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 #ifdef CONFIG_X86_SMP
206 if (smp_num_siblings > 1)
207 printk("WARNING: polling idle and HT enabled, performance may degrade.\n");
209 } else if (!strncmp(str, "halt", 4)) {
210 printk("using halt in idle threads.\n");
211 pm_idle = default_idle;
217 __setup("idle=", idle_setup);
219 void show_regs(struct pt_regs * regs)
221 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L;
224 printk("Pid: %d, comm: %20s\n", current->pid, current->comm);
225 printk("EIP: %04x:[<%08lx>] CPU: %d\n",0xffff & regs->xcs,regs->eip, smp_processor_id());
226 print_symbol("EIP is at %s\n", regs->eip);
229 printk(" ESP: %04x:%08lx",0xffff & regs->xss,regs->esp);
230 printk(" EFLAGS: %08lx %s (%s)\n",regs->eflags, print_tainted(),UTS_RELEASE);
231 printk("EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
232 regs->eax,regs->ebx,regs->ecx,regs->edx);
233 printk("ESI: %08lx EDI: %08lx EBP: %08lx",
234 regs->esi, regs->edi, regs->ebp);
235 printk(" DS: %04x ES: %04x\n",
236 0xffff & regs->xds,0xffff & regs->xes);
238 __asm__("movl %%cr0, %0": "=r" (cr0));
239 __asm__("movl %%cr2, %0": "=r" (cr2));
240 __asm__("movl %%cr3, %0": "=r" (cr3));
241 /* This could fault if %cr4 does not exist */
242 __asm__("1: movl %%cr4, %0 \n"
244 ".section __ex_table,\"a\" \n"
247 : "=r" (cr4): "0" (0));
248 printk("CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n", cr0, cr2, cr3, cr4);
249 show_trace(NULL, ®s->esp);
253 * This gets run with %ebx containing the
254 * function to call, and %edx containing
257 extern void kernel_thread_helper(void);
258 __asm__(".section .text\n"
260 "kernel_thread_helper:\n\t"
269 * Create a kernel thread
271 int kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
275 memset(®s, 0, sizeof(regs));
277 regs.ebx = (unsigned long) fn;
278 regs.edx = (unsigned long) arg;
280 regs.xds = __USER_DS;
281 regs.xes = __USER_DS;
283 regs.eip = (unsigned long) kernel_thread_helper;
284 regs.xcs = __KERNEL_CS;
285 regs.eflags = X86_EFLAGS_IF | X86_EFLAGS_SF | X86_EFLAGS_PF | 0x2;
287 /* Ok, create the new process.. */
288 return do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0, ®s, 0, NULL, NULL);
292 * Free current thread data structures etc..
294 void exit_thread(void)
296 struct task_struct *tsk = current;
298 /* The process may have allocated an io port bitmap... nuke it. */
299 if (unlikely(NULL != tsk->thread.io_bitmap_ptr)) {
301 struct tss_struct *tss = init_tss + cpu;
302 kfree(tsk->thread.io_bitmap_ptr);
303 tsk->thread.io_bitmap_ptr = NULL;
304 tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
309 void flush_thread(void)
311 struct task_struct *tsk = current;
313 memset(tsk->thread.debugreg, 0, sizeof(unsigned long)*8);
314 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
316 * Forget coprocessor state..
322 void release_thread(struct task_struct *dead_task)
325 // temporary debugging check
326 if (dead_task->mm->context.size) {
327 printk("WARNING: dead process %8s still has LDT? <%p/%d>\n",
329 dead_task->mm->context.ldt,
330 dead_task->mm->context.size);
335 release_x86_irqs(dead_task);
339 * This gets called before we allocate a new thread and copy
340 * the current task into it.
