2 * Copyright (C) 1995 Linus Torvalds
4 * Pentium III FXSR, SSE support
5 * Gareth Hughes <gareth@valinux.com>, May 2000
9 * This file handles the architecture-dependent parts of process handling..
14 #include <linux/cpu.h>
15 #include <linux/errno.h>
16 #include <linux/sched.h>
18 #include <linux/kernel.h>
20 #include <linux/elfcore.h>
21 #include <linux/smp.h>
22 #include <linux/stddef.h>
23 #include <linux/slab.h>
24 #include <linux/vmalloc.h>
25 #include <linux/user.h>
26 #include <linux/interrupt.h>
27 #include <linux/utsname.h>
28 #include <linux/delay.h>
29 #include <linux/reboot.h>
30 #include <linux/init.h>
31 #include <linux/mc146818rtc.h>
32 #include <linux/module.h>
33 #include <linux/kallsyms.h>
34 #include <linux/ptrace.h>
35 #include <linux/random.h>
36 #include <linux/personality.h>
37 #include <linux/tick.h>
38 #include <linux/percpu.h>
39 #include <linux/prctl.h>
41 #include <asm/uaccess.h>
42 #include <asm/pgtable.h>
43 #include <asm/system.h>
46 #include <asm/processor.h>
49 #ifdef CONFIG_MATH_EMULATION
50 #include <asm/math_emu.h>
53 #include <linux/err.h>
55 #include <asm/tlbflush.h>
57 #include <asm/kdebug.h>
58 #include <asm/syscalls.h>
61 asmlinkage void ret_from_fork(void) __asm__("ret_from_fork");
63 DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task;
64 EXPORT_PER_CPU_SYMBOL(current_task);
66 DEFINE_PER_CPU(int, cpu_number);
67 EXPORT_PER_CPU_SYMBOL(cpu_number);
70 * Return saved PC of a blocked thread.
72 unsigned long thread_saved_pc(struct task_struct *tsk)
74 return ((unsigned long *)tsk->thread.sp)[3];
77 #ifdef CONFIG_HOTPLUG_CPU
80 static void cpu_exit_clear(void)
82 int cpu = raw_smp_processor_id();
89 cpu_clear(cpu, cpu_callout_map);
90 cpu_clear(cpu, cpu_callin_map);
95 /* We don't actually take CPU down, just spin without interrupts. */
96 static inline void play_dead(void)
98 /* This must be done before dead CPU ack */
103 __get_cpu_var(cpu_state) = CPU_DEAD;
106 * With physical CPU hotplug, we should halt the cpu
113 static inline void play_dead(void)
117 #endif /* CONFIG_HOTPLUG_CPU */
120 * The idle thread. There's no useful work to be
121 * done, so just try to conserve power and have a
122 * low exit latency (ie sit in a loop waiting for
123 * somebody to say that they'd like to reschedule)
127 int cpu = smp_processor_id();
129 current_thread_info()->status |= TS_POLLING;
131 /* endless idle loop with no priority at all */
133 tick_nohz_stop_sched_tick(1);
134 while (!need_resched()) {
139 if (rcu_pending(cpu))
140 rcu_check_callbacks(cpu, 0);
142 if (cpu_is_offline(cpu))
146 __get_cpu_var(irq_stat).idle_timestamp = jiffies;
147 /* Don't trace irqs off for idle */
148 stop_critical_timings();
150 start_critical_timings();
152 tick_nohz_restart_sched_tick();
153 preempt_enable_no_resched();
159 void __show_registers(struct pt_regs *regs, int all)
161 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L;
162 unsigned long d0, d1, d2, d3, d6, d7;
164 unsigned short ss, gs;
166 if (user_mode_vm(regs)) {
168 ss = regs->ss & 0xffff;
171 sp = (unsigned long) (®s->sp);
177 printk("Pid: %d, comm: %s %s (%s %.*s)\n",
178 task_pid_nr(current), current->comm,
179 print_tainted(), init_utsname()->release,
180 (int)strcspn(init_utsname()->version, " "),
181 init_utsname()->version);
183 printk("EIP: %04x:[<%08lx>] EFLAGS: %08lx CPU: %d\n",
184 (u16)regs->cs, regs->ip, regs->flags,
186 print_symbol("EIP is at %s\n", regs->ip);
188 printk("EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
189 regs->ax, regs->bx, regs->cx, regs->dx);
190 printk("ESI: %08lx EDI: %08lx EBP: %08lx ESP: %08lx\n",
191 regs->si, regs->di, regs->bp, sp);
192 printk(" DS: %04x ES: %04x FS: %04x GS: %04x SS: %04x\n",
193 (u16)regs->ds, (u16)regs->es, (u16)regs->fs, gs, ss);
201 cr4 = read_cr4_safe();
202 printk("CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n",
209 printk("DR0: %08lx DR1: %08lx DR2: %08lx DR3: %08lx\n",
214 printk("DR6: %08lx DR7: %08lx\n",
218 void show_regs(struct pt_regs *regs)
220 __show_registers(regs, 1);
221 show_trace(NULL, regs, ®s->sp, regs->bp);
225 * This gets run with %bx containing the
226 * function to call, and %dx containing
229 extern void kernel_thread_helper(void);
232 * Create a kernel thread
234 int kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
238 memset(®s, 0, sizeof(regs));
240 regs.bx = (unsigned long) fn;
241 regs.dx = (unsigned long) arg;
245 regs.fs = __KERNEL_PERCPU;
247 regs.ip = (unsigned long) kernel_thread_helper;
248 regs.cs = __KERNEL_CS | get_kernel_rpl();
249 regs.flags = X86_EFLAGS_IF | X86_EFLAGS_SF | X86_EFLAGS_PF | 0x2;
251 /* Ok, create the new process.. */
252 return do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0, ®s, 0, NULL, NULL);
254 EXPORT_SYMBOL(kernel_thread);
257 * Free current thread data structures etc..
