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 #include <asm/atomic_kmap.h>
52 #ifdef CONFIG_MATH_EMULATION
53 #include <asm/math_emu.h>
56 #include <linux/irq.h>
57 #include <linux/err.h>
59 asmlinkage void ret_from_fork(void) __asm__("ret_from_fork");
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);
76 void disable_hlt(void)
81 EXPORT_SYMBOL(disable_hlt);
88 EXPORT_SYMBOL(enable_hlt);
91 * We use this if we don't have any better
94 void default_idle(void)
96 if (!hlt_counter && current_cpu_data.hlt_works_ok) {
106 * On SMP it's slightly faster (but much more power-consuming!)
107 * to poll the ->work.need_resched flag instead of waiting for the
108 * cross-CPU IPI to arrive. Use this option with caution.
110 static void poll_idle (void)
117 * Deal with another CPU just having chosen a thread to
120 oldval = test_and_clear_thread_flag(TIF_NEED_RESCHED);
123 set_thread_flag(TIF_POLLING_NRFLAG);
129 : : "i"(_TIF_NEED_RESCHED), "m" (current_thread_info()->flags));
131 clear_thread_flag(TIF_POLLING_NRFLAG);
138 * The idle thread. There's no useful work to be
139 * done, so just try to conserve power and have a
140 * low exit latency (ie sit in a loop waiting for
141 * somebody to say that they'd like to reschedule)
145 /* endless idle loop with no priority at all */
147 while (!need_resched()) {
148 void (*idle)(void) = pm_idle;
153 irq_stat[smp_processor_id()].idle_timestamp = jiffies;
161 * This uses new MONITOR/MWAIT instructions on P4 processors with PNI,
162 * which can obviate IPI to trigger checking of need_resched.
163 * We execute MONITOR against need_resched and enter optimized wait state
164 * through MWAIT. Whenever someone changes need_resched, we would be woken
165 * up from MWAIT (without an IPI).
167 static void mwait_idle(void)
171 if (!need_resched()) {
172 set_thread_flag(TIF_POLLING_NRFLAG);
174 __monitor((void *)¤t_thread_info()->flags, 0, 0);
178 } while (!need_resched());
179 clear_thread_flag(TIF_POLLING_NRFLAG);
183 void __init select_idle_routine(const struct cpuinfo_x86 *c)
185 if (cpu_has(c, X86_FEATURE_MWAIT)) {
186 printk("monitor/mwait feature present.\n");
188 * Skip, if setup has overridden idle.
189 * Also, take care of system with asymmetric CPUs.
190 * Use, mwait_idle only if all cpus support it.
191 * If not, we fallback to default_idle()
194 printk("using mwait in idle threads.\n");
195 pm_idle = mwait_idle;
199 pm_idle = default_idle;
203 static int __init idle_setup (char *str)
205 if (!strncmp(str, "poll", 4)) {
206 printk("using polling idle threads.\n");
208 #ifdef CONFIG_X86_SMP
209 if (smp_num_siblings > 1)
210 printk("WARNING: polling idle and HT enabled, performance may degrade.\n");
212 } else if (!strncmp(str, "halt", 4)) {
213 printk("using halt in idle threads.\n");
214 pm_idle = default_idle;
220 __setup("idle=", idle_setup);
222 void show_regs(struct pt_regs * regs)
224 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L;
227 printk("Pid: %d, comm: %20s\n", current->pid, current->comm);
228 printk("EIP: %04x:[<%08lx>] CPU: %d\n",0xffff & regs->xcs,regs->eip, smp_processor_id());
229 print_symbol("EIP is at %s\n", regs->eip);
232 printk(" ESP: %04x:%08lx",0xffff & regs->xss,regs->esp);
233 printk(" EFLAGS: %08lx %s (%s)\n",regs->eflags, print_tainted(),UTS_RELEASE);
234 printk("EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
235 regs->eax,regs->ebx,regs->ecx,regs->edx);
236 printk("ESI: %08lx EDI: %08lx EBP: %08lx",
237 regs->esi, regs->edi, regs->ebp);
238 printk(" DS: %04x ES: %04x\n",
239 0xffff & regs->xds,0xffff & regs->xes);
241 __asm__("movl %%cr0, %0": "=r" (cr0));
242 __asm__("movl %%cr2, %0": "=r" (cr2));
243 __asm__("movl %%cr3, %0": "=r" (cr3));
244 /* This could fault if %cr4 does not exist */
245 __asm__("1: movl %%cr4, %0 \n"
247 ".section __ex_table,\"a\" \n"
250 : "=r" (cr4): "0" (0));
251 printk("CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n", cr0, cr2, cr3, cr4);
252 show_trace(NULL, ®s->esp);
255 EXPORT_SYMBOL_GPL(show_regs);
258 * This gets run with %ebx containing the
259 * function to call, and %edx containing
262 extern void kernel_thread_helper(void);
263 __asm__(".section .text\n"
265 "kernel_thread_helper:\n\t"
274 * Create a kernel thread
276 int kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
280 memset(®s, 0, sizeof(regs));
282 regs.ebx = (unsigned long) fn;
283 regs.edx = (unsigned long) arg;
285 regs.xds = __USER_DS;
286 regs.xes = __USER_DS;
288 regs.eip = (unsigned long) kernel_thread_helper;
289 regs.xcs = __KERNEL_CS;
290 regs.eflags = X86_EFLAGS_IF | X86_EFLAGS_SF | X86_EFLAGS_PF | 0x2;
292 /* Ok, create the new process.. */
293 return do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0, ®s, 0, NULL, NULL);
297 * Free current thread data structures etc..
