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>
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/irq.h>
54 #include <linux/err.h>
56 asmlinkage void ret_from_fork(void) __asm__("ret_from_fork");
60 unsigned long boot_option_idle_override = 0;
61 EXPORT_SYMBOL(boot_option_idle_override);
64 * Return saved PC of a blocked thread.
66 unsigned long thread_saved_pc(struct task_struct *tsk)
68 return ((unsigned long *)tsk->thread.esp)[3];
72 * Powermanagement idle function, if any..
74 void (*pm_idle)(void);
75 static cpumask_t cpu_idle_map;
77 void disable_hlt(void)
82 EXPORT_SYMBOL(disable_hlt);
89 EXPORT_SYMBOL(enable_hlt);
92 * We use this if we don't have any better
95 void default_idle(void)
97 if (!hlt_counter && boot_cpu_data.hlt_works_ok) {
109 * On SMP it's slightly faster (but much more power-consuming!)
110 * to poll the ->work.need_resched flag instead of waiting for the
111 * cross-CPU IPI to arrive. Use this option with caution.
113 static void poll_idle (void)
120 * Deal with another CPU just having chosen a thread to
123 oldval = test_and_clear_thread_flag(TIF_NEED_RESCHED);
126 set_thread_flag(TIF_POLLING_NRFLAG);
132 : : "i"(_TIF_NEED_RESCHED), "m" (current_thread_info()->flags));
134 clear_thread_flag(TIF_POLLING_NRFLAG);
141 * The idle thread. There's no useful work to be
142 * done, so just try to conserve power and have a
143 * low exit latency (ie sit in a loop waiting for
144 * somebody to say that they'd like to reschedule)
148 int cpu = _smp_processor_id();
150 /* endless idle loop with no priority at all */
152 while (!need_resched()) {
155 if (cpu_isset(cpu, cpu_idle_map))
156 cpu_clear(cpu, cpu_idle_map);
163 irq_stat[cpu].idle_timestamp = jiffies;
170 void cpu_idle_wait(void)
175 for_each_online_cpu(cpu)
176 cpu_set(cpu, cpu_idle_map);
181 cpus_and(map, cpu_idle_map, cpu_online_map);
182 } while (!cpus_empty(map));
184 EXPORT_SYMBOL_GPL(cpu_idle_wait);
187 * This uses new MONITOR/MWAIT instructions on P4 processors with PNI,
188 * which can obviate IPI to trigger checking of need_resched.
189 * We execute MONITOR against need_resched and enter optimized wait state
190 * through MWAIT. Whenever someone changes need_resched, we would be woken
191 * up from MWAIT (without an IPI).
193 static void mwait_idle(void)
197 if (!need_resched()) {
198 set_thread_flag(TIF_POLLING_NRFLAG);
200 __monitor((void *)¤t_thread_info()->flags, 0, 0);
204 } while (!need_resched());
205 clear_thread_flag(TIF_POLLING_NRFLAG);
209 void __init select_idle_routine(const struct cpuinfo_x86 *c)
211 if (cpu_has(c, X86_FEATURE_MWAIT)) {
212 printk("monitor/mwait feature present.\n");
214 * Skip, if setup has overridden idle.
