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/mman.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/irq.h>
56 #include <linux/err.h>
58 asmlinkage void ret_from_fork(void) __asm__("ret_from_fork");
62 unsigned long boot_option_idle_override = 0;
63 EXPORT_SYMBOL(boot_option_idle_override);
66 * Return saved PC of a blocked thread.
68 unsigned long thread_saved_pc(struct task_struct *tsk)
70 return ((unsigned long *)tsk->thread.esp)[3];
74 * Powermanagement idle function, if any..
76 void (*pm_idle)(void);
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 && current_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 * Mark this as an RCU critical section so that
155 * synchronize_kernel() in the unload path waits
156 * for our completion.
164 irq_stat[smp_processor_id()].idle_timestamp = jiffies;
173 * This uses new MONITOR/MWAIT instructions on P4 processors with PNI,
174 * which can obviate IPI to trigger checking of need_resched.
175 * We execute MONITOR against need_resched and enter optimized wait state
176 * through MWAIT. Whenever someone changes need_resched, we would be woken
177 * up from MWAIT (without an IPI).
179 static void mwait_idle(void)
183 if (!need_resched()) {
184 set_thread_flag(TIF_POLLING_NRFLAG);
186 __monitor((void *)¤t_thread_info()->flags, 0, 0);
190 } while (!need_resched());
191 clear_thread_flag(TIF_POLLING_NRFLAG);
195 void __init select_idle_routine(const struct cpuinfo_x86 *c)
197 if (cpu_has(c, X86_FEATURE_MWAIT)) {
198 printk("monitor/mwait feature present.\n");
200 * Skip, if setup has overridden idle.
201 * One CPU supports mwait => All CPUs supports mwait
204 printk("using mwait in idle threads.\n");
205 pm_idle = mwait_idle;
210 static int __init idle_setup (char *str)
212 if (!strncmp(str, "poll", 4)) {
213 printk("using polling idle threads.\n");
215 #ifdef CONFIG_X86_SMP
216 if (smp_num_siblings > 1)
217 printk("WARNING: polling idle and HT enabled, performance may degrade.\n");
219 } else if (!strncmp(str, "halt", 4)) {
220 printk("using halt in idle threads.\n");
221 pm_idle = default_idle;
224 boot_option_idle_override = 1;
228 __setup("idle=", idle_setup);
230 void show_regs(struct pt_regs * regs)
232 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L;
235 printk("Pid: %d, comm: %20s\n", current->pid, current->comm);
236 printk("EIP: %04x:[<%08lx>] CPU: %d\n",0xffff & regs->xcs,regs->eip, smp_processor_id());
237 print_symbol("EIP is at %s\n", regs->eip);
240 printk(" ESP: %04x:%08lx",0xffff & regs->xss,regs->esp);
241 printk(" EFLAGS: %08lx %s (%s)\n",
242 regs->eflags, print_tainted(), system_utsname.release);
243 printk("EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
244 regs->eax,regs->ebx,regs->ecx,regs->edx);
245 printk("ESI: %08lx EDI: %08lx EBP: %08lx",
246 regs->esi, regs->edi, regs->ebp);
247 printk(" DS: %04x ES: %04x\n",
248 0xffff & regs->xds,0xffff & regs->xes);
250 __asm__("movl %%cr0, %0": "=r" (cr0));
251 __asm__("movl %%cr2, %0": "=r" (cr2));
252 __asm__("movl %%cr3, %0": "=r" (cr3));
253 /* This could fault if %cr4 does not exist */
254 __asm__("1: movl %%cr4, %0 \n"
256 ".section __ex_table,\"a\" \n"
259 : "=r" (cr4): "0" (0));
260 printk("CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n", cr0, cr2, cr3, cr4);
261 show_trace(NULL, ®s->esp);
264 EXPORT_SYMBOL_GPL(show_regs);
267 * This gets run with %ebx containing the
268 * function to call, and %edx containing
271 extern void kernel_thread_helper(void);
272 __asm__(".section .text\n"
274 "kernel_thread_helper:\n\t"
283 * Create a kernel thread
285 int kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
289 memset(®s, 0, sizeof(regs));
291 regs.ebx = (unsigned long) fn;
292 regs.edx = (unsigned long) arg;
294 regs.xds = __USER_DS;
295 regs.xes = __USER_DS;
297 regs.eip = (unsigned long) kernel_thread_helper;
298 regs.xcs = __KERNEL_CS;
299 regs.eflags = X86_EFLAGS_IF | X86_EFLAGS_SF | X86_EFLAGS_PF | 0x2;
301 /* Ok, create the new process.. */
302 return do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0, ®s, 0, NULL, NULL);
306 * Free current thread data structures etc..
