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>
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");
61 * Return saved PC of a blocked thread.
63 unsigned long thread_saved_pc(struct task_struct *tsk)
65 return ((unsigned long *)tsk->thread.esp)[3];
69 * Powermanagement idle function, if any..
71 void (*pm_idle)(void);
73 void disable_hlt(void)
78 EXPORT_SYMBOL(disable_hlt);
85 EXPORT_SYMBOL(enable_hlt);
88 * We use this if we don't have any better
91 void default_idle(void)
93 if (!hlt_counter && current_cpu_data.hlt_works_ok) {
103 * On SMP it's slightly faster (but much more power-consuming!)
104 * to poll the ->work.need_resched flag instead of waiting for the
105 * cross-CPU IPI to arrive. Use this option with caution.
107 static void poll_idle (void)
114 * Deal with another CPU just having chosen a thread to
117 oldval = test_and_clear_thread_flag(TIF_NEED_RESCHED);
120 set_thread_flag(TIF_POLLING_NRFLAG);
126 : : "i"(_TIF_NEED_RESCHED), "m" (current_thread_info()->flags));
128 clear_thread_flag(TIF_POLLING_NRFLAG);
135 * The idle thread. There's no useful work to be
136 * done, so just try to conserve power and have a
137 * low exit latency (ie sit in a loop waiting for
138 * somebody to say that they'd like to reschedule)
142 /* endless idle loop with no priority at all */
144 while (!need_resched()) {
145 void (*idle)(void) = pm_idle;
150 irq_stat[smp_processor_id()].idle_timestamp = jiffies;
158 * This uses new MONITOR/MWAIT instructions on P4 processors with PNI,
159 * which can obviate IPI to trigger checking of need_resched.
160 * We execute MONITOR against need_resched and enter optimized wait state
161 * through MWAIT. Whenever someone changes need_resched, we would be woken
162 * up from MWAIT (without an IPI).
164 static void mwait_idle(void)
168 if (!need_resched()) {
169 set_thread_flag(TIF_POLLING_NRFLAG);
171 __monitor((void *)¤t_thread_info()->flags, 0, 0);
175 } while (!need_resched());
176 clear_thread_flag(TIF_POLLING_NRFLAG);
180 void __init select_idle_routine(const struct cpuinfo_x86 *c)
182 if (cpu_has(c, X86_FEATURE_MWAIT)) {
183 printk("monitor/mwait feature present.\n");
185 * Skip, if setup has overridden idle.
186 * Also, take care of system with asymmetric CPUs.
187 * Use, mwait_idle only if all cpus support it.
188 * If not, we fallback to default_idle()
191 printk("using mwait in idle threads.\n");
192 pm_idle = mwait_idle;
196 pm_idle = default_idle;
200 static int __init idle_setup (char *str)
202 if (!strncmp(str, "poll", 4)) {
203 printk("using polling idle threads.\n");
205 } else if (!strncmp(str, "halt", 4)) {
206 printk("using halt in idle threads.\n");
207 pm_idle = default_idle;
213 __setup("idle=", idle_setup);
215 void show_regs(struct pt_regs * regs)
217 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L;
220 printk("Pid: %d, comm: %20s\n", current->pid, current->comm);
221 printk("EIP: %04x:[<%08lx>] CPU: %d\n",0xffff & regs->xcs,regs->eip, smp_processor_id());
222 print_symbol("EIP is at %s\n", regs->eip);
225 printk(" ESP: %04x:%08lx",0xffff & regs->xss,regs->esp);
226 printk(" EFLAGS: %08lx %s (%s)\n",regs->eflags, print_tainted(),UTS_RELEASE);
227 printk("EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
228 regs->eax,regs->ebx,regs->ecx,regs->edx);