259 void exit_thread(void)
261 /* The process may have allocated an io port bitmap... nuke it. */
262 if (unlikely(test_thread_flag(TIF_IO_BITMAP))) {
263 struct task_struct *tsk = current;
264 struct thread_struct *t = &tsk->thread;
266 struct tss_struct *tss = &per_cpu(init_tss, cpu);
268 kfree(t->io_bitmap_ptr);
269 t->io_bitmap_ptr = NULL;
270 clear_thread_flag(TIF_IO_BITMAP);
272 * Careful, clear this in the TSS too:
274 memset(tss->io_bitmap, 0xff, tss->io_bitmap_max);
275 t->io_bitmap_max = 0;
276 tss->io_bitmap_owner = NULL;
277 tss->io_bitmap_max = 0;
278 tss->x86_tss.io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
283 void flush_thread(void)
285 struct task_struct *tsk = current;
287 tsk->thread.debugreg0 = 0;
288 tsk->thread.debugreg1 = 0;
289 tsk->thread.debugreg2 = 0;
290 tsk->thread.debugreg3 = 0;
291 tsk->thread.debugreg6 = 0;
292 tsk->thread.debugreg7 = 0;
293 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
294 clear_tsk_thread_flag(tsk, TIF_DEBUG);
296 * Forget coprocessor state..
298 tsk->fpu_counter = 0;
303 void release_thread(struct task_struct *dead_task)
305 BUG_ON(dead_task->mm);
306 release_vm86_irqs(dead_task);
310 * This gets called before we allocate a new thread and copy
311 * the current task into it.
313 void prepare_to_copy(struct task_struct *tsk)
318 int copy_thread(int nr, unsigned long clone_flags, unsigned long sp,
319 unsigned long unused,
320 struct task_struct * p, struct pt_regs * regs)
322 struct pt_regs * childregs;
323 struct task_struct *tsk;
326 childregs = task_pt_regs(p);
331 p->thread.sp = (unsigned long) childregs;
332 p->thread.sp0 = (unsigned long) (childregs+1);
334 p->thread.ip = (unsigned long) ret_from_fork;
336 savesegment(gs, p->thread.gs);
339 if (unlikely(test_tsk_thread_flag(tsk, TIF_IO_BITMAP))) {
340 p->thread.io_bitmap_ptr = kmemdup(tsk->thread.io_bitmap_ptr,
341 IO_BITMAP_BYTES, GFP_KERNEL);
342 if (!p->thread.io_bitmap_ptr) {
343 p->thread.io_bitmap_max = 0;
346 set_tsk_thread_flag(p, TIF_IO_BITMAP);
352 * Set a new TLS for the child thread?
354 if (clone_flags & CLONE_SETTLS)
355 err = do_set_thread_area(p, -1,
356 (struct user_desc __user *)childregs->si, 0);
358 if (err && p->thread.io_bitmap_ptr) {
359 kfree(p->thread.io_bitmap_ptr);
360 p->thread.io_bitmap_max = 0;
366 start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp)
368 __asm__("movl %0, %%gs" :: "r"(0));
371 regs->ds = __USER_DS;
372 regs->es = __USER_DS;
373 regs->ss = __USER_DS;
374 regs->cs = __USER_CS;
378 * Free the old FP and other extended state
380 free_thread_xstate(current);
382 EXPORT_SYMBOL_GPL(start_thread);
384 static void hard_disable_TSC(void)
386 write_cr4(read_cr4() | X86_CR4_TSD);
389 void disable_TSC(void)
392 if (!test_and_set_thread_flag(TIF_NOTSC))
394 * Must flip the CPU state synchronously with
395 * TIF_NOTSC in the current running context.