308 void exit_thread(void)
310 struct task_struct *tsk = current;
311 struct thread_struct *t = &tsk->thread;
313 /* The process may have allocated an io port bitmap... nuke it. */
314 if (unlikely(NULL != t->io_bitmap_ptr)) {
316 struct tss_struct *tss = &per_cpu(init_tss, cpu);
318 kfree(t->io_bitmap_ptr);
319 t->io_bitmap_ptr = NULL;
321 * Careful, clear this in the TSS too:
323 memset(tss->io_bitmap, 0xff, tss->io_bitmap_max);
324 t->io_bitmap_max = 0;
325 tss->io_bitmap_owner = NULL;
326 tss->io_bitmap_max = 0;
327 tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
332 void flush_thread(void)
334 struct task_struct *tsk = current;
336 memset(tsk->thread.debugreg, 0, sizeof(unsigned long)*8);
337 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
339 * Forget coprocessor state..
345 void release_thread(struct task_struct *dead_task)
348 // temporary debugging check
349 if (dead_task->mm->context.size) {
350 printk("WARNING: dead process %8s still has LDT? <%p/%d>\n",
352 dead_task->mm->context.ldt,
353 dead_task->mm->context.size);
358 release_vm86_irqs(dead_task);
362 * This gets called before we allocate a new thread and copy
363 * the current task into it.
365 void prepare_to_copy(struct task_struct *tsk)
370 int copy_thread(int nr, unsigned long clone_flags, unsigned long esp,
371 unsigned long unused,
372 struct task_struct * p, struct pt_regs * regs)
374 struct pt_regs * childregs;
375 struct task_struct *tsk;
378 childregs = ((struct pt_regs *) (THREAD_SIZE + (unsigned long) p->thread_info)) - 1;
381 childregs->esp = esp;
383 p->thread.esp = (unsigned long) childregs;
384 p->thread.esp0 = (unsigned long) (childregs+1);
386 p->thread.eip = (unsigned long) ret_from_fork;
388 savesegment(fs,p->thread.fs);
389 savesegment(gs,p->thread.gs);
392 if (unlikely(NULL != tsk->thread.io_bitmap_ptr)) {
393 p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL);
394 if (!p->thread.io_bitmap_ptr) {
395 p->thread.io_bitmap_max = 0;
398 memcpy(p->thread.io_bitmap_ptr, tsk->thread.io_bitmap_ptr,
403 * Set a new TLS for the child thread?
405 if (clone_flags & CLONE_SETTLS) {
406 struct desc_struct *desc;
407 struct user_desc info;
411 if (copy_from_user(&info, (void __user *)childregs->esi, sizeof(info)))
414 if (LDT_empty(&info))
417 idx = info.entry_number;
418 if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
421 desc = p->thread.tls_array + idx - GDT_ENTRY_TLS_MIN;
422 desc->a = LDT_entry_a(&info);
423 desc->b = LDT_entry_b(&info);
428 if (err && p->thread.io_bitmap_ptr) {
429 kfree(p->thread.io_bitmap_ptr);
430 p->thread.io_bitmap_max = 0;
436 * fill in the user structure for a core dump..
