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/version.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");
63 * Return saved PC of a blocked thread.
65 unsigned long thread_saved_pc(struct task_struct *tsk)
67 return ((unsigned long *)tsk->thread.esp)[3];
71 * Powermanagement idle function, if any..
73 void (*pm_idle)(void);
75 void disable_hlt(void)
80 EXPORT_SYMBOL(disable_hlt);
87 EXPORT_SYMBOL(enable_hlt);
90 * We use this if we don't have any better
93 void default_idle(void)
95 if (!hlt_counter && current_cpu_data.hlt_works_ok) {
105 * On SMP it's slightly faster (but much more power-consuming!)
106 * to poll the ->work.need_resched flag instead of waiting for the
107 * cross-CPU IPI to arrive. Use this option with caution.
109 static void poll_idle (void)
116 * Deal with another CPU just having chosen a thread to
119 oldval = test_and_clear_thread_flag(TIF_NEED_RESCHED);
122 set_thread_flag(TIF_POLLING_NRFLAG);
128 : : "i"(_TIF_NEED_RESCHED), "m" (current_thread_info()->flags));
130 clear_thread_flag(TIF_POLLING_NRFLAG);
137 * The idle thread. There's no useful work to be
138 * done, so just try to conserve power and have a
139 * low exit latency (ie sit in a loop waiting for
140 * somebody to say that they'd like to reschedule)
144 /* endless idle loop with no priority at all */
146 while (!need_resched()) {
149 * Mark this as an RCU critical section so that
150 * synchronize_kernel() in the unload path waits
151 * for our completion.
159 irq_stat[smp_processor_id()].idle_timestamp = jiffies;
168 * This uses new MONITOR/MWAIT instructions on P4 processors with PNI,
169 * which can obviate IPI to trigger checking of need_resched.
170 * We execute MONITOR against need_resched and enter optimized wait state
171 * through MWAIT. Whenever someone changes need_resched, we would be woken
172 * up from MWAIT (without an IPI).
174 static void mwait_idle(void)
178 if (!need_resched()) {
179 set_thread_flag(TIF_POLLING_NRFLAG);
181 __monitor((void *)¤t_thread_info()->flags, 0, 0);
185 } while (!need_resched());
186 clear_thread_flag(TIF_POLLING_NRFLAG);
190 void __init select_idle_routine(const struct cpuinfo_x86 *c)
192 if (cpu_has(c, X86_FEATURE_MWAIT)) {
193 printk("monitor/mwait feature present.\n");
195 * Skip, if setup has overridden idle.
196 * One CPU supports mwait => All CPUs supports mwait
199 printk("using mwait in idle threads.\n");
200 pm_idle = mwait_idle;
205 static int __init idle_setup (char *str)
207 if (!strncmp(str, "poll", 4)) {
208 printk("using polling idle threads.\n");
210 #ifdef CONFIG_X86_SMP
211 if (smp_num_siblings > 1)
212 printk("WARNING: polling idle and HT enabled, performance may degrade.\n");
214 } else if (!strncmp(str, "halt", 4)) {
215 printk("using halt in idle threads.\n");
216 pm_idle = default_idle;
222 __setup("idle=", idle_setup);
224 void stack_overflow(void)
226 unsigned long esp = current_stack_pointer();
227 int panicing = ((esp&(THREAD_SIZE-1)) <= STACK_PANIC);
229 oops_in_progress = 1;
230 printk( "esp: 0x%lx masked: 0x%lx STACK_PANIC:0x%lx %d %d\n",
231 esp, (esp&(THREAD_SIZE-1)), STACK_PANIC,
232 (((esp&(THREAD_SIZE-1)) <= STACK_PANIC)), panicing);
233 show_trace(current,(void*)esp);
236 panic("stack overflow\n");
238 oops_in_progress = 0;
240 /* Just let it happen once per task, as otherwise it goes nuts
241 * in printing stack traces. This means that I need to dump
242 * the stack_overflowed boolean into the task or thread_info
243 * structure. For now just turn it off all together.