401 static void hard_enable_TSC(void)
403 write_cr4(read_cr4() & ~X86_CR4_TSD);
406 static void enable_TSC(void)
409 if (test_and_clear_thread_flag(TIF_NOTSC))
411 * Must flip the CPU state synchronously with
412 * TIF_NOTSC in the current running context.
418 int get_tsc_mode(unsigned long adr)
422 if (test_thread_flag(TIF_NOTSC))
423 val = PR_TSC_SIGSEGV;
427 return put_user(val, (unsigned int __user *)adr);
430 int set_tsc_mode(unsigned int val)
432 if (val == PR_TSC_SIGSEGV)
434 else if (val == PR_TSC_ENABLE)
443 __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p,
444 struct tss_struct *tss)
446 struct thread_struct *prev, *next;
447 unsigned long debugctl;
449 prev = &prev_p->thread;
450 next = &next_p->thread;
452 debugctl = prev->debugctlmsr;
453 if (next->ds_area_msr != prev->ds_area_msr) {
454 /* we clear debugctl to make sure DS
455 * is not in use when we change it */
457 update_debugctlmsr(0);
458 wrmsr(MSR_IA32_DS_AREA, next->ds_area_msr, 0);
461 if (next->debugctlmsr != debugctl)
462 update_debugctlmsr(next->debugctlmsr);
464 if (test_tsk_thread_flag(next_p, TIF_DEBUG)) {
465 set_debugreg(next->debugreg0, 0);
466 set_debugreg(next->debugreg1, 1);
467 set_debugreg(next->debugreg2, 2);
468 set_debugreg(next->debugreg3, 3);
470 set_debugreg(next->debugreg6, 6);
471 set_debugreg(next->debugreg7, 7);
474 if (test_tsk_thread_flag(prev_p, TIF_NOTSC) ^
475 test_tsk_thread_flag(next_p, TIF_NOTSC)) {
476 /* prev and next are different */
477 if (test_tsk_thread_flag(next_p, TIF_NOTSC))
484 if (test_tsk_thread_flag(prev_p, TIF_BTS_TRACE_TS))
485 ptrace_bts_take_timestamp(prev_p, BTS_TASK_DEPARTS);
487 if (test_tsk_thread_flag(next_p, TIF_BTS_TRACE_TS))
488 ptrace_bts_take_timestamp(next_p, BTS_TASK_ARRIVES);
492 if (!test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
494 * Disable the bitmap via an invalid offset. We still cache
495 * the previous bitmap owner and the IO bitmap contents:
497 tss->x86_tss.io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
501 if (likely(next == tss->io_bitmap_owner)) {
503 * Previous owner of the bitmap (hence the bitmap content)
504 * matches the next task, we dont have to do anything but
505 * to set a valid offset in the TSS:
507 tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET;
511 * Lazy TSS's I/O bitmap copy. We set an invalid offset here
512 * and we let the task to get a GPF in case an I/O instruction
513 * is performed. The handler of the GPF will verify that the
514 * faulting task has a valid I/O bitmap and, it true, does the
515 * real copy and restart the instruction. This will save us
516 * redundant copies when the currently switched task does not
517 * perform any I/O during its timeslice.
519 tss->x86_tss.io_bitmap_base = INVALID_IO_BITMAP_OFFSET_LAZY;
523 * switch_to(x,yn) should switch tasks from x to y.
525 * We fsave/fwait so that an exception goes off at the right time
526 * (as a call from the fsave or fwait in effect) rather than to
527 * the wrong process. Lazy FP saving no longer makes any sense
528 * with modern CPU's, and this simplifies a lot of things (SMP
529 * and UP become the same).
531 * NOTE! We used to use the x86 hardware context switching. The
532 * reason for not using it any more becomes apparent when you
533 * try to recover gracefully from saved state that is no longer
534 * valid (stale segment register values in particular). With the
535 * hardware task-switch, there is no way to fix up bad state in
536 * a reasonable manner.
538 * The fact that Intel documents the hardware task-switching to
539 * be slow is a fairly red herring - this code is not noticeably
540 * faster. However, there _is_ some room for improvement here,
541 * so the performance issues may eventually be a valid point.
