",0xffff & regs->xcs,regs->eip, smp_processor_id());
263 print_symbol("EIP is at %s\n", regs->eip);
266 printk(" ESP: %04x:%08lx",0xffff & regs->xss,regs->esp);
267 printk(" EFLAGS: %08lx %s (%s)\n",
268 regs->eflags, print_tainted(), system_utsname.release);
269 printk("EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
270 regs->eax,regs->ebx,regs->ecx,regs->edx);
271 printk("ESI: %08lx EDI: %08lx EBP: %08lx",
272 regs->esi, regs->edi, regs->ebp);
273 printk(" DS: %04x ES: %04x\n",
274 0xffff & regs->xds,0xffff & regs->xes);
276 __asm__("movl %%cr0, %0": "=r" (cr0));
277 __asm__("movl %%cr2, %0": "=r" (cr2));
278 __asm__("movl %%cr3, %0": "=r" (cr3));
279 /* This could fault if %cr4 does not exist */
280 __asm__("1: movl %%cr4, %0 \n"
282 ".section __ex_table,\"a\" \n"
285 : "=r" (cr4): "0" (0));
286 printk("CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n", cr0, cr2, cr3, cr4);
287 show_trace(NULL, ®s->esp);
291 * This gets run with %ebx containing the
292 * function to call, and %edx containing
295 extern void kernel_thread_helper(void);
296 __asm__(".section .text\n"
298 "kernel_thread_helper:\n\t"
307 * Create a kernel thread
309 int kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
313 memset(®s, 0, sizeof(regs));
315 regs.ebx = (unsigned long) fn;
316 regs.edx = (unsigned long) arg;
318 regs.xds = __USER_DS;
319 regs.xes = __USER_DS;
321 regs.eip = (unsigned long) kernel_thread_helper;
322 regs.xcs = __KERNEL_CS;
323 regs.eflags = X86_EFLAGS_IF | X86_EFLAGS_SF | X86_EFLAGS_PF | 0x2;
325 /* Ok, create the new process.. */
326 return do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0, ®s, 0, NULL, NULL);
330 * Free current thread data structures etc..
332 void exit_thread(void)
334 struct task_struct *tsk = current;
335 struct thread_struct *t = &tsk->thread;
337 /* The process may have allocated an io port bitmap... nuke it. */
338 if (unlikely(NULL != t->io_bitmap_ptr)) {
340 struct tss_struct *tss = &per_cpu(init_tss, cpu);
342 kfree(t->io_bitmap_ptr);
343 t->io_bitmap_ptr = NULL;
345 * Careful, clear this in the TSS too:
347 memset(tss->io_bitmap, 0xff, tss->io_bitmap_max);
348 t->io_bitmap_max = 0;
349 tss->io_bitmap_owner = NULL;
350 tss->io_bitmap_max = 0;
351 tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
356 void flush_thread(void)
358 struct task_struct *tsk = current;
360 memset(tsk->thread.debugreg, 0, sizeof(unsigned long)*8);
361 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
363 * Forget coprocessor state..
369 void release_thread(struct task_struct *dead_task)
372 // temporary debugging check
373 if (dead_task->mm->context.size) {
374 printk("WARNING: dead process %8s still has LDT? <%p/%d>\n",
376 dead_task->mm->context.ldt,
377 dead_task->mm->context.size);
382 release_vm86_irqs(dead_task);
386 * This gets called before we allocate a new thread and copy
387 * the current task into it.
389 void prepare_to_copy(struct task_struct *tsk)
394 int copy_thread(int nr, unsigned long clone_flags, unsigned long esp,
395 unsigned long unused,
396 struct task_struct * p, struct pt_regs * regs)
398 struct pt_regs * childregs;
399 struct task_struct *tsk;
402 childregs = ((struct pt_regs *) (THREAD_SIZE + (unsigned long) p->thread_info)) - 1;
404 * The below -8 is to reserve 8 bytes on top of the ring0 stack.
405 * This is necessary to guarantee that the entire "struct pt_regs"
406 * is accessable even if the CPU haven't stored the SS/ESP registers
407 * on the stack (interrupt gate does not save these registers
408 * when switching to the same priv ring).