229 printk("ESI: %08lx EDI: %08lx EBP: %08lx",
230 regs->esi, regs->edi, regs->ebp);
231 printk(" DS: %04x ES: %04x\n",
232 0xffff & regs->xds,0xffff & regs->xes);
234 __asm__("movl %%cr0, %0": "=r" (cr0));
235 __asm__("movl %%cr2, %0": "=r" (cr2));
236 __asm__("movl %%cr3, %0": "=r" (cr3));
237 /* This could fault if %cr4 does not exist */
238 __asm__("1: movl %%cr4, %0 \n"
240 ".section __ex_table,\"a\" \n"
243 : "=r" (cr4): "0" (0));
244 printk("CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n", cr0, cr2, cr3, cr4);
245 show_trace(NULL, ®s->esp);
249 * This gets run with %ebx containing the
250 * function to call, and %edx containing
253 extern void kernel_thread_helper(void);
254 __asm__(".section .text\n"
256 "kernel_thread_helper:\n\t"
265 * Create a kernel thread
267 int kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
271 memset(®s, 0, sizeof(regs));
273 regs.ebx = (unsigned long) fn;
274 regs.edx = (unsigned long) arg;
276 regs.xds = __USER_DS;
277 regs.xes = __USER_DS;
279 regs.eip = (unsigned long) kernel_thread_helper;
280 regs.xcs = __KERNEL_CS;
281 regs.eflags = X86_EFLAGS_IF | X86_EFLAGS_SF | X86_EFLAGS_PF | 0x2;
283 /* Ok, create the new process.. */
284 return do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0, ®s, 0, NULL, NULL);
288 * Free current thread data structures etc..
290 void exit_thread(void)
292 struct task_struct *tsk = current;
294 /* The process may have allocated an io port bitmap... nuke it. */
295 if (unlikely(NULL != tsk->thread.io_bitmap_ptr)) {
297 struct tss_struct *tss = init_tss + cpu;
298 kfree(tsk->thread.io_bitmap_ptr);
299 tsk->thread.io_bitmap_ptr = NULL;
300 tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
305 void flush_thread(void)
307 struct task_struct *tsk = current;
309 memset(tsk->thread.debugreg, 0, sizeof(unsigned long)*8);
310 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
312 * Forget coprocessor state..
318 void release_thread(struct task_struct *dead_task)
321 // temporary debugging check
322 if (dead_task->mm->context.size) {
323 printk("WARNING: dead process %8s still has LDT? <%p/%d>\n",
325 dead_task->mm->context.ldt,
326 dead_task->mm->context.size);
331 release_x86_irqs(dead_task);
335 * This gets called before we allocate a new thread and copy
336 * the current task into it.
338 void prepare_to_copy(struct task_struct *tsk)
343 int copy_thread(int nr, unsigned long clone_flags, unsigned long esp,
344 unsigned long unused,
345 struct task_struct * p, struct pt_regs * regs)
347 struct pt_regs * childregs;
348 struct task_struct *tsk;
351 childregs = ((struct pt_regs *) (THREAD_SIZE + (unsigned long) p->thread_info)) - 1;
352 struct_cpy(childregs, regs);
354 childregs->esp = esp;
355 p->set_child_tid = p->clear_child_tid = NULL;
357 p->thread.esp = (unsigned long) childregs;
358 p->thread.esp0 = (unsigned long) (childregs+1);
360 p->thread.eip = (unsigned long) ret_from_fork;
362 savesegment(fs,p->thread.fs);
363 savesegment(gs,p->thread.gs);
366 if (unlikely(NULL != tsk->thread.io_bitmap_ptr)) {
367 p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL);
368 if (!p->thread.io_bitmap_ptr)
370 memcpy(p->thread.io_bitmap_ptr, tsk->thread.io_bitmap_ptr,
375 * Set a new TLS for the child thread?