342 void prepare_to_copy(struct task_struct *tsk)
347 int copy_thread(int nr, unsigned long clone_flags, unsigned long esp,
348 unsigned long unused,
349 struct task_struct * p, struct pt_regs * regs)
351 struct pt_regs * childregs;
352 struct task_struct *tsk;
355 childregs = ((struct pt_regs *) (THREAD_SIZE + (unsigned long) p->thread_info)) - 1;
358 childregs->esp = esp;
359 p->set_child_tid = p->clear_child_tid = NULL;
361 p->thread.esp = (unsigned long) childregs;
362 p->thread.esp0 = (unsigned long) (childregs+1);
364 p->thread.eip = (unsigned long) ret_from_fork;
366 savesegment(fs,p->thread.fs);
367 savesegment(gs,p->thread.gs);
370 if (unlikely(NULL != tsk->thread.io_bitmap_ptr)) {
371 p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL);
372 if (!p->thread.io_bitmap_ptr)
374 memcpy(p->thread.io_bitmap_ptr, tsk->thread.io_bitmap_ptr,
379 * Set a new TLS for the child thread?
381 if (clone_flags & CLONE_SETTLS) {
382 struct desc_struct *desc;
383 struct user_desc info;
387 if (copy_from_user(&info, (void __user *)childregs->esi, sizeof(info)))
390 if (LDT_empty(&info))
393 idx = info.entry_number;
394 if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
397 desc = p->thread.tls_array + idx - GDT_ENTRY_TLS_MIN;
398 desc->a = LDT_entry_a(&info);
399 desc->b = LDT_entry_b(&info);
404 if (err && p->thread.io_bitmap_ptr)
405 kfree(p->thread.io_bitmap_ptr);
410 * fill in the user structure for a core dump..
412 void dump_thread(struct pt_regs * regs, struct user * dump)
416 /* changed the size calculations - should hopefully work better. lbt */
417 dump->magic = CMAGIC;
418 dump->start_code = 0;
419 dump->start_stack = regs->esp & ~(PAGE_SIZE - 1);
420 dump->u_tsize = ((unsigned long) current->mm->end_code) >> PAGE_SHIFT;
421 dump->u_dsize = ((unsigned long) (current->mm->brk + (PAGE_SIZE-1))) >> PAGE_SHIFT;
422 dump->u_dsize -= dump->u_tsize;
424 for (i = 0; i < 8; i++)
425 dump->u_debugreg[i] = current->thread.debugreg[i];
427 if (dump->start_stack < TASK_SIZE)
428 dump->u_ssize = ((unsigned long) (TASK_SIZE - dump->start_stack)) >> PAGE_SHIFT;
430 dump->regs.ebx = regs->ebx;
431 dump->regs.ecx = regs->ecx;
432 dump->regs.edx = regs->edx;
433 dump->regs.esi = regs->esi;
434 dump->regs.edi = regs->edi;
435 dump->regs.ebp = regs->ebp;
436 dump->regs.eax = regs->eax;
437 dump->regs.ds = regs->xds;
438 dump->regs.es = regs->xes;
439 savesegment(fs,dump->regs.fs);
440 savesegment(gs,dump->regs.gs);
441 dump->regs.orig_eax = regs->orig_eax;
442 dump->regs.eip = regs->eip;
443 dump->regs.cs = regs->xcs;
444 dump->regs.eflags = regs->eflags;
445 dump->regs.esp = regs->esp;
446 dump->regs.ss = regs->xss;
448 dump->u_fpvalid = dump_fpu (regs, &dump->i387);
452 * Capture the user space registers if the task is not running (in user space)
454 int dump_task_regs(struct task_struct *tsk, elf_gregset_t *regs)
456 struct pt_regs ptregs;
458 ptregs = *(struct pt_regs *)
459 ((unsigned long)tsk->thread_info+THREAD_SIZE - sizeof(ptregs));
460 ptregs.xcs &= 0xffff;
461 ptregs.xds &= 0xffff;
462 ptregs.xes &= 0xffff;
463 ptregs.xss &= 0xffff;
465 elf_core_copy_regs(regs, &ptregs);
471 * This special macro can be used to load a debugging register
473 #define loaddebug(thread,register) \
474 __asm__("movl %0,%%db" #register \
476 :"r" (thread->debugreg[register]))
479 * switch_to(x,yn) should switch tasks from x to y.
