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>
59 #include <asm/syscalls.h>
62 asmlinkage void ret_from_fork(void) __asm__("ret_from_fork");
64 DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task;
65 EXPORT_PER_CPU_SYMBOL(current_task);
67 DEFINE_PER_CPU(int, cpu_number);
68 EXPORT_PER_CPU_SYMBOL(cpu_number);
71 * Return saved PC of a blocked thread.
73 unsigned long thread_saved_pc(struct task_struct *tsk)
75 return ((unsigned long *)tsk->thread.sp)[3];
78 #ifdef CONFIG_HOTPLUG_CPU
81 static void cpu_exit_clear(void)
83 int cpu = raw_smp_processor_id();
90 cpu_clear(cpu, cpu_callout_map);
91 cpu_clear(cpu, cpu_callin_map);
97 /* We don't actually take CPU down, just spin without interrupts. */
98 static inline void play_dead(void)
100 /* This must be done before dead CPU ack */
104 __get_cpu_var(cpu_state) = CPU_DEAD;
107 * With physical CPU hotplug, we should halt the cpu
110 /* mask all interrupts, flush any and all caches, and halt */
114 static inline void play_dead(void)
118 #endif /* CONFIG_HOTPLUG_CPU */
121 * The idle thread. There's no useful work to be
122 * done, so just try to conserve power and have a
123 * low exit latency (ie sit in a loop waiting for
124 * somebody to say that they'd like to reschedule)
128 int cpu = smp_processor_id();
130 current_thread_info()->status |= TS_POLLING;
132 /* endless idle loop with no priority at all */
134 tick_nohz_stop_sched_tick(1);
135 while (!need_resched()) {
140 if (rcu_pending(cpu))
141 rcu_check_callbacks(cpu, 0);
143 if (cpu_is_offline(cpu))
147 __get_cpu_var(irq_stat).idle_timestamp = jiffies;
148 /* Don't trace irqs off for idle */
149 stop_critical_timings();
151 start_critical_timings();
153 tick_nohz_restart_sched_tick();
154 preempt_enable_no_resched();
160 void __show_registers(struct pt_regs *regs, int all)
162 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L;
163 unsigned long d0, d1, d2, d3, d6, d7;
165 unsigned short ss, gs;
167 if (user_mode_vm(regs)) {
169 ss = regs->ss & 0xffff;
172 sp = (unsigned long) (®s->sp);
178 printk("Pid: %d, comm: %s %s (%s %.*s)\n",
179 task_pid_nr(current), current->comm,
180 print_tainted(), init_utsname()->release,
181 (int)strcspn(init_utsname()->version, " "),
182 init_utsname()->version);
184 printk("EIP: %04x:[<%08lx>] EFLAGS: %08lx CPU: %d\n",
185 (u16)regs->cs, regs->ip, regs->flags,
187 print_symbol("EIP is at %s\n", regs->ip);
189 printk("EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
190 regs->ax, regs->bx, regs->cx, regs->dx);
191 printk("ESI: %08lx EDI: %08lx EBP: %08lx ESP: %08lx\n",
192 regs->si, regs->di, regs->bp, sp);
193 printk(" DS: %04x ES: %04x FS: %04x GS: %04x SS: %04x\n",
194 (u16)regs->ds, (u16)regs->es, (u16)regs->fs, gs, ss);
202 cr4 = read_cr4_safe();
203 printk("CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n",
210 printk("DR0: %08lx DR1: %08lx DR2: %08lx DR3: %08lx\n",
215 printk("DR6: %08lx DR7: %08lx\n",
219 void show_regs(struct pt_regs *regs)
221 __show_registers(regs, 1);
222 show_trace(NULL, regs, ®s->sp, regs->bp);
226 * This gets run with %bx containing the
227 * function to call, and %dx containing
230 extern void kernel_thread_helper(void);
233 * Create a kernel thread
235 int kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
239 memset(®s, 0, sizeof(regs));
241 regs.bx = (unsigned long) fn;
242 regs.dx = (unsigned long) arg;
246 regs.fs = __KERNEL_PERCPU;
248 regs.ip = (unsigned long) kernel_thread_helper;
249 regs.cs = __KERNEL_CS | get_kernel_rpl();
250 regs.flags = X86_EFLAGS_IF | X86_EFLAGS_SF | X86_EFLAGS_PF | 0x2;
252 /* Ok, create the new process.. */
253 return do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0, ®s, 0, NULL, NULL);
255 EXPORT_SYMBOL(kernel_thread);
258 * Free current thread data structures etc..