299 void exit_thread(void)
301 struct task_struct *tsk = current;
303 /* The process may have allocated an io port bitmap... nuke it. */
304 if (unlikely(NULL != tsk->thread.io_bitmap_ptr)) {
306 struct tss_struct *tss = init_tss + cpu;
307 kfree(tsk->thread.io_bitmap_ptr);
308 tsk->thread.io_bitmap_ptr = NULL;
309 tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
314 void flush_thread(void)
316 struct task_struct *tsk = current;
318 memset(tsk->thread.debugreg, 0, sizeof(unsigned long)*8);
319 #ifdef CONFIG_X86_HIGH_ENTRY
320 clear_thread_flag(TIF_DB7);
322 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
324 * Forget coprocessor state..
330 void release_thread(struct task_struct *dead_task)
333 // temporary debugging check
334 if (dead_task->mm->context.size) {
335 printk("WARNING: dead process %8s still has LDT? <%d>\n",
337 dead_task->mm->context.size);
342 release_x86_irqs(dead_task);
346 * This gets called before we allocate a new thread and copy
347 * the current task into it.
349 void prepare_to_copy(struct task_struct *tsk)
354 int copy_thread(int nr, unsigned long clone_flags, unsigned long esp,
355 unsigned long unused,
356 struct task_struct * p, struct pt_regs * regs)
358 struct pt_regs * childregs;
359 struct task_struct *tsk;
362 childregs = ((struct pt_regs *) (THREAD_SIZE + (unsigned long) p->thread_info)) - 1;
365 childregs->esp = esp;
366 p->set_child_tid = p->clear_child_tid = NULL;
368 p->thread.esp = (unsigned long) childregs;
369 p->thread.esp0 = (unsigned long) (childregs+1);
372 * get the two stack pages, for the virtual stack.
374 * IMPORTANT: this code relies on the fact that the task
375 * structure is an THREAD_SIZE aligned piece of physical memory.
377 for (i = 0; i < ARRAY_SIZE(p->thread.stack_page); i++)
378 p->thread.stack_page[i] =
379 virt_to_page((unsigned long)p->thread_info + (i*PAGE_SIZE));
381 p->thread.eip = (unsigned long) ret_from_fork;
382 p->thread_info->real_stack = p->thread_info;
384 savesegment(fs,p->thread.fs);
385 savesegment(gs,p->thread.gs);
388 if (unlikely(NULL != tsk->thread.io_bitmap_ptr)) {
389 p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL);
390 if (!p->thread.io_bitmap_ptr)
392 memcpy(p->thread.io_bitmap_ptr, tsk->thread.io_bitmap_ptr,
397 * Set a new TLS for the child thread?
399 if (clone_flags & CLONE_SETTLS) {
400 struct desc_struct *desc;
401 struct user_desc info;
405 if (copy_from_user(&info, (void __user *)childregs->esi, sizeof(info)))
408 if (LDT_empty(&info))
411 idx = info.entry_number;
412 if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
415 desc = p->thread.tls_array + idx - GDT_ENTRY_TLS_MIN;
416 desc->a = LDT_entry_a(&info);
417 desc->b = LDT_entry_b(&info);
422 if (err && p->thread.io_bitmap_ptr)
423 kfree(p->thread.io_bitmap_ptr);
428 * fill in the user structure for a core dump..