215 * One CPU supports mwait => All CPUs supports mwait
218 printk("using mwait in idle threads.\n");
219 pm_idle = mwait_idle;
224 static int __init idle_setup (char *str)
226 if (!strncmp(str, "poll", 4)) {
227 printk("using polling idle threads.\n");
229 #ifdef CONFIG_X86_SMP
230 if (smp_num_siblings > 1)
231 printk("WARNING: polling idle and HT enabled, performance may degrade.\n");
233 } else if (!strncmp(str, "halt", 4)) {
234 printk("using halt in idle threads.\n");
235 pm_idle = default_idle;
238 boot_option_idle_override = 1;
242 __setup("idle=", idle_setup);
244 void show_regs(struct pt_regs * regs)
246 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L;
249 printk("Pid: %d, comm: %20s\n", current->pid, current->comm);
250 printk("EIP: %04x:[<%08lx>] CPU: %d\n",0xffff & regs->xcs,regs->eip, smp_processor_id());
251 print_symbol("EIP is at %s\n", regs->eip);
254 printk(" ESP: %04x:%08lx",0xffff & regs->xss,regs->esp);
255 printk(" EFLAGS: %08lx %s (%s)\n",
256 regs->eflags, print_tainted(), system_utsname.release);
257 printk("EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
258 regs->eax,regs->ebx,regs->ecx,regs->edx);
259 printk("ESI: %08lx EDI: %08lx EBP: %08lx",
260 regs->esi, regs->edi, regs->ebp);
261 printk(" DS: %04x ES: %04x\n",
262 0xffff & regs->xds,0xffff & regs->xes);
264 __asm__("movl %%cr0, %0": "=r" (cr0));
265 __asm__("movl %%cr2, %0": "=r" (cr2));
266 __asm__("movl %%cr3, %0": "=r" (cr3));
267 /* This could fault if %cr4 does not exist */
268 __asm__("1: movl %%cr4, %0 \n"
270 ".section __ex_table,\"a\" \n"
273 : "=r" (cr4): "0" (0));
274 printk("CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n", cr0, cr2, cr3, cr4);
275 show_trace(NULL, ®s->esp);
279 * This gets run with %ebx containing the
280 * function to call, and %edx containing
283 extern void kernel_thread_helper(void);
284 __asm__(".section .text\n"
286 "kernel_thread_helper:\n\t"
295 * Create a kernel thread
297 int kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
301 memset(®s, 0, sizeof(regs));
303 regs.ebx = (unsigned long) fn;
304 regs.edx = (unsigned long) arg;
306 regs.xds = __USER_DS;
307 regs.xes = __USER_DS;
309 regs.eip = (unsigned long) kernel_thread_helper;
310 regs.xcs = __KERNEL_CS;
311 regs.eflags = X86_EFLAGS_IF | X86_EFLAGS_SF | X86_EFLAGS_PF | 0x2;
313 /* Ok, create the new process.. */
314 return do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0, ®s, 0, NULL, NULL);
318 * Free current thread data structures etc..
320 void exit_thread(void)
322 struct task_struct *tsk = current;
323 struct thread_struct *t = &tsk->thread;
325 /* The process may have allocated an io port bitmap... nuke it. */
326 if (unlikely(NULL != t->io_bitmap_ptr)) {
328 struct tss_struct *tss = &per_cpu(init_tss, cpu);
330 kfree(t->io_bitmap_ptr);
331 t->io_bitmap_ptr = NULL;
333 * Careful, clear this in the TSS too:
335 memset(tss->io_bitmap, 0xff, tss->io_bitmap_max);
336 t->io_bitmap_max = 0;
337 tss->io_bitmap_owner = NULL;
338 tss->io_bitmap_max = 0;
339 tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
344 void flush_thread(void)
346 struct task_struct *tsk = current;
348 memset(tsk->thread.debugreg, 0, sizeof(unsigned long)*8);
349 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
351 * Forget coprocessor state..
357 void release_thread(struct task_struct *dead_task)
360 // temporary debugging check
361 if (dead_task->mm->context.size) {
362 printk("WARNING: dead process %8s still has LDT? <%p/%d>\n",
364 dead_task->mm->context.ldt,
365 dead_task->mm->context.size);
370 release_vm86_irqs(dead_task);
374 * This gets called before we allocate a new thread and copy
375 * the current task into it.
377 void prepare_to_copy(struct task_struct *tsk)
382 int copy_thread(int nr, unsigned long clone_flags, unsigned long esp,
383 unsigned long unused,
384 struct task_struct * p, struct pt_regs * regs)
386 struct pt_regs * childregs;
387 struct task_struct *tsk;
390 childregs = ((struct pt_regs *) (THREAD_SIZE + (unsigned long) p->thread_info)) - 1;
393 childregs->esp = esp;
395 p->thread.esp = (unsigned long) childregs;
396 p->thread.esp0 = (unsigned long) (childregs+1);
398 p->thread.eip = (unsigned long) ret_from_fork;
400 savesegment(fs,p->thread.fs);
401 savesegment(gs,p->thread.gs);
404 if (unlikely(NULL != tsk->thread.io_bitmap_ptr)) {
405 p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL);
406 if (!p->thread.io_bitmap_ptr) {
407 p->thread.io_bitmap_max = 0;
410 memcpy(p->thread.io_bitmap_ptr, tsk->thread.io_bitmap_ptr,
415 * Set a new TLS for the child thread?