260 void exit_thread(void)
262 /* The process may have allocated an io port bitmap... nuke it. */
263 if (unlikely(test_thread_flag(TIF_IO_BITMAP))) {
264 struct task_struct *tsk = current;
265 struct thread_struct *t = &tsk->thread;
267 struct tss_struct *tss = &per_cpu(init_tss, cpu);
269 kfree(t->io_bitmap_ptr);
270 t->io_bitmap_ptr = NULL;
271 clear_thread_flag(TIF_IO_BITMAP);
273 * Careful, clear this in the TSS too:
275 memset(tss->io_bitmap, 0xff, tss->io_bitmap_max);
276 t->io_bitmap_max = 0;
277 tss->io_bitmap_owner = NULL;
278 tss->io_bitmap_max = 0;
279 tss->x86_tss.io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
283 /* Free any DS contexts that have not been properly released. */
284 if (unlikely(current->thread.ds_ctx)) {
285 /* we clear debugctl to make sure DS is not used. */
286 update_debugctlmsr(0);
287 ds_free(current->thread.ds_ctx);
289 #endif /* CONFIG_X86_DS */
292 void flush_thread(void)
294 struct task_struct *tsk = current;
296 tsk->thread.debugreg0 = 0;
297 tsk->thread.debugreg1 = 0;
298 tsk->thread.debugreg2 = 0;
299 tsk->thread.debugreg3 = 0;
300 tsk->thread.debugreg6 = 0;
301 tsk->thread.debugreg7 = 0;
302 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
303 clear_tsk_thread_flag(tsk, TIF_DEBUG);
305 * Forget coprocessor state..
307 tsk->fpu_counter = 0;
312 void release_thread(struct task_struct *dead_task)
314 BUG_ON(dead_task->mm);
315 release_vm86_irqs(dead_task);
319 * This gets called before we allocate a new thread and copy
320 * the current task into it.
322 void prepare_to_copy(struct task_struct *tsk)
327 int copy_thread(int nr, unsigned long clone_flags, unsigned long sp,
328 unsigned long unused,
329 struct task_struct * p, struct pt_regs * regs)
331 struct pt_regs * childregs;
332 struct task_struct *tsk;
335 childregs = task_pt_regs(p);
340 p->thread.sp = (unsigned long) childregs;
341 p->thread.sp0 = (unsigned long) (childregs+1);
343 p->thread.ip = (unsigned long) ret_from_fork;
345 savesegment(gs, p->thread.gs);
348 if (unlikely(test_tsk_thread_flag(tsk, TIF_IO_BITMAP))) {
349 p->thread.io_bitmap_ptr = kmemdup(tsk->thread.io_bitmap_ptr,
350 IO_BITMAP_BYTES, GFP_KERNEL);
351 if (!p->thread.io_bitmap_ptr) {
352 p->thread.io_bitmap_max = 0;
355 set_tsk_thread_flag(p, TIF_IO_BITMAP);
361 * Set a new TLS for the child thread?
363 if (clone_flags & CLONE_SETTLS)
364 err = do_set_thread_area(p, -1,
365 (struct user_desc __user *)childregs->si, 0);
367 if (err && p->thread.io_bitmap_ptr) {
368 kfree(p->thread.io_bitmap_ptr);
369 p->thread.io_bitmap_max = 0;
375 start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp)
377 __asm__("movl %0, %%gs" :: "r"(0));
380 regs->ds = __USER_DS;
381 regs->es = __USER_DS;
382 regs->ss = __USER_DS;
383 regs->cs = __USER_CS;
387 * Free the old FP and other extended state
389 free_thread_xstate(current);
391 EXPORT_SYMBOL_GPL(start_thread);
393 static void hard_disable_TSC(void)
395 write_cr4(read_cr4() | X86_CR4_TSD);
398 void disable_TSC(void)
401 if (!test_and_set_thread_flag(TIF_NOTSC))
403 * Must flip the CPU state synchronously with
404 * TIF_NOTSC in the current running context.
410 static void hard_enable_TSC(void)
412 write_cr4(read_cr4() & ~X86_CR4_TSD);
415 static void enable_TSC(void)
418 if (test_and_clear_thread_flag(TIF_NOTSC))
420 * Must flip the CPU state synchronously with
421 * TIF_NOTSC in the current running context.
427 int get_tsc_mode(unsigned long adr)
431 if (test_thread_flag(TIF_NOTSC))
432 val = PR_TSC_SIGSEGV;
436 return put_user(val, (unsigned int __user *)adr);
439 int set_tsc_mode(unsigned int val)
441 if (val == PR_TSC_SIGSEGV)
443 else if (val == PR_TSC_ENABLE)
452 static int update_debugctl(struct thread_struct *prev,
453 struct thread_struct *next, unsigned long debugctl)
455 unsigned long ds_prev = 0;
456 unsigned long ds_next = 0;
459 ds_prev = (unsigned long)prev->ds_ctx->ds;
461 ds_next = (unsigned long)next->ds_ctx->ds;
463 if (ds_next != ds_prev) {
464 /* we clear debugctl to make sure DS
465 * is not in use when we change it */
467 update_debugctlmsr(0);
468 wrmsr(MSR_IA32_DS_AREA, ds_next, 0);
473 static int update_debugctl(struct thread_struct *prev,
474 struct thread_struct *next, unsigned long debugctl)
478 #endif /* CONFIG_X86_DS */
481 __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p,
482 struct tss_struct *tss)
484 struct thread_struct *prev, *next;
485 unsigned long debugctl;
487 prev = &prev_p->thread;
488 next = &next_p->thread;
490 debugctl = update_debugctl(prev, next, prev->debugctlmsr);
492 if