438 void dump_thread(struct pt_regs * regs, struct user * dump)
442 /* changed the size calculations - should hopefully work better. lbt */
443 dump->magic = CMAGIC;
444 dump->start_code = 0;
445 dump->start_stack = regs->esp & ~(PAGE_SIZE - 1);
446 dump->u_tsize = ((unsigned long) current->mm->end_code) >> PAGE_SHIFT;
447 dump->u_dsize = ((unsigned long) (current->mm->brk + (PAGE_SIZE-1))) >> PAGE_SHIFT;
448 dump->u_dsize -= dump->u_tsize;
450 for (i = 0; i < 8; i++)
451 dump->u_debugreg[i] = current->thread.debugreg[i];
453 if (dump->start_stack < TASK_SIZE)
454 dump->u_ssize = ((unsigned long) (TASK_SIZE - dump->start_stack)) >> PAGE_SHIFT;
456 dump->regs.ebx = regs->ebx;
457 dump->regs.ecx = regs->ecx;
458 dump->regs.edx = regs->edx;
459 dump->regs.esi = regs->esi;
460 dump->regs.edi = regs->edi;
461 dump->regs.ebp = regs->ebp;
462 dump->regs.eax = regs->eax;
463 dump->regs.ds = regs->xds;
464 dump->regs.es = regs->xes;
465 savesegment(fs,dump->regs.fs);
466 savesegment(gs,dump->regs.gs);
467 dump->regs.orig_eax = regs->orig_eax;
468 dump->regs.eip = regs->eip;
469 dump->regs.cs = regs->xcs;
470 dump->regs.eflags = regs->eflags;
471 dump->regs.esp = regs->esp;
472 dump->regs.ss = regs->xss;
474 dump->u_fpvalid = dump_fpu (regs, &dump->i387);
478 * Capture the user space registers if the task is not running (in user space)
480 int dump_task_regs(struct task_struct *tsk, elf_gregset_t *regs)
482 struct pt_regs ptregs;
484 ptregs = *(struct pt_regs *)
485 ((unsigned long)tsk->thread_info+THREAD_SIZE - sizeof(ptregs));
486 ptregs.xcs &= 0xffff;
487 ptregs.xds &= 0xffff;
488 ptregs.xes &= 0xffff;
489 ptregs.xss &= 0xffff;
491 elf_core_copy_regs(regs, &ptregs);
497 handle_io_bitmap(struct thread_struct *next, struct tss_struct *tss)
499 if (!next->io_bitmap_ptr) {
501 * Disable the bitmap via an invalid offset. We still cache
502 * the previous bitmap owner and the IO bitmap contents:
504 tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
507 if (likely(next == tss->io_bitmap_owner)) {
509 * Previous owner of the bitmap (hence the bitmap content)
510 * matches the next task, we dont have to do anything but
511 * to set a valid offset in the TSS:
513 tss->io_bitmap_base = IO_BITMAP_OFFSET;
517 * Lazy TSS's I/O bitmap copy. We set an invalid offset here
518 * and we let the task to get a GPF in case an I/O instruction
519 * is performed. The handler of the GPF will verify that the
520 * faulting task has a valid I/O bitmap and, it true, does the
521 * real copy and restart the instruction. This will save us
522 * redundant copies when the currently switched task does not
523 * perform any I/O during its timeslice.
525 tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET_LAZY;
528 * This special macro can be used to load a debugging register
530 #define loaddebug(thread,register) \
531 __asm__("movl %0,%%db" #register \
533 :"r" (thread->debugreg[register]))
536 * switch_to(x,yn) should switch tasks from x to y.
538 * We fsave/fwait so that an exception goes off at the right time
539 * (as a call from the fsave or fwait in effect) rather than to
540 * the wrong process. Lazy FP saving no longer makes any sense
541 * with modern CPU's, and this simplifies a lot of things (SMP
542 * and UP become the same).
544 * NOTE! We used to use the x86 hardware context switching. The
545 * reason for not using it any more becomes apparent when you
546 * try to recover gracefully from saved state that is no longer
547 * valid (stale segment register values in particular). With the
548 * hardware task-switch, there is no way to fix up bad state in
549 * a reasonable manner.
551 * The fact that Intel documents the hardware task-switching to
552 * be slow is a fairly red herring - this code is not noticeably
553 * faster. However, there _is_ some room for improvement here,
554 * so the performance issues may eventually be a valid point.