417 if (clone_flags & CLONE_SETTLS) {
418 struct desc_struct *desc;
419 struct user_desc info;
423 if (copy_from_user(&info, (void __user *)childregs->esi, sizeof(info)))
426 if (LDT_empty(&info))
429 idx = info.entry_number;
430 if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
433 desc = p->thread.tls_array + idx - GDT_ENTRY_TLS_MIN;
434 desc->a = LDT_entry_a(&info);
435 desc->b = LDT_entry_b(&info);
440 if (err && p->thread.io_bitmap_ptr) {
441 kfree(p->thread.io_bitmap_ptr);
442 p->thread.io_bitmap_max = 0;
448 * fill in the user structure for a core dump..
450 void dump_thread(struct pt_regs * regs, struct user * dump)
454 /* changed the size calculations - should hopefully work better. lbt */
455 dump->magic = CMAGIC;
456 dump->start_code = 0;
457 dump->start_stack = regs->esp & ~(PAGE_SIZE - 1);
458 dump->u_tsize = ((unsigned long) current->mm->end_code) >> PAGE_SHIFT;
459 dump->u_dsize = ((unsigned long) (current->mm->brk + (PAGE_SIZE-1))) >> PAGE_SHIFT;
460 dump->u_dsize -= dump->u_tsize;
462 for (i = 0; i < 8; i++)
463 dump->u_debugreg[i] = current->thread.debugreg[i];
465 if (dump->start_stack < TASK_SIZE)
466 dump->u_ssize = ((unsigned long) (TASK_SIZE - dump->start_stack)) >> PAGE_SHIFT;
468 dump->regs.ebx = regs->ebx;
469 dump->regs.ecx = regs->ecx;
470 dump->regs.edx = regs->edx;
471 dump->regs.esi = regs->esi;
472 dump->regs.edi = regs->edi;
473 dump->regs.ebp = regs->ebp;
474 dump->regs.eax = regs->eax;
475 dump->regs.ds = regs->xds;
476 dump->regs.es = regs->xes;
477 savesegment(fs,dump->regs.fs);
478 savesegment(gs,dump->regs.gs);
479 dump->regs.orig_eax = regs->orig_eax;
480 dump->regs.eip = regs->eip;
481 dump->regs.cs = regs->xcs;
482 dump->regs.eflags = regs->eflags;
483 dump->regs.esp = regs->esp;
484 dump->regs.ss = regs->xss;
486 dump->u_fpvalid = dump_fpu (regs, &dump->i387);
490 * Capture the user space registers if the task is not running (in user space)
492 int dump_task_regs(struct task_struct *tsk, elf_gregset_t *regs)
494 struct pt_regs ptregs;
496 ptregs = *(struct pt_regs *)
497 ((unsigned long)tsk->thread_info+THREAD_SIZE - sizeof(ptregs));
498 ptregs.xcs &= 0xffff;
499 ptregs.xds &= 0xffff;
500 ptregs.xes &= 0xffff;
501 ptregs.xss &= 0xffff;
503 elf_core_copy_regs(regs, &ptregs);
509 handle_io_bitmap(struct thread_struct *next, struct tss_struct *tss)
511 if (!next->io_bitmap_ptr) {
513 * Disable the bitmap via an invalid offset. We still cache
514 * the previous bitmap owner and the IO bitmap contents:
516 tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
519 if (likely(next == tss->io_bitmap_owner)) {
521 * Previous owner of the bitmap (hence the bitmap content)
522 * matches the next task, we dont have to do anything but
523 * to set a valid offset in the TSS:
525 tss->io_bitmap_base = IO_BITMAP_OFFSET;
529 * Lazy TSS's I/O bitmap copy. We set an invalid offset here
530 * and we let the task to get a GPF in case an I/O instruction
531 * is performed. The handler of the GPF will verify that the
532 * faulting task has a valid I/O bitmap and, it true, does the
533 * real copy and restart the instruction. This will save us
534 * redundant copies when the currently switched task does not
535 * perform any I/O during its timeslice.