377 if (clone_flags & CLONE_SETTLS) {
378 struct desc_struct *desc;
379 struct user_desc info;
383 if (copy_from_user(&info, (void __user *)childregs->esi, sizeof(info)))
386 if (LDT_empty(&info))
389 idx = info.entry_number;
390 if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
393 desc = p->thread.tls_array + idx - GDT_ENTRY_TLS_MIN;
394 desc->a = LDT_entry_a(&info);
395 desc->b = LDT_entry_b(&info);
400 if (err && p->thread.io_bitmap_ptr)
401 kfree(p->thread.io_bitmap_ptr);
406 * fill in the user structure for a core dump..
408 void dump_thread(struct pt_regs * regs, struct user * dump)
412 /* changed the size calculations - should hopefully work better. lbt */
413 dump->magic = CMAGIC;
414 dump->start_code = 0;
415 dump->start_stack = regs->esp & ~(PAGE_SIZE - 1);
416 dump->u_tsize = ((unsigned long) current->mm->end_code) >> PAGE_SHIFT;
417 dump->u_dsize = ((unsigned long) (current->mm->brk + (PAGE_SIZE-1))) >> PAGE_SHIFT;
418 dump->u_dsize -= dump->u_tsize;
420 for (i = 0; i < 8; i++)
421 dump->u_debugreg[i] = current->thread.debugreg[i];
423 if (dump->start_stack < TASK_SIZE)
424 dump->u_ssize = ((unsigned long) (TASK_SIZE - dump->start_stack)) >> PAGE_SHIFT;
426 dump->regs.ebx = regs->ebx;
427 dump->regs.ecx = regs->ecx;
428 dump->regs.edx = regs->edx;
429 dump->regs.esi = regs->esi;
430 dump->regs.edi = regs->edi;
431 dump->regs.ebp = regs->ebp;
432 dump->regs.eax = regs->eax;
433 dump->regs.ds = regs->xds;
434 dump->regs.es = regs->xes;
435 savesegment(fs,dump->regs.fs);
436 savesegment(gs,dump->regs.gs);
437 dump->regs.orig_eax = regs->orig_eax;
438 dump->regs.eip = regs->eip;
439 dump->regs.cs = regs->xcs;
440 dump->regs.eflags = regs->eflags;
441 dump->regs.esp = regs->esp;
442 dump->regs.ss = regs->xss;
444 dump->u_fpvalid = dump_fpu (regs, &dump->i387);
448 * Capture the user space registers if the task is not running (in user space)
450 int dump_task_regs(struct task_struct *tsk, elf_gregset_t *regs)
452 struct pt_regs ptregs;
454 ptregs = *(struct pt_regs *)
455 ((unsigned long)tsk->thread_info+THREAD_SIZE - sizeof(ptregs));
456 ptregs.xcs &= 0xffff;
457 ptregs.xds &= 0xffff;
458 ptregs.xes &= 0xffff;
459 ptregs.xss &= 0xffff;
461 elf_core_copy_regs(regs, &ptregs);
467 * This special macro can be used to load a debugging register
469 #define loaddebug(thread,register) \
470 __asm__("movl %0,%%db" #register \
472 :"r" (thread->debugreg[register]))
475 * switch_to(x,yn) should switch tasks from x to y.
477 * We fsave/fwait so that an exception goes off at the right time
478 * (as a call from the fsave or fwait in effect) rather than to
479 * the wrong process. Lazy FP saving no longer makes any sense
480 * with modern CPU's, and this simplifies a lot of things (SMP
481 * and UP become the same).
483 * NOTE! We used to use the x86 hardware context switching. The
484 * reason for not using it any more becomes apparent when you
485 * try to recover gracefully from saved state that is no longer
486 * valid (stale segment register values in particular). With the
487 * hardware task-switch, there is no way to fix up bad state in
488 * a reasonable manner.
490 * The fact that Intel documents the hardware task-switching to
491 * be slow is a fairly red herring - this code is not noticeably
492 * faster. However, there _is_ some room for improvement here,
493 * so the performance issues may eventually be a valid point.