555 * More important, however, is the fact that this allows us much
558 * The return value (in %eax) will be the "prev" task after
559 * the task-switch, and shows up in ret_from_fork in entry.S,
562 struct task_struct fastcall * __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
564 struct thread_struct *prev = &prev_p->thread,
565 *next = &next_p->thread;
566 int cpu = smp_processor_id();
567 struct tss_struct *tss = &per_cpu(init_tss, cpu);
569 /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
571 __unlazy_fpu(prev_p);
573 load_user_cs_desc(cpu, next_p->mm);
576 * Reload esp0, LDT and the page table pointer:
578 load_esp0(tss, next);
581 * Load the per-thread Thread-Local Storage descriptor.
586 * Save away %fs and %gs. No need to save %es and %ds, as
587 * those are always kernel segments while inside the kernel.
589 asm volatile("movl %%fs,%0":"=m" (*(int *)&prev->fs));
590 asm volatile("movl %%gs,%0":"=m" (*(int *)&prev->gs));
593 * Restore %fs and %gs if needed.
595 if (unlikely(prev->fs | prev->gs | next->fs | next->gs)) {
596 loadsegment(fs, next->fs);
597 loadsegment(gs, next->gs);
601 * Now maybe reload the debug registers
603 if (unlikely(next->debugreg[7])) {
613 if (unlikely(prev->io_bitmap_ptr || next->io_bitmap_ptr))
614 handle_io_bitmap(next, tss);
619 asmlinkage int sys_fork(struct pt_regs regs)
621 return do_fork(SIGCHLD, regs.esp, ®s, 0, NULL, NULL);
624 asmlinkage int sys_clone(struct pt_regs regs)
626 unsigned long clone_flags;
628 int __user *parent_tidptr, *child_tidptr;
630 clone_flags = regs.ebx;
632 parent_tidptr = (int __user *)regs.edx;
633 child_tidptr = (int __user *)regs.edi;
636 return do_fork(clone_flags, newsp, ®s, 0, parent_tidptr, child_tidptr);
640 * This is trivial, and on the face of it looks like it
641 * could equally well be done in user mode.
643 * Not so, for quite unobvious reasons - register pressure.
644 * In user mode vfork() cannot have a stack frame, and if
645 * done by calling the "clone()" system call directly, you
646 * do not have enough call-clobbered registers to hold all
647 * the information you need.
649 asmlinkage int sys_vfork(struct pt_regs regs)
651 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs.esp, ®s, 0, NULL, NULL);
655 * sys_execve() executes a new program.
657 asmlinkage int sys_execve(struct pt_regs regs)
662 filename = getname((char __user *) regs.ebx);
663 error = PTR_ERR(filename);
664 if (IS_ERR(filename))
666 error = do_execve(filename,
667 (char __user * __user *) regs.ecx,
668 (char __user * __user *) regs.edx,
672 current->ptrace &= ~PT_DTRACE;
673 task_unlock(current);
674 /* Make sure we don't return using sysenter.. */
675 set_thread_flag(TIF_IRET);
682 #define top_esp (THREAD_SIZE - sizeof(unsigned long))
683 #define top_ebp (THREAD_SIZE - 2*sizeof(unsigned long))
685 unsigned long get_wchan(struct task_struct *p)
687 unsigned long ebp, esp, eip;
688 unsigned long stack_page;
690 if (!p || p == current || p->state == TASK_RUNNING)
692 stack_page = (unsigned long)p->thread_info;
694 if (!stack_page || esp < stack_page || esp > top_esp+stack_page)
696 /* include/asm-i386/system.h:switch_to() pushes ebp last. */
697 ebp = *(unsigned long *) esp;
699 if (ebp < stack_page || ebp > top_ebp+stack_page)
701 eip = *(unsigned long *) (ebp+4);
702 if (!in_sched_functions(eip))
704 ebp = *(unsigned long *) ebp;
705 } while (count++ < 16);
710 * sys_alloc_thread_area: get a yet unused TLS descriptor index.