537 tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET_LAZY;
540 * This special macro can be used to load a debugging register
542 #define loaddebug(thread,register) \
543 __asm__("movl %0,%%db" #register \
545 :"r" (thread->debugreg[register]))
548 * switch_to(x,yn) should switch tasks from x to y.
550 * We fsave/fwait so that an exception goes off at the right time
551 * (as a call from the fsave or fwait in effect) rather than to
552 * the wrong process. Lazy FP saving no longer makes any sense
553 * with modern CPU's, and this simplifies a lot of things (SMP
554 * and UP become the same).
556 * NOTE! We used to use the x86 hardware context switching. The
557 * reason for not using it any more becomes apparent when you
558 * try to recover gracefully from saved state that is no longer
559 * valid (stale segment register values in particular). With the
560 * hardware task-switch, there is no way to fix up bad state in
561 * a reasonable manner.
563 * The fact that Intel documents the hardware task-switching to
564 * be slow is a fairly red herring - this code is not noticeably
565 * faster. However, there _is_ some room for improvement here,
566 * so the performance issues may eventually be a valid point.
567 * More important, however, is the fact that this allows us much
570 * The return value (in %eax) will be the "prev" task after
571 * the task-switch, and shows up in ret_from_fork in entry.S,
574 struct task_struct fastcall * __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
576 struct thread_struct *prev = &prev_p->thread,
577 *next = &next_p->thread;
578 int cpu = smp_processor_id();
579 struct tss_struct *tss = &per_cpu(init_tss, cpu);
581 /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
583 __unlazy_fpu(prev_p);
586 * Reload esp0, LDT and the page table pointer:
588 load_esp0(tss, next);
591 * Load the per-thread Thread-Local Storage descriptor.
596 * Save away %fs and %gs. No need to save %es and %ds, as
597 * those are always kernel segments while inside the kernel.
599 asm volatile("movl %%fs,%0":"=m" (*(int *)&prev->fs));
600 asm volatile("movl %%gs,%0":"=m" (*(int *)&prev->gs));
603 * Restore %fs and %gs if needed.
605 if (unlikely(prev->fs | prev->gs | next->fs | next->gs)) {
606 loadsegment(fs, next->fs);
607 loadsegment(gs, next->gs);
611 * Now maybe reload the debug registers
613 if (unlikely(next->debugreg[7])) {
623 if (unlikely(prev->io_bitmap_ptr || next->io_bitmap_ptr))
624 handle_io_bitmap(next, tss);
629 asmlinkage int sys_fork(struct pt_regs regs)
631 return do_fork(SIGCHLD, regs.esp, ®s, 0, NULL, NULL);
634 asmlinkage int sys_clone(struct pt_regs regs)
636 unsigned long clone_flags;
638 int __user *parent_tidptr, *child_tidptr;
640 clone_flags = regs.ebx;
642 parent_tidptr = (int __user *)regs.edx;
643 child_tidptr = (int __user *)regs.edi;
646 return do_fork(clone_flags, newsp, ®s, 0, parent_tidptr, child_tidptr);
650 * This is trivial, and on the face of it looks like it
651 * could equally well be done in user mode.
653 * Not so, for quite unobvious reasons - register pressure.
654 * In user mode vfork() cannot have a stack frame, and if
655 * done by calling the "clone()" system call directly, you
656 * do not have enough call-clobbered registers to hold all
657 * the information you need.
659 asmlinkage int sys_vfork(struct pt_regs regs)
661 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs.esp, ®s, 0, NULL, NULL);
665 * sys_execve() executes a new program.