494 * More important, however, is the fact that this allows us much
497 * The return value (in %eax) will be the "prev" task after
498 * the task-switch, and shows up in ret_from_fork in entry.S,
501 struct task_struct fastcall * __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
503 struct thread_struct *prev = &prev_p->thread,
504 *next = &next_p->thread;
505 int cpu = smp_processor_id();
506 struct tss_struct *tss = init_tss + cpu;
508 /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
510 __unlazy_fpu(prev_p);
513 * Reload esp0, LDT and the page table pointer:
515 load_esp0(tss, next);
518 * Load the per-thread Thread-Local Storage descriptor.
523 * Save away %fs and %gs. No need to save %es and %ds, as
524 * those are always kernel segments while inside the kernel.
526 asm volatile("movl %%fs,%0":"=m" (*(int *)&prev->fs));
527 asm volatile("movl %%gs,%0":"=m" (*(int *)&prev->gs));
530 * Restore %fs and %gs if needed.
532 if (unlikely(prev->fs | prev->gs | next->fs | next->gs)) {
533 loadsegment(fs, next->fs);
534 loadsegment(gs, next->gs);
538 * Now maybe reload the debug registers
540 if (unlikely(next->debugreg[7])) {
550 if (unlikely(prev->io_bitmap_ptr || next->io_bitmap_ptr)) {
551 if (next->io_bitmap_ptr) {
553 * 4 cachelines copy ... not good, but not that
554 * bad either. Anyone got something better?
555 * This only affects processes which use ioperm().
556 * [Putting the TSSs into 4k-tlb mapped regions
557 * and playing VM tricks to switch the IO bitmap
558 * is not really acceptable.]
560 memcpy(tss->io_bitmap, next->io_bitmap_ptr,
562 tss->io_bitmap_base = IO_BITMAP_OFFSET;
565 * a bitmap offset pointing outside of the TSS limit
566 * causes a nicely controllable SIGSEGV if a process
567 * tries to use a port IO instruction. The first
568 * sys_ioperm() call sets up the bitmap properly.
570 tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
575 asmlinkage int sys_fork(struct pt_regs regs)
577 return do_fork(SIGCHLD, regs.esp, ®s, 0, NULL, NULL);
580 asmlinkage int sys_clone(struct pt_regs regs)
582 unsigned long clone_flags;
584 int __user *parent_tidptr, *child_tidptr;
586 clone_flags = regs.ebx;
588 parent_tidptr = (int __user *)regs.edx;
589 child_tidptr = (int __user *)regs.edi;
592 return do_fork(clone_flags & ~CLONE_IDLETASK, newsp, ®s, 0, parent_tidptr, child_tidptr);
596 * This is trivial, and on the face of it looks like it
597 * could equally well be done in user mode.
599 * Not so, for quite unobvious reasons - register pressure.
600 * In user mode vfork() cannot have a stack frame, and if
601 * done by calling the "clone()" system call directly, you
602 * do not have enough call-clobbered registers to hold all
603 * the information you need.
605 asmlinkage int sys_vfork(struct pt_regs regs)
607 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs.esp, ®s, 0, NULL, NULL);
611 * sys_execve() executes a new program.
613 asmlinkage int sys_execve(struct pt_regs regs)
618 filename = getname((char __user *) regs.ebx);
619 error = PTR_ERR(filename);
620 if (IS_ERR(filename))
622 error = do_execve(filename,
623 (char __user * __user *) regs.ecx,
624 (char __user * __user *) regs.edx,
627 current->ptrace &= ~PT_DTRACE;
628 /* Make sure we don't return using sysenter.. */
629 set_thread_flag(TIF_IRET);
637 * These bracket the sleeping functions..