246 /* stack_overflowed = 0; */
249 void show_regs(struct pt_regs * regs)
251 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L;
254 printk("Pid: %d, comm: %20s\n", current->pid, current->comm);
255 printk("EIP: %04x:[<%08lx>] CPU: %d\n",0xffff & regs->xcs,regs->eip, smp_processor_id());
256 print_symbol("EIP is at %s\n", regs->eip);
259 printk(" ESP: %04x:%08lx",0xffff & regs->xss,regs->esp);
260 printk(" EFLAGS: %08lx %s (%s)\n",regs->eflags, print_tainted(),UTS_RELEASE);
261 printk("EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
262 regs->eax,regs->ebx,regs->ecx,regs->edx);
263 printk("ESI: %08lx EDI: %08lx EBP: %08lx",
264 regs->esi, regs->edi, regs->ebp);
265 printk(" DS: %04x ES: %04x\n",
266 0xffff & regs->xds,0xffff & regs->xes);
268 __asm__("movl %%cr0, %0": "=r" (cr0));
269 __asm__("movl %%cr2, %0": "=r" (cr2));
270 __asm__("movl %%cr3, %0": "=r" (cr3));
271 /* This could fault if %cr4 does not exist */
272 __asm__("1: movl %%cr4, %0 \n"
274 ".section __ex_table,\"a\" \n"
277 : "=r" (cr4): "0" (0));
278 printk("CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n", cr0, cr2, cr3, cr4);
279 show_trace(NULL, ®s->esp);
283 * This gets run with %ebx containing the
284 * function to call, and %edx containing
287 extern void kernel_thread_helper(void);
288 __asm__(".section .text\n"
290 "kernel_thread_helper:\n\t"
299 * Create a kernel thread
301 int kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
305 memset(®s, 0, sizeof(regs));
307 regs.ebx = (unsigned long) fn;
308 regs.edx = (unsigned long) arg;
310 regs.xds = __USER_DS;
311 regs.xes = __USER_DS;
313 regs.eip = (unsigned long) kernel_thread_helper;
314 regs.xcs = __KERNEL_CS;
315 regs.eflags = X86_EFLAGS_IF | X86_EFLAGS_SF | X86_EFLAGS_PF | 0x2;
317 /* Ok, create the new process.. */
318 return do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0, ®s, 0, NULL, NULL);
322 * Free current thread data structures etc..
324 void exit_thread(void)
326 struct task_struct *tsk = current;
327 struct thread_struct *t = &tsk->thread;
329 /* The process may have allocated an io port bitmap... nuke it. */
330 if (unlikely(NULL != t->io_bitmap_ptr)) {
332 struct tss_struct *tss = &per_cpu(init_tss, cpu);
334 kfree(t->io_bitmap_ptr);
335 t->io_bitmap_ptr = NULL;
337 * Careful, clear this in the TSS too:
339 memset(tss->io_bitmap, 0xff, tss->io_bitmap_max);
340 t->io_bitmap_max = 0;
341 tss->io_bitmap_owner = NULL;
342 tss->io_bitmap_max = 0;
343 tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
348 void flush_thread(void)
350 struct task_struct *tsk = current;
352 memset(tsk->thread.debugreg, 0, sizeof(unsigned long)*8);
353 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
355 * Forget coprocessor state..
361 void release_thread(struct task_struct *dead_task)
364 // temporary debugging check
365 if (dead_task->mm->context.size) {
366 printk("WARNING: dead process %8s still has LDT? <%p/%d>\n",
368 dead_task->mm->context.ldt,
369 dead_task->mm->context.size);
374 release_x86_irqs(dead_task);
378 * This gets called before we allocate a new thread and copy
379 * the current task into it.
381 void prepare_to_copy(struct task_struct *tsk)
386 int copy_thread(int nr, unsigned long clone_flags, unsigned long esp,
387 unsigned long unused,
388 struct task_struct * p, struct pt_regs * regs)
390 struct pt_regs * childregs;
391 struct task_struct *tsk;
394 childregs = ((struct pt_regs *) (THREAD_SIZE + (unsigned long) p->thread_info)) - 1;
397 childregs->esp = esp;
398 p->set_child_tid = p->clear_child_tid = NULL;
400 p->thread.esp = (unsigned long) childregs;
401 p->thread.esp0 = (unsigned long) (childregs+1);
403 p->thread.eip = (unsigned long) ret_from_fork;
405 savesegment(fs,p->thread.fs);
406 savesegment(gs,p->thread.gs);
409 if (unlikely(NULL != tsk->thread.io_bitmap_ptr)) {
410 p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL);
411 if (!p->thread.io_bitmap_ptr) {
412 p->thread.io_bitmap_max = 0;
415 memcpy(p->thread.io_bitmap_ptr, tsk->thread.io_bitmap_ptr,
420 * Set a new TLS for the child thread?