667 asmlinkage int sys_execve(struct pt_regs regs)
672 filename = getname((char __user *) regs.ebx);
673 error = PTR_ERR(filename);
674 if (IS_ERR(filename))
676 error = do_execve(filename,
677 (char __user * __user *) regs.ecx,
678 (char __user * __user *) regs.edx,
682 current->ptrace &= ~PT_DTRACE;
683 task_unlock(current);
684 /* Make sure we don't return using sysenter.. */
685 set_thread_flag(TIF_IRET);
692 #define top_esp (THREAD_SIZE - sizeof(unsigned long))
693 #define top_ebp (THREAD_SIZE - 2*sizeof(unsigned long))
695 unsigned long get_wchan(struct task_struct *p)
697 unsigned long ebp, esp, eip;
698 unsigned long stack_page;
700 if (!p || p == current || p->state == TASK_RUNNING)
702 stack_page = (unsigned long)p->thread_info;
704 if (!stack_page || esp < stack_page || esp > top_esp+stack_page)
706 /* include/asm-i386/system.h:switch_to() pushes ebp last. */
707 ebp = *(unsigned long *) esp;
709 if (ebp < stack_page || ebp > top_ebp+stack_page)
711 eip = *(unsigned long *) (ebp+4);
712 if (!in_sched_functions(eip))
714 ebp = *(unsigned long *) ebp;
715 } while (count++ < 16);
720 * sys_alloc_thread_area: get a yet unused TLS descriptor index.
722 static int get_free_idx(void)
724 struct thread_struct *t = ¤t->thread;
727 for (idx = 0; idx < GDT_ENTRY_TLS_ENTRIES; idx++)
728 if (desc_empty(t->tls_array + idx))
729 return idx + GDT_ENTRY_TLS_MIN;
734 * Set a given TLS descriptor:
736 asmlinkage int sys_set_thread_area(struct user_desc __user *u_info)
738 struct thread_struct *t = ¤t->thread;
739 struct user_desc info;
740 struct desc_struct *desc;
743 if (copy_from_user(&info, u_info, sizeof(info)))
745 idx = info.entry_number;
748 * index -1 means the kernel should try to find and
749 * allocate an empty descriptor:
752 idx = get_free_idx();
755 if (put_user(idx, &u_info->entry_number))
759 if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
762 desc = t->tls_array + idx - GDT_ENTRY_TLS_MIN;
765 * We must not get preempted while modifying the TLS.
769 if (LDT_empty(&info)) {
773 desc->a = LDT_entry_a(&info);
774 desc->b = LDT_entry_b(&info);
784 * Get the current Thread-Local Storage area:
787 #define GET_BASE(desc) ( \
788 (((desc)->a >> 16) & 0x0000ffff) | \
789 (((desc)->b << 16) & 0x00ff0000) | \
790 ( (desc)->b & 0xff000000) )
792 #define GET_LIMIT(desc) ( \
793 ((desc)->a & 0x0ffff) | \
794 ((desc)->b & 0xf0000) )
796 #define GET_32BIT(desc) (((desc)->b >> 22) & 1)
797 #define GET_CONTENTS(desc) (((desc)->b >> 10) & 3)
798 #define GET_WRITABLE(desc) (((desc)->b >> 9) & 1)
799 #define GET_LIMIT_PAGES(desc) (((desc)->b >> 23) & 1)
800 #define GET_PRESENT(desc) (((desc)->b >> 15) & 1)
801 #define GET_USEABLE(desc) (((desc)->b >> 20) & 1)
803 asmlinkage int sys_get_thread_area(struct user_desc __user *u_info)
805 struct user_desc info;
806 struct desc_struct *desc;
809 if (get_user(idx, &u_info->entry_number))
811 if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
814 desc = current->thread.tls_array + idx - GDT_ENTRY_TLS_MIN;
816 info.entry_number = idx;
817 info.base_addr = GET_BASE(desc);
818 info.limit = GET_LIMIT(desc);
819 info.seg_32bit = GET_32BIT(desc);
820 info.contents = GET_CONTENTS(desc);
821 info.read_exec_only = !GET_WRITABLE(desc);
822 info.limit_in_pages = GET_LIMIT_PAGES(desc);
823 info.seg_not_present = !GET_PRESENT(desc);
824 info.useable = GET_USEABLE(desc);
826 if (copy_to_user(u_info, &info, sizeof(info)))