712 static int get_free_idx(void)
714 struct thread_struct *t = ¤t->thread;
717 for (idx = 0; idx < GDT_ENTRY_TLS_ENTRIES; idx++)
718 if (desc_empty(t->tls_array + idx))
719 return idx + GDT_ENTRY_TLS_MIN;
724 * Set a given TLS descriptor:
726 asmlinkage int sys_set_thread_area(struct user_desc __user *u_info)
728 struct thread_struct *t = ¤t->thread;
729 struct user_desc info;
730 struct desc_struct *desc;
733 if (copy_from_user(&info, u_info, sizeof(info)))
735 idx = info.entry_number;
738 * index -1 means the kernel should try to find and
739 * allocate an empty descriptor:
742 idx = get_free_idx();
745 if (put_user(idx, &u_info->entry_number))
749 if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
752 desc = t->tls_array + idx - GDT_ENTRY_TLS_MIN;
755 * We must not get preempted while modifying the TLS.
759 if (LDT_empty(&info)) {
763 desc->a = LDT_entry_a(&info);
764 desc->b = LDT_entry_b(&info);
774 * Get the current Thread-Local Storage area:
777 #define GET_BASE(desc) ( \
778 (((desc)->a >> 16) & 0x0000ffff) | \
779 (((desc)->b << 16) & 0x00ff0000) | \
780 ( (desc)->b & 0xff000000) )
782 #define GET_LIMIT(desc) ( \
783 ((desc)->a & 0x0ffff) | \
784 ((desc)->b & 0xf0000) )
786 #define GET_32BIT(desc) (((desc)->b >> 22) & 1)
787 #define GET_CONTENTS(desc) (((desc)->b >> 10) & 3)
788 #define GET_WRITABLE(desc) (((desc)->b >> 9) & 1)
789 #define GET_LIMIT_PAGES(desc) (((desc)->b >> 23) & 1)
790 #define GET_PRESENT(desc) (((desc)->b >> 15) & 1)
791 #define GET_USEABLE(desc) (((desc)->b >> 20) & 1)
793 asmlinkage int sys_get_thread_area(struct user_desc __user *u_info)
795 struct user_desc info;
796 struct desc_struct *desc;
799 if (get_user(idx, &u_info->entry_number))
801 if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
804 desc = current->thread.tls_array + idx - GDT_ENTRY_TLS_MIN;
806 info.entry_number = idx;
807 info.base_addr = GET_BASE(desc);
808 info.limit = GET_LIMIT(desc);
809 info.seg_32bit = GET_32BIT(desc);
810 info.contents = GET_CONTENTS(desc);
811 info.read_exec_only = !GET_WRITABLE(desc);
812 info.limit_in_pages = GET_LIMIT_PAGES(desc);
813 info.seg_not_present = !GET_PRESENT(desc);
814 info.useable = GET_USEABLE(desc);
816 if (copy_to_user(u_info, &info, sizeof(info)))
822 unsigned long arch_align_stack(unsigned long sp)
824 if (current->flags & PF_RELOCEXEC)
825 sp -= ((get_random_int() % 65536) << 4);
830 void arch_add_exec_range(struct mm_struct *mm, unsigned long limit)
832 if (limit > mm->context.exec_limit) {
833 mm->context.exec_limit = limit;
834 set_user_cs(&mm->context.user_cs, limit);
835 if (mm == current->mm)
836 load_user_cs_desc(smp_processor_id(), mm);
840 void arch_remove_exec_range(struct mm_struct *mm, unsigned long old_end)
842 struct vm_area_struct *vma;
843 unsigned long limit = 0;
845 if (old_end == mm->context.exec_limit) {
846 for (vma = mm->mmap; vma; vma = vma->vm_next)
847 if ((vma->vm_flags & VM_EXEC) && (vma->vm_end > limit))
850 mm->context.exec_limit = limit;
851 set_user_cs(&mm->context.user_cs, limit);
852 if (mm == current->mm)
853 load_user_cs_desc(smp_processor_id(), mm);
857 void arch_flush_exec_range(struct mm_struct *mm)
859 mm->context.exec_limit = 0;
860 set_user_cs(&mm->context.user_cs, 0);
864 * Generate random brk address between 128MB and 196MB. (if the layout
867 void randomize_brk(unsigned long old_brk)
869 unsigned long new_brk, range_start, range_end;
871 range_start = 0x08000000;
872 if (current->mm->brk >= range_start)
873 range_start = current->mm->brk;
874 range_end = range_start + 0x02000000;
875 new_brk = randomize_range(range_start, range_end, 0);
877 current->mm->brk = new_brk;