409 * Therefore beware: accessing the xss/esp fields of the
410 * "struct pt_regs" is possible, but they may contain the
411 * completely wrong values.
413 childregs = (struct pt_regs *) ((unsigned long) childregs - 8);
416 childregs->esp = esp;
418 p->thread.esp = (unsigned long) childregs;
419 p->thread.esp0 = (unsigned long) (childregs+1);
421 p->thread.eip = (unsigned long) ret_from_fork;
423 savesegment(fs,p->thread.fs);
424 savesegment(gs,p->thread.gs);
427 if (unlikely(NULL != tsk->thread.io_bitmap_ptr)) {
428 p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL);
429 if (!p->thread.io_bitmap_ptr) {
430 p->thread.io_bitmap_max = 0;
433 memcpy(p->thread.io_bitmap_ptr, tsk->thread.io_bitmap_ptr,
438 * Set a new TLS for the child thread?
440 if (clone_flags & CLONE_SETTLS) {
441 struct desc_struct *desc;
442 struct user_desc info;
446 if (copy_from_user(&info, (void __user *)childregs->esi, sizeof(info)))
449 if (LDT_empty(&info))
452 idx = info.entry_number;
453 if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
456 desc = p->thread.tls_array + idx - GDT_ENTRY_TLS_MIN;
457 desc->a = LDT_entry_a(&info);
458 desc->b = LDT_entry_b(&info);
463 if (err && p->thread.io_bitmap_ptr) {
464 kfree(p->thread.io_bitmap_ptr);
465 p->thread.io_bitmap_max = 0;
471 * fill in the user structure for a core dump..
473 void dump_thread(struct pt_regs * regs, struct user * dump)
477 /* changed the size calculations - should hopefully work better. lbt */
478 dump->magic = CMAGIC;
479 dump->start_code = 0;
480 dump->start_stack = regs->esp & ~(PAGE_SIZE - 1);
481 dump->u_tsize = ((unsigned long) current->mm->end_code) >> PAGE_SHIFT;
482 dump->u_dsize = ((unsigned long) (current->mm->brk + (PAGE_SIZE-1))) >> PAGE_SHIFT;
483 dump->u_dsize -= dump->u_tsize;
485 for (i = 0; i < 8; i++)
486 dump->u_debugreg[i] = current->thread.debugreg[i];
488 if (dump->start_stack < TASK_SIZE)
489 dump->u_ssize = ((unsigned long) (TASK_SIZE - dump->start_stack)) >> PAGE_SHIFT;
491 dump->regs.ebx = regs->ebx;
492 dump->regs.ecx = regs->ecx;
493 dump->regs.edx = regs->edx;
494 dump->regs.esi = regs->esi;
495 dump->regs.edi = regs->edi;
496 dump->regs.ebp = regs->ebp;
497 dump->regs.eax = regs->eax;
498 dump->regs.ds = regs->xds;
499 dump->regs.es = regs->xes;
500 savesegment(fs,dump->regs.fs);
501 savesegment(gs,dump->regs.gs);
502 dump->regs.orig_eax = regs->orig_eax;
503 dump->regs.eip = regs->eip;
504 dump->regs.cs = regs->xcs;
505 dump->regs.eflags = regs->eflags;
506 dump->regs.esp = regs->esp;
507 dump->regs.ss = regs->xss;
509 dump->u_fpvalid = dump_fpu (regs, &dump->i387);
513 * Capture the user space registers if the task is not running (in user space)
515 int dump_task_regs(struct task_struct *tsk, elf_gregset_t *regs)
517 struct pt_regs ptregs;
519 ptregs = *(struct pt_regs *)
520 ((unsigned long)tsk->thread_info+THREAD_SIZE - sizeof(ptregs));
521 ptregs.xcs &= 0xffff;
522 ptregs.xds &= 0xffff;
523 ptregs.xes &= 0xffff;
524 ptregs.xss &= 0xffff;
526 elf_core_copy_regs(regs, &ptregs);
532 handle_io_bitmap(struct thread_struct *next, struct tss_struct *tss)
534 if (!next->io_bitmap_ptr) {
536 * Disable the bitmap via an invalid offset. We still cache
537 * the previous bitmap owner and the IO bitmap contents:
539 tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
542 if (likely(next == tss->io_bitmap_owner)) {
544 * Previous owner of the bitmap (hence the bitmap content)
545 * matches the next task, we dont have to do anything but
546 * to set a valid offset in the TSS:
548 tss->io_bitmap_base = IO_BITMAP_OFFSET;
552 * Lazy TSS's I/O bitmap copy. We set an invalid offset here
553 * and we let the task to get a GPF in case an I/O instruction
554 * is performed. The handler of the GPF will verify that the
555 * faulting task has a valid I/O bitmap and, it true, does the
556 * real copy and restart the instruction. This will save us
557 * redundant copies when the currently switched task does not
558 * perform any I/O during its timeslice.