422 if (clone_flags & CLONE_SETTLS) {
423 struct desc_struct *desc;
424 struct user_desc info;
428 if (copy_from_user(&info, (void __user *)childregs->esi, sizeof(info)))
431 if (LDT_empty(&info))
434 idx = info.entry_number;
435 if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
438 desc = p->thread.tls_array + idx - GDT_ENTRY_TLS_MIN;
439 desc->a = LDT_entry_a(&info);
440 desc->b = LDT_entry_b(&info);
445 if (err && p->thread.io_bitmap_ptr) {
446 kfree(p->thread.io_bitmap_ptr);
447 p->thread.io_bitmap_max = 0;
453 * fill in the user structure for a core dump..
455 void dump_thread(struct pt_regs * regs, struct user * dump)
459 /* changed the size calculations - should hopefully work better. lbt */
460 dump->magic = CMAGIC;
461 dump->start_code = 0;
462 dump->start_stack = regs->esp & ~(PAGE_SIZE - 1);
463 dump->u_tsize = ((unsigned long) current->mm->end_code) >> PAGE_SHIFT;
464 dump->u_dsize = ((unsigned long) (current->mm->brk + (PAGE_SIZE-1))) >> PAGE_SHIFT;
465 dump->u_dsize -= dump->u_tsize;
467 for (i = 0; i < 8; i++)
468 dump->u_debugreg[i] = current->thread.debugreg[i];
470 if (dump->start_stack < TASK_SIZE)
471 dump->u_ssize = ((unsigned long) (TASK_SIZE - dump->start_stack)) >> PAGE_SHIFT;
473 dump->regs.ebx = regs->ebx;
474 dump->regs.ecx = regs->ecx;
475 dump->regs.edx = regs->edx;
476 dump->regs.esi = regs->esi;
477 dump->regs.edi = regs->edi;
478 dump->regs.ebp = regs->ebp;
479 dump->regs.eax = regs->eax;
480 dump->regs.ds = regs->xds;
481 dump->regs.es = regs->xes;
482 savesegment(fs,dump->regs.fs);
483 savesegment(gs,dump->regs.gs);
484 dump->regs.orig_eax = regs->orig_eax;
485 dump->regs.eip = regs->eip;
486 dump->regs.cs = regs->xcs;
487 dump->regs.eflags = regs->eflags;
488 dump->regs.esp = regs->esp;
489 dump->regs.ss = regs->xss;
491 dump->u_fpvalid = dump_fpu (regs, &dump->i387);
495 * Capture the user space registers if the task is not running (in user space)
497 int dump_task_regs(struct task_struct *tsk, elf_gregset_t *regs)
499 struct pt_regs ptregs;
501 ptregs = *(struct pt_regs *)
502 ((unsigned long)tsk->thread_info+THREAD_SIZE - sizeof(ptregs));
503 ptregs.xcs &= 0xffff;
504 ptregs.xds &= 0xffff;
505 ptregs.xes &= 0xffff;
506 ptregs.xss &= 0xffff;
508 elf_core_copy_regs(regs, &ptregs);
514 handle_io_bitmap(struct thread_struct *next, struct tss_struct *tss)
516 if (!next->io_bitmap_ptr) {
518 * Disable the bitmap via an invalid offset. We still cache
519 * the previous bitmap owner and the IO bitmap contents:
521 tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
524 if (likely(next == tss->io_bitmap_owner)) {
526 * Previous owner of the bitmap (hence the bitmap content)
527 * matches the next task, we dont have to do anything but
528 * to set a valid offset in the TSS:
530 tss->io_bitmap_base = IO_BITMAP_OFFSET;
534 * Lazy TSS's I/O bitmap copy. We set an invalid offset here
535 * and we let the task to get a GPF in case an I/O instruction
536 * is performed. The handler of the GPF will verify that the
537 * faulting task has a valid I/O bitmap and, it true, does the
538 * real copy and restart the instruction. This will save us
539 * redundant copies when the currently switched task does not
540 * perform any I/O during its timeslice.
542 tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET_LAZY;
545 * This special macro can be used to load a debugging register
547 #define loaddebug(thread,register) \
548 __asm__("movl %0,%%db" #register \
550 :"r" (thread->debugreg[register]))
553 * switch_to(x,yn) should switch tasks from x to y.
555 * We fsave/fwait so that an exception goes off at the right time
556 * (as a call from the fsave or fwait in effect) rather than to
557 * the wrong process. Lazy FP saving no longer makes any sense
558 * with modern CPU's, and this simplifies a lot of things (SMP
559 * and UP become the same).