481 * We fsave/fwait so that an exception goes off at the right time
482 * (as a call from the fsave or fwait in effect) rather than to
483 * the wrong process. Lazy FP saving no longer makes any sense
484 * with modern CPU's, and this simplifies a lot of things (SMP
485 * and UP become the same).
487 * NOTE! We used to use the x86 hardware context switching. The
488 * reason for not using it any more becomes apparent when you
489 * try to recover gracefully from saved state that is no longer
490 * valid (stale segment register values in particular). With the
491 * hardware task-switch, there is no way to fix up bad state in
492 * a reasonable manner.
494 * The fact that Intel documents the hardware task-switching to
495 * be slow is a fairly red herring - this code is not noticeably
496 * faster. However, there _is_ some room for improvement here,
497 * so the performance issues may eventually be a valid point.
498 * More important, however, is the fact that this allows us much
501 * The return value (in %eax) will be the "prev" task after
502 * the task-switch, and shows up in ret_from_fork in entry.S,
505 struct task_struct fastcall * __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
507 struct thread_struct *prev = &prev_p->thread,
508 *next = &next_p->thread;
509 int cpu = smp_processor_id();
510 struct tss_struct *tss = init_tss + cpu;
512 /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
514 __unlazy_fpu(prev_p);
517 * Reload esp0, LDT and the page table pointer:
519 load_esp0(tss, next);
522 * Load the per-thread Thread-Local Storage descriptor.
527 * Save away %fs and %gs. No need to save %es and %ds, as
528 * those are always kernel segments while inside the kernel.
530 asm volatile("movl %%fs,%0":"=m" (*(int *)&prev->fs));
531 asm volatile("movl %%gs,%0":"=m" (*(int *)&prev->gs));
534 * Restore %fs and %gs if needed.
536 if (unlikely(prev->fs | prev->gs | next->fs | next->gs)) {
537 loadsegment(fs, next->fs);
538 loadsegment(gs, next->gs);
542 * Now maybe reload the debug registers
544 if (unlikely(next->debugreg[7])) {
554 if (unlikely(prev->io_bitmap_ptr || next->io_bitmap_ptr)) {
555 if (next->io_bitmap_ptr) {
557 * 4 cachelines copy ... not good, but not that
558 * bad either. Anyone got something better?
559 * This only affects processes which use ioperm().
560 * [Putting the TSSs into 4k-tlb mapped regions
561 * and playing VM tricks to switch the IO bitmap
562 * is not really acceptable.]
564 memcpy(tss->io_bitmap, next->io_bitmap_ptr,
566 tss->io_bitmap_base = IO_BITMAP_OFFSET;
569 * a bitmap offset pointing outside of the TSS limit
570 * causes a nicely controllable SIGSEGV if a process
571 * tries to use a port IO instruction. The first
572 * sys_ioperm() call sets up the bitmap properly.
574 tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
579 asmlinkage int sys_fork(struct pt_regs regs)
581 return do_fork(SIGCHLD, regs.esp, ®s, 0, NULL, NULL);
584 asmlinkage int sys_clone(struct pt_regs regs)
586 unsigned long clone_flags;
588 int __user *parent_tidptr, *child_tidptr;
590 clone_flags = regs.ebx;
592 parent_tidptr = (int __user *)regs.edx;
593 child_tidptr = (int __user *)regs.edi;
596 return do_fork(clone_flags & ~CLONE_IDLETASK, newsp, ®s, 0, parent_tidptr, child_tidptr);
600 * This is trivial, and on the face of it looks like it
601 * could equally well be done in user mode.
603 * Not so, for quite unobvious reasons - register pressure.
604 * In user mode vfork() cannot have a stack frame, and if
605 * done by calling the "clone()" system call directly, you
606 * do not have enough call-clobbered registers to hold all
607 * the information you need.
609 asmlinkage int sys_vfork(struct pt_regs regs)
611 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs.esp, ®s, 0, NULL, NULL);
615 * sys_execve() executes a new program.