560 tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET_LAZY;
564 * switch_to(x,yn) should switch tasks from x to y.
566 * We fsave/fwait so that an exception goes off at the right time
567 * (as a call from the fsave or fwait in effect) rather than to
568 * the wrong process. Lazy FP saving no longer makes any sense
569 * with modern CPU's, and this simplifies a lot of things (SMP
570 * and UP become the same).
572 * NOTE! We used to use the x86 hardware context switching. The
573 * reason for not using it any more becomes apparent when you
574 * try to recover gracefully from saved state that is no longer
575 * valid (stale segment register values in particular). With the
576 * hardware task-switch, there is no way to fix up bad state in
577 * a reasonable manner.
579 * The fact that Intel documents the hardware task-switching to
580 * be slow is a fairly red herring - this code is not noticeably
581 * faster. However, there _is_ some room for improvement here,
582 * so the performance issues may eventually be a valid point.
583 * More important, however, is the fact that this allows us much
586 * The return value (in %eax) will be the "prev" task after
587 * the task-switch, and shows up in ret_from_fork in entry.S,
590 struct task_struct fastcall * __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
592 struct thread_struct *prev = &prev_p->thread,
593 *next = &next_p->thread;
594 int cpu = smp_processor_id();
595 struct tss_struct *tss = &per_cpu(init_tss, cpu);
597 /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
599 __unlazy_fpu(prev_p);
602 * Reload esp0, LDT and the page table pointer:
604 load_esp0(tss, next);
607 * Load the per-thread Thread-Local Storage descriptor.
612 * Save away %fs and %gs. No need to save %es and %ds, as
613 * those are always kernel segments while inside the kernel.
615 asm volatile("mov %%fs,%0":"=m" (prev->fs));
616 asm volatile("mov %%gs,%0":"=m" (prev->gs));
619 * Restore %fs and %gs if needed.
621 if (unlikely(prev->fs | prev->gs | next->fs | next->gs)) {
622 loadsegment(fs, next->fs);
623 loadsegment(gs, next->gs);
627 * Now maybe reload the debug registers
629 if (unlikely(next->debugreg[7])) {
639 if (unlikely(prev->io_bitmap_ptr || next->io_bitmap_ptr))
640 handle_io_bitmap(next, tss);
645 asmlinkage int sys_fork(struct pt_regs regs)
647 return do_fork(SIGCHLD, regs.esp, ®s, 0, NULL, NULL);
650 asmlinkage int sys_clone(struct pt_regs regs)
652 unsigned long clone_flags;
654 int __user *parent_tidptr, *child_tidptr;
656 clone_flags = regs.ebx;
658 parent_tidptr = (int __user *)regs.edx;
659 child_tidptr = (int __user *)regs.edi;
662 return do_fork(clone_flags, newsp, ®s, 0, parent_tidptr, child_tidptr);
666 * This is trivial, and on the face of it looks like it
667 * could equally well be done in user mode.
669 * Not so, for quite unobvious reasons - register pressure.
670 * In user mode vfork() cannot have a stack frame, and if
671 * done by calling the "clone()" system call directly, you
672 * do not have enough call-clobbered registers to hold all
673 * the information you need.
675 asmlinkage int sys_vfork(struct pt_regs regs)
677 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs.esp, ®s, 0, NULL, NULL);
681 * sys_execve() executes a new program.