561 * NOTE! We used to use the x86 hardware context switching. The
562 * reason for not using it any more becomes apparent when you
563 * try to recover gracefully from saved state that is no longer
564 * valid (stale segment register values in particular). With the
565 * hardware task-switch, there is no way to fix up bad state in
566 * a reasonable manner.
568 * The fact that Intel documents the hardware task-switching to
569 * be slow is a fairly red herring - this code is not noticeably
570 * faster. However, there _is_ some room for improvement here,
571 * so the performance issues may eventually be a valid point.
572 * More important, however, is the fact that this allows us much
575 * The return value (in %eax) will be the "prev" task after
576 * the task-switch, and shows up in ret_from_fork in entry.S,
579 struct task_struct fastcall * __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
581 struct thread_struct *prev = &prev_p->thread,
582 *next = &next_p->thread;
583 int cpu = smp_processor_id();
584 struct tss_struct *tss = &per_cpu(init_tss, cpu);
586 /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
588 __unlazy_fpu(prev_p);
590 load_user_cs_desc(cpu, next_p->mm);
593 * Reload esp0, LDT and the page table pointer:
595 load_esp0(tss, next);
598 * Load the per-thread Thread-Local Storage descriptor.
603 * Save away %fs and %gs. No need to save %es and %ds, as
604 * those are always kernel segments while inside the kernel.
606 asm volatile("movl %%fs,%0":"=m" (*(int *)&prev->fs));
607 asm volatile("movl %%gs,%0":"=m" (*(int *)&prev->gs));
610 * Restore %fs and %gs if needed.
612 if (unlikely(prev->fs | prev->gs | next->fs | next->gs)) {
613 loadsegment(fs, next->fs);
614 loadsegment(gs, next->gs);
618 * Now maybe reload the debug registers
620 if (unlikely(next->debugreg[7])) {
630 if (unlikely(prev->io_bitmap_ptr || next->io_bitmap_ptr))
631 handle_io_bitmap(next, tss);
636 asmlinkage int sys_fork(struct pt_regs regs)
638 return do_fork(SIGCHLD, regs.esp, ®s, 0, NULL, NULL);
641 asmlinkage int sys_clone(struct pt_regs regs)
643 unsigned long clone_flags;
645 int __user *parent_tidptr, *child_tidptr;
647 clone_flags = regs.ebx;
649 parent_tidptr = (int __user *)regs.edx;
650 child_tidptr = (int __user *)regs.edi;
653 return do_fork(clone_flags, newsp, ®s, 0, parent_tidptr, child_tidptr);
657 * This is trivial, and on the face of it looks like it
658 * could equally well be done in user mode.
660 * Not so, for quite unobvious reasons - register pressure.
661 * In user mode vfork() cannot have a stack frame, and if
662 * done by calling the "clone()" system call directly, you
663 * do not have enough call-clobbered registers to hold all
664 * the information you need.
666 asmlinkage int sys_vfork(struct pt_regs regs)
668 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs.esp, ®s, 0, NULL, NULL);
672 * sys_execve() executes a new program.
674 asmlinkage int sys_execve(struct pt_regs regs)
679 filename = getname((char __user *) regs.ebx);
680 error = PTR_ERR(filename);
681 if (IS_ERR(filename))
683 error = do_execve(filename,
684 (char __user * __user *) regs.ecx,
685 (char __user * __user *) regs.edx,
688 current->ptrace &= ~PT_DTRACE;
689 /* Make sure we don't return using sysenter.. */
690 set_thread_flag(TIF_IRET);
697 #define top_esp (THREAD_SIZE - sizeof(unsigned long))
698 #define top_ebp (THREAD_SIZE - 2*sizeof(unsigned long))
700 unsigned long get_wchan(struct task_struct *p)
702 unsigned long ebp, esp, eip;
703 unsigned long stack_page;
705 if (!p || p == current || p->state == TASK_RUNNING)
707 stack_page = (unsigned long)p->thread_info;
709 if (!stack_page || esp < stack_page || esp > top_esp+stack_page)
711 /* include/asm-i386/system.h:switch_to() pushes ebp last. */
712 ebp = *(unsigned long *) esp;
714 if (ebp < stack_page || ebp > top_ebp+stack_page)
716 eip = *(unsigned long *) (ebp+4);
717 if (!in_sched_functions(eip))
719 ebp = *(unsigned long *) ebp;
720 } while (count++ < 16);
725 * sys_alloc_thread_area: get a yet unused TLS descriptor index.