639 #define first_sched ((unsigned long) scheduling_functions_start_here)
640 #define last_sched ((unsigned long) scheduling_functions_end_here)
641 #define top_esp (THREAD_SIZE - sizeof(unsigned long))
642 #define top_ebp (THREAD_SIZE - 2*sizeof(unsigned long))
644 unsigned long get_wchan(struct task_struct *p)
646 unsigned long ebp, esp, eip;
647 unsigned long stack_page;
649 if (!p || p == current || p->state == TASK_RUNNING)
651 stack_page = (unsigned long)p->thread_info;
653 if (!stack_page || esp < stack_page || esp > top_esp+stack_page)
655 /* include/asm-i386/system.h:switch_to() pushes ebp last. */
656 ebp = *(unsigned long *) esp;
658 if (ebp < stack_page || ebp > top_ebp+stack_page)
660 eip = *(unsigned long *) (ebp+4);
661 if (eip < first_sched || eip >= last_sched)
663 ebp = *(unsigned long *) ebp;
664 } while (count++ < 16);
671 * sys_alloc_thread_area: get a yet unused TLS descriptor index.
673 static int get_free_idx(void)
675 struct thread_struct *t = ¤t->thread;
678 for (idx = 0; idx < GDT_ENTRY_TLS_ENTRIES; idx++)
679 if (desc_empty(t->tls_array + idx))
680 return idx + GDT_ENTRY_TLS_MIN;
685 * Set a given TLS descriptor:
687 asmlinkage int sys_set_thread_area(struct user_desc __user *u_info)
689 struct thread_struct *t = ¤t->thread;
690 struct user_desc info;
691 struct desc_struct *desc;
694 if (copy_from_user(&info, u_info, sizeof(info)))
696 idx = info.entry_number;
699 * index -1 means the kernel should try to find and
700 * allocate an empty descriptor:
703 idx = get_free_idx();
706 if (put_user(idx, &u_info->entry_number))
710 if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
713 desc = t->tls_array + idx - GDT_ENTRY_TLS_MIN;
716 * We must not get preempted while modifying the TLS.
720 if (LDT_empty(&info)) {
724 desc->a = LDT_entry_a(&info);
725 desc->b = LDT_entry_b(&info);
735 * Get the current Thread-Local Storage area:
738 #define GET_BASE(desc) ( \
739 (((desc)->a >> 16) & 0x0000ffff) | \
740 (((desc)->b << 16) & 0x00ff0000) | \
741 ( (desc)->b & 0xff000000) )
743 #define GET_LIMIT(desc) ( \
744 ((desc)->a & 0x0ffff) | \
745 ((desc)->b & 0xf0000) )
747 #define GET_32BIT(desc) (((desc)->b >> 23) & 1)
748 #define GET_CONTENTS(desc) (((desc)->b >> 10) & 3)
749 #define GET_WRITABLE(desc) (((desc)->b >> 9) & 1)
750 #define GET_LIMIT_PAGES(desc) (((desc)->b >> 23) & 1)
751 #define GET_PRESENT(desc) (((desc)->b >> 15) & 1)
752 #define GET_USEABLE(desc) (((desc)->b >> 20) & 1)
754 asmlinkage int sys_get_thread_area(struct user_desc __user *u_info)
756 struct user_desc info;
757 struct desc_struct *desc;
760 if (get_user(idx, &u_info->entry_number))
762 if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
765 desc = current->thread.tls_array + idx - GDT_ENTRY_TLS_MIN;
767 info.entry_number = idx;
768 info.base_addr = GET_BASE(desc);
769 info.limit = GET_LIMIT(desc);
770 info.seg_32bit = GET_32BIT(desc);
771 info.contents = GET_CONTENTS(desc);
772 info.read_exec_only = !GET_WRITABLE(desc);
773 info.limit_in_pages = GET_LIMIT_PAGES(desc);
774 info.seg_not_present = !GET_PRESENT(desc);
775 info.useable = GET_USEABLE(desc);
777 if (copy_to_user(u_info, &info, sizeof(info)))