430 void dump_thread(struct pt_regs * regs, struct user * dump)
434 /* changed the size calculations - should hopefully work better. lbt */
435 dump->magic = CMAGIC;
436 dump->start_code = 0;
437 dump->start_stack = regs->esp & ~(PAGE_SIZE - 1);
438 dump->u_tsize = ((unsigned long) current->mm->end_code) >> PAGE_SHIFT;
439 dump->u_dsize = ((unsigned long) (current->mm->brk + (PAGE_SIZE-1))) >> PAGE_SHIFT;
440 dump->u_dsize -= dump->u_tsize;
442 for (i = 0; i < 8; i++)
443 dump->u_debugreg[i] = current->thread.debugreg[i];
445 if (dump->start_stack < TASK_SIZE)
446 dump->u_ssize = ((unsigned long) (TASK_SIZE - dump->start_stack)) >> PAGE_SHIFT;
448 dump->regs.ebx = regs->ebx;
449 dump->regs.ecx = regs->ecx;
450 dump->regs.edx = regs->edx;
451 dump->regs.esi = regs->esi;
452 dump->regs.edi = regs->edi;
453 dump->regs.ebp = regs->ebp;
454 dump->regs.eax = regs->eax;
455 dump->regs.ds = regs->xds;
456 dump->regs.es = regs->xes;
457 savesegment(fs,dump->regs.fs);
458 savesegment(gs,dump->regs.gs);
459 dump->regs.orig_eax = regs->orig_eax;
460 dump->regs.eip = regs->eip;
461 dump->regs.cs = regs->xcs;
462 dump->regs.eflags = regs->eflags;
463 dump->regs.esp = regs->esp;
464 dump->regs.ss = regs->xss;
466 dump->u_fpvalid = dump_fpu (regs, &dump->i387);
470 * Capture the user space registers if the task is not running (in user space)
472 int dump_task_regs(struct task_struct *tsk, elf_gregset_t *regs)
474 struct pt_regs ptregs;
476 ptregs = *(struct pt_regs *)
477 ((unsigned long)tsk->thread_info+THREAD_SIZE - sizeof(ptregs));
478 ptregs.xcs &= 0xffff;
479 ptregs.xds &= 0xffff;
480 ptregs.xes &= 0xffff;
481 ptregs.xss &= 0xffff;
483 elf_core_copy_regs(regs, &ptregs);
489 * This special macro can be used to load a debugging register
491 #define loaddebug(thread,register) \
492 __asm__("movl %0,%%db" #register \
494 :"r" (thread->debugreg[register]))
497 * switch_to(x,yn) should switch tasks from x to y.
499 * We fsave/fwait so that an exception goes off at the right time
500 * (as a call from the fsave or fwait in effect) rather than to
501 * the wrong process. Lazy FP saving no longer makes any sense
502 * with modern CPU's, and this simplifies a lot of things (SMP
503 * and UP become the same).
505 * NOTE! We used to use the x86 hardware context switching. The
506 * reason for not using it any more becomes apparent when you
507 * try to recover gracefully from saved state that is no longer
508 * valid (stale segment register values in particular). With the
509 * hardware task-switch, there is no way to fix up bad state in
510 * a reasonable manner.
512 * The fact that Intel documents the hardware task-switching to
513 * be slow is a fairly red herring - this code is not noticeably
514 * faster. However, there _is_ some room for improvement here,
515 * so the performance issues may eventually be a valid point.
516 * More important, however, is the fact that this allows us much
519 * The return value (in %eax) will be the "prev" task after
520 * the task-switch, and shows up in ret_from_fork in entry.S,
523 struct task_struct fastcall * __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
525 struct thread_struct *prev = &prev_p->thread,
526 *next = &next_p->thread;
527 int cpu = smp_processor_id();
528 struct tss_struct *tss = init_tss + cpu;
530 /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
532 __unlazy_fpu(prev_p);
534 load_user_cs_desc(cpu, next_p->mm);
536 #ifdef CONFIG_X86_HIGH_ENTRY
540 * Set the ptes of the virtual stack. (NOTE: a one-page TLB flush is
541 * needed because otherwise NMIs could interrupt the
542 * user-return code with a virtual stack and stale TLBs.)
544 for (i = 0; i < ARRAY_SIZE(next->stack_page); i++) {
545 __kunmap_atomic_type(KM_VSTACK_TOP-i);
546 __kmap_atomic(next->stack_page[i], KM_VSTACK_TOP-i);
549 * NOTE: here we rely on the task being the stack as well
551 next_p->thread_info->virtual_stack =
552 (void *)__kmap_atomic_vaddr(KM_VSTACK_TOP);
554 #if defined(CONFIG_PREEMPT) && defined(CONFIG_SMP)
556 * If next was preempted on entry from userspace to kernel,
557 * and now it's on a different cpu, we need to adjust %esp.
558 * This assumes that entry.S does not copy %esp while on the
559 * virtual stack (with interrupts enabled): which is so,
560 * except within __SWITCH_KERNELSPACE itself.
562 if (unlikely(next->esp >= TASK_SIZE)) {
563 next->esp &= THREAD_SIZE - 1;
564 next->esp |= (unsigned long) next_p->thread_info->virtual_stack;
569 * Reload esp0, LDT and the page table pointer:
571 load_virtual_esp0(tss, next_p);
574 * Load the per-thread Thread-Local Storage descriptor.