727 static int get_free_idx(void)
729 struct thread_struct *t = ¤t->thread;
732 for (idx = 0; idx < GDT_ENTRY_TLS_ENTRIES; idx++)
733 if (desc_empty(t->tls_array + idx))
734 return idx + GDT_ENTRY_TLS_MIN;
739 * Set a given TLS descriptor:
741 asmlinkage int sys_set_thread_area(struct user_desc __user *u_info)
743 struct thread_struct *t = ¤t->thread;
744 struct user_desc info;
745 struct desc_struct *desc;
748 if (copy_from_user(&info, u_info, sizeof(info)))
750 idx = info.entry_number;
753 * index -1 means the kernel should try to find and
754 * allocate an empty descriptor:
757 idx = get_free_idx();
760 if (put_user(idx, &u_info->entry_number))
764 if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
767 desc = t->tls_array + idx - GDT_ENTRY_TLS_MIN;
770 * We must not get preempted while modifying the TLS.
774 if (LDT_empty(&info)) {
778 desc->a = LDT_entry_a(&info);
779 desc->b = LDT_entry_b(&info);
789 * Get the current Thread-Local Storage area:
792 #define GET_BASE(desc) ( \
793 (((desc)->a >> 16) & 0x0000ffff) | \
794 (((desc)->b << 16) & 0x00ff0000) | \
795 ( (desc)->b & 0xff000000) )
797 #define GET_LIMIT(desc) ( \
798 ((desc)->a & 0x0ffff) | \
799 ((desc)->b & 0xf0000) )
801 #define GET_32BIT(desc) (((desc)->b >> 22) & 1)
802 #define GET_CONTENTS(desc) (((desc)->b >> 10) & 3)
803 #define GET_WRITABLE(desc) (((desc)->b >> 9) & 1)
804 #define GET_LIMIT_PAGES(desc) (((desc)->b >> 23) & 1)
805 #define GET_PRESENT(desc) (((desc)->b >> 15) & 1)
806 #define GET_USEABLE(desc) (((desc)->b >> 20) & 1)
808 asmlinkage int sys_get_thread_area(struct user_desc __user *u_info)
810 struct user_desc info;
811 struct desc_struct *desc;
814 if (get_user(idx, &u_info->entry_number))
816 if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
819 desc = current->thread.tls_array + idx - GDT_ENTRY_TLS_MIN;
821 info.entry_number = idx;
822 info.base_addr = GET_BASE(desc);
823 info.limit = GET_LIMIT(desc);
824 info.seg_32bit = GET_32BIT(desc);
825 info.contents = GET_CONTENTS(desc);
826 info.read_exec_only = !GET_WRITABLE(desc);
827 info.limit_in_pages = GET_LIMIT_PAGES(desc);
828 info.seg_not_present = !GET_PRESENT(desc);
829 info.useable = GET_USEABLE(desc);
831 if (copy_to_user(u_info, &info, sizeof(info)))
837 unsigned long arch_align_stack(unsigned long sp)
839 if (current->flags & PF_RELOCEXEC)
840 sp -= ((get_random_int() % 65536) << 4);
845 void arch_add_exec_range(struct mm_struct *mm, unsigned long limit)
847 if (limit > mm->context.exec_limit) {
848 mm->context.exec_limit = limit;
849 set_user_cs(&mm->context.user_cs, limit);
850 if (mm == current->mm)
851 load_user_cs_desc(smp_processor_id(), mm);
855 void arch_remove_exec_range(struct mm_struct *mm, unsigned long old_end)
857 struct vm_area_struct *vma;
858 unsigned long limit = 0;
860 if (old_end == mm->context.exec_limit) {
861 for (vma = mm->mmap; vma; vma = vma->vm_next)
862 if ((vma->vm_flags & VM_EXEC) && (vma->vm_end > limit))
865 mm->context.exec_limit = limit;
866 set_user_cs(&mm->context.user_cs, limit);
867 if (mm == current->mm)
868 load_user_cs_desc(smp_processor_id(), mm);
872 void arch_flush_exec_range(struct mm_struct *mm)
874 mm->context.exec_limit = 0;
875 set_user_cs(&mm->context.user_cs, 0);
879 * Generate random brk address between 128MB and 196MB. (if the layout
882 void randomize_brk(unsigned long old_brk)
884 unsigned long new_brk, range_start, range_end;
886 range_start = 0x08000000;
887 if (current->mm->brk >= range_start)
888 range_start = current->mm->brk;
889 range_end = range_start + 0x02000000;
890 new_brk = randomize_range(range_start, range_end, 0);
892 current->mm->brk = new_brk;