617 asmlinkage int sys_execve(struct pt_regs regs)
622 filename = getname((char __user *) regs.ebx);
623 error = PTR_ERR(filename);
624 if (IS_ERR(filename))
626 error = do_execve(filename,
627 (char __user * __user *) regs.ecx,
628 (char __user * __user *) regs.edx,
631 current->ptrace &= ~PT_DTRACE;
632 /* Make sure we don't return using sysenter.. */
633 set_thread_flag(TIF_IRET);
640 #define top_esp (THREAD_SIZE - sizeof(unsigned long))
641 #define top_ebp (THREAD_SIZE - 2*sizeof(unsigned long))
643 unsigned long get_wchan(struct task_struct *p)
645 unsigned long ebp, esp, eip;
646 unsigned long stack_page;
648 if (!p || p == current || p->state == TASK_RUNNING)
650 stack_page = (unsigned long)p->thread_info;
652 if (!stack_page || esp < stack_page || esp > top_esp+stack_page)
654 /* include/asm-i386/system.h:switch_to() pushes ebp last. */
655 ebp = *(unsigned long *) esp;
657 if (ebp < stack_page || ebp > top_ebp+stack_page)
659 eip = *(unsigned long *) (ebp+4);
660 if (!in_sched_functions(eip))
662 ebp = *(unsigned long *) ebp;
663 } while (count++ < 16);
668 * sys_alloc_thread_area: get a yet unused TLS descriptor index.
670 static int get_free_idx(void)
672 struct thread_struct *t = ¤t->thread;
675 for (idx = 0; idx < GDT_ENTRY_TLS_ENTRIES; idx++)
676 if (desc_empty(t->tls_array + idx))
677 return idx + GDT_ENTRY_TLS_MIN;
682 * Set a given TLS descriptor:
684 asmlinkage int sys_set_thread_area(struct user_desc __user *u_info)
686 struct thread_struct *t = ¤t->thread;
687 struct user_desc info;
688 struct desc_struct *desc;
691 if (copy_from_user(&info, u_info, sizeof(info)))
693 idx = info.entry_number;
696 * index -1 means the kernel should try to find and
697 * allocate an empty descriptor:
700 idx = get_free_idx();
703 if (put_user(idx, &u_info->entry_number))
707 if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
710 desc = t->tls_array + idx - GDT_ENTRY_TLS_MIN;
713 * We must not get preempted while modifying the TLS.
717 if (LDT_empty(&info)) {
721 desc->a = LDT_entry_a(&info);
722 desc->b = LDT_entry_b(&info);
732 * Get the current Thread-Local Storage area:
735 #define GET_BASE(desc) ( \
736 (((desc)->a >> 16) & 0x0000ffff) | \
737 (((desc)->b << 16) & 0x00ff0000) | \
738 ( (desc)->b & 0xff000000) )
740 #define GET_LIMIT(desc) ( \
741 ((desc)->a & 0x0ffff) | \
742 ((desc)->b & 0xf0000) )
744 #define GET_32BIT(desc) (((desc)->b >> 22) & 1)
745 #define GET_CONTENTS(desc) (((desc)->b >> 10) & 3)
746 #define GET_WRITABLE(desc) (((desc)->b >> 9) & 1)
747 #define GET_LIMIT_PAGES(desc) (((desc)->b >> 23) & 1)
748 #define GET_PRESENT(desc) (((desc)->b >> 15) & 1)
749 #define GET_USEABLE(desc) (((desc)->b >> 20) & 1)
751 asmlinkage int sys_get_thread_area(struct user_desc __user *u_info)
753 struct user_desc info;
754 struct desc_struct *desc;
757 if (get_user(idx, &u_info->entry_number))
759 if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
762 desc = current->thread.tls_array + idx - GDT_ENTRY_TLS_MIN;
764 info.entry_number = idx;
765 info.base_addr = GET_BASE(desc);
766 info.limit = GET_LIMIT(desc);
767 info.seg_32bit = GET_32BIT(desc);
768 info.contents = GET_CONTENTS(desc);
769 info.read_exec_only = !GET_WRITABLE(desc);
770 info.limit_in_pages = GET_LIMIT_PAGES(desc);
771 info.seg_not_present = !GET_PRESENT(desc);
772 info.useable = GET_USEABLE(desc);
774 if (copy_to_user(u_info, &info, sizeof(info)))