579 * Save away %fs and %gs. No need to save %es and %ds, as
580 * those are always kernel segments while inside the kernel.
582 asm volatile("movl %%fs,%0":"=m" (*(int *)&prev->fs));
583 asm volatile("movl %%gs,%0":"=m" (*(int *)&prev->gs));
586 * Restore %fs and %gs if needed.
588 if (unlikely(prev->fs | prev->gs | next->fs | next->gs)) {
589 loadsegment(fs, next->fs);
590 loadsegment(gs, next->gs);
594 * Now maybe reload the debug registers
596 if (unlikely(next->debugreg[7])) {
606 if (unlikely(prev->io_bitmap_ptr || next->io_bitmap_ptr)) {
607 if (next->io_bitmap_ptr) {
609 * 4 cachelines copy ... not good, but not that
610 * bad either. Anyone got something better?
611 * This only affects processes which use ioperm().
612 * [Putting the TSSs into 4k-tlb mapped regions
613 * and playing VM tricks to switch the IO bitmap
614 * is not really acceptable.]
616 memcpy(tss->io_bitmap, next->io_bitmap_ptr,
618 tss->io_bitmap_base = IO_BITMAP_OFFSET;
621 * a bitmap offset pointing outside of the TSS limit
622 * causes a nicely controllable SIGSEGV if a process
623 * tries to use a port IO instruction. The first
624 * sys_ioperm() call sets up the bitmap properly.
626 tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
631 asmlinkage int sys_fork(struct pt_regs regs)
633 return do_fork(SIGCHLD, regs.esp, ®s, 0, NULL, NULL);
636 asmlinkage int sys_clone(struct pt_regs regs)
638 unsigned long clone_flags;
640 int __user *parent_tidptr, *child_tidptr;
642 clone_flags = regs.ebx;
644 parent_tidptr = (int __user *)regs.edx;
645 child_tidptr = (int __user *)regs.edi;
648 return do_fork(clone_flags & ~CLONE_IDLETASK, newsp, ®s, 0, parent_tidptr, child_tidptr);
652 * This is trivial, and on the face of it looks like it
653 * could equally well be done in user mode.
655 * Not so, for quite unobvious reasons - register pressure.
656 * In user mode vfork() cannot have a stack frame, and if
657 * done by calling the "clone()" system call directly, you
658 * do not have enough call-clobbered registers to hold all
659 * the information you need.
661 asmlinkage int sys_vfork(struct pt_regs regs)
663 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs.esp, ®s, 0, NULL, NULL);
667 * sys_execve() executes a new program.
669 asmlinkage int sys_execve(struct pt_regs regs)
674 filename = getname((char __user *) regs.ebx);
675 error = PTR_ERR(filename);
676 if (IS_ERR(filename))
678 error = do_execve(filename,
679 (char __user * __user *) regs.ecx,
680 (char __user * __user *) regs.edx,
683 current->ptrace &= ~PT_DTRACE;
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)))
832 unsigned long arch_align_stack(unsigned long sp)
834 if (current->flags & PF_RELOCEXEC)
835 sp -= ((get_random_int() % 65536) << 4);
840 void arch_add_exec_range(struct mm_struct *mm, unsigned long limit)
842 if (limit > mm->context.exec_limit) {
843 mm->context.exec_limit = limit;
844 set_user_cs(&mm->context.user_cs, limit);
845 if (mm == current->mm)
846 load_user_cs_desc(smp_processor_id(), mm);
850 void arch_remove_exec_range(struct mm_struct *mm, unsigned long old_end)
852 struct vm_area_struct *vma;
853 unsigned long limit = 0;
855 if (old_end == mm->context.exec_limit) {
856 for (vma = mm->mmap; vma; vma = vma->vm_next)
857 if ((vma->vm_flags & VM_EXEC) && (vma->vm_end > limit))
860 mm->context.exec_limit = limit;
861 set_user_cs(&mm->context.user_cs, limit);
862 if (mm == current->mm)
863 load_user_cs_desc(smp_processor_id(), mm);
867 void arch_flush_exec_range(struct mm_struct *mm)
869 mm->context.exec_limit = 0;
870 set_user_cs(&mm->context.user_cs, 0);
874 * Generate random brk address between 128MB and 196MB. (if the layout
877 void randomize_brk(unsigned long old_brk)
879 unsigned long new_brk, range_start, range_end;
881 range_start = 0x08000000;
882 if (current->mm->brk >= range_start)
883 range_start = current->mm->brk;
884 range_end = range_start + 0x02000000;
885 new_brk = randomize_range(range_start, range_end, 0);
887 current->mm->brk = new_brk;