542 * More important, however, is the fact that this allows us much
545 * The return value (in %ax) will be the "prev" task after
546 * the task-switch, and shows up in ret_from_fork in entry.S,
549 struct task_struct * __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
551 struct thread_struct *prev = &prev_p->thread,
552 *next = &next_p->thread;
553 int cpu = smp_processor_id();
554 struct tss_struct *tss = &per_cpu(init_tss, cpu);
556 /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
558 __unlazy_fpu(prev_p);
561 /* we're going to use this soon, after a few expensive things */
562 if (next_p->fpu_counter > 5)
563 prefetch(next->xstate);
571 * Save away %gs. No need to save %fs, as it was saved on the
572 * stack on entry. No need to save %es and %ds, as those are
573 * always kernel segments while inside the kernel. Doing this
574 * before setting the new TLS descriptors avoids the situation
575 * where we temporarily have non-reloadable segments in %fs
576 * and %gs. This could be an issue if the NMI handler ever
577 * used %fs or %gs (it does not today), or if the kernel is
578 * running inside of a hypervisor layer.
580 savesegment(gs, prev->gs);
583 * Load the per-thread Thread-Local Storage descriptor.
588 * Restore IOPL if needed. In normal use, the flags restore
589 * in the switch assembly will handle this. But if the kernel
590 * is running virtualized at a non-zero CPL, the popf will
591 * not restore flags, so it must be done in a separate step.
593 if (get_kernel_rpl() && unlikely(prev->iopl != next->iopl))
594 set_iopl_mask(next->iopl);
597 * Now maybe handle debug registers and/or IO bitmaps
599 if (unlikely(task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV ||
600 task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT))
601 __switch_to_xtra(prev_p, next_p, tss);
604 * Leave lazy mode, flushing any hypercalls made here.
605 * This must be done before restoring TLS segments so
606 * the GDT and LDT are properly updated, and must be
607 * done before math_state_restore, so the TS bit is up
610 arch_leave_lazy_cpu_mode();
612 /* If the task has used fpu the last 5 timeslices, just do a full
613 * restore of the math state immediately to avoid the trap; the
614 * chances of needing FPU soon are obviously high now
616 * tsk_used_math() checks prevent calling math_state_restore(),
617 * which can sleep in the case of !tsk_used_math()
619 if (tsk_used_math(next_p) && next_p->fpu_counter > 5)
620 math_state_restore();
623 * Restore %gs if needed (which is common)
625 if (prev->gs | next->gs)
626 loadsegment(gs, next->gs);
628 x86_write_percpu(current_task, next_p);
633 asmlinkage int sys_fork(struct pt_regs regs)
635 return do_fork(SIGCHLD, regs.sp, ®s, 0, NULL, NULL);
638 asmlinkage int sys_clone(struct pt_regs regs)
640 unsigned long clone_flags;
642 int __user *parent_tidptr, *child_tidptr;
644 clone_flags = regs.bx;
646 parent_tidptr = (int __user *)regs.dx;
647 child_tidptr = (int __user *)regs.di;
650 return do_fork(clone_flags, newsp, ®s, 0, parent_tidptr, child_tidptr);
654 * This is trivial, and on the face of it looks like it
655 * could equally well be done in user mode.
657 * Not so, for quite unobvious reasons - register pressure.
658 * In user mode vfork() cannot have a stack frame, and if
659 * done by calling the "clone()" system call directly, you
660 * do not have enough call-clobbered registers to hold all
661 * the information you need.
663 asmlinkage int sys_vfork(struct pt_regs regs)
665 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs.sp, ®s, 0, NULL, NULL);
669 * sys_execve() executes a new program.
671 asmlinkage int sys_execve(struct pt_regs regs)
676 filename = getname((char __user *) regs.bx);
677 error = PTR_ERR(filename);
678 if (IS_ERR(filename))
680 error = do_execve(filename,
681 (char __user * __user *) regs.cx,
682 (char __user * __user *) regs.dx,
685 /* Make sure we don't return using sysenter.. */
686 set_thread_flag(TIF_IRET);
693 #define top_esp (THREAD_SIZE - sizeof(unsigned long))
694 #define top_ebp (THREAD_SIZE - 2*sizeof(unsigned long))
696 unsigned long get_wchan(struct task_struct *p)
698 unsigned long bp, sp, ip;
699 unsigned long stack_page;
701 if (!p || p == current || p->state == TASK_RUNNING)
703 stack_page = (unsigned long)task_stack_page(p);
705 if (!stack_page || sp < stack_page || sp > top_esp+stack_page)
707 /* include/asm-i386/system.h:switch_to() pushes bp last. */
708 bp = *(unsigned long *) sp;
710 if (bp < stack_page || bp > top_ebp+stack_page)
712 ip = *(unsigned long *) (bp+4);
713 if (!in_sched_functions(ip))
715 bp = *(unsigned long *) bp;
716 } while (count++ < 16);
720 unsigned long arch_align_stack(unsigned long sp)
722 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
723 sp -= get_random_int() % 8192;
727 unsigned long arch_randomize_brk(struct mm_struct *mm)
729 unsigned long range_end = mm->brk + 0x02000000;
730 return randomize_range(mm->brk, range_end, 0) ? : mm->brk;