683 asmlinkage int sys_execve(struct pt_regs regs)
688 filename = getname((char __user *) regs.ebx);
689 error = PTR_ERR(filename);
690 if (IS_ERR(filename))
692 error = do_execve(filename,
693 (char __user * __user *) regs.ecx,
694 (char __user * __user *) regs.edx,
698 current->ptrace &= ~PT_DTRACE;
699 task_unlock(current);
700 /* Make sure we don't return using sysenter.. */
701 set_thread_flag(TIF_IRET);
708 #define top_esp (THREAD_SIZE - sizeof(unsigned long))
709 #define top_ebp (THREAD_SIZE - 2*sizeof(unsigned long))
711 unsigned long get_wchan(struct task_struct *p)
713 unsigned long ebp, esp, eip;
714 unsigned long stack_page;
716 if (!p || p == current || p->state == TASK_RUNNING)
718 stack_page = (unsigned long)p->thread_info;
720 if (!stack_page || esp < stack_page || esp > top_esp+stack_page)
722 /* include/asm-i386/system.h:switch_to() pushes ebp last. */
723 ebp = *(unsigned long *) esp;
725 if (ebp < stack_page || ebp > top_ebp+stack_page)
727 eip = *(unsigned long *) (ebp+4);
728 if (!in_sched_functions(eip))
730 ebp = *(unsigned long *) ebp;
731 } while (count++ < 16);
736 * sys_alloc_thread_area: get a yet unused TLS descriptor index.
738 static int get_free_idx(void)
740 struct thread_struct *t = ¤t->thread;
743 for (idx = 0; idx < GDT_ENTRY_TLS_ENTRIES; idx++)
744 if (desc_empty(t->tls_array + idx))
745 return idx + GDT_ENTRY_TLS_MIN;
750 * Set a given TLS descriptor:
752 asmlinkage int sys_set_thread_area(struct user_desc __user *u_info)
754 struct thread_struct *t = ¤t->thread;
755 struct user_desc info;
756 struct desc_struct *desc;
759 if (copy_from_user(&info, u_info, sizeof(info)))
761 idx = info.entry_number;
764 * index -1 means the kernel should try to find and
765 * allocate an empty descriptor:
768 idx = get_free_idx();
771 if (put_user(idx, &u_info->entry_number))
775 if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
778 desc = t->tls_array + idx - GDT_ENTRY_TLS_MIN;
781 * We must not get preempted while modifying the TLS.
785 if (LDT_empty(&info)) {
789 desc->a = LDT_entry_a(&info);
790 desc->b = LDT_entry_b(&info);
800 * Get the current Thread-Local Storage area:
803 #define GET_BASE(desc) ( \
804 (((desc)->a >> 16) & 0x0000ffff) | \
805 (((desc)->b << 16) & 0x00ff0000) | \
806 ( (desc)->b & 0xff000000) )
808 #define GET_LIMIT(desc) ( \
809 ((desc)->a & 0x0ffff) | \
810 ((desc)->b & 0xf0000) )
812 #define GET_32BIT(desc) (((desc)->b >> 22) & 1)
813 #define GET_CONTENTS(desc) (((desc)->b >> 10) & 3)
814 #define GET_WRITABLE(desc) (((desc)->b >> 9) & 1)
815 #define GET_LIMIT_PAGES(desc) (((desc)->b >> 23) & 1)
816 #define GET_PRESENT(desc) (((desc)->b >> 15) & 1)
817 #define GET_USEABLE(desc) (((desc)->b >> 20) & 1)
819 asmlinkage int sys_get_thread_area(struct user_desc __user *u_info)
821 struct user_desc info;
822 struct desc_struct *desc;
825 if (get_user(idx, &u_info->entry_number))
827 if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
830 desc = current->thread.tls_array + idx - GDT_ENTRY_TLS_MIN;
832 info.entry_number = idx;
833 info.base_addr = GET_BASE(desc);
834 info.limit = GET_LIMIT(desc);
835 info.seg_32bit = GET_32BIT(desc);
836 info.contents = GET_CONTENTS(desc);
837 info.read_exec_only = !GET_WRITABLE(desc);
838 info.limit_in_pages = GET_LIMIT_PAGES(desc);
839 info.seg_not_present = !GET_PRESENT(desc);
840 info.useable = GET_USEABLE(desc);
842 if (copy_to_user(u_info, &info, sizeof(info)))
847 unsigned long arch_align_stack(unsigned long sp)
849 if (randomize_va_space)
850 sp -= get_random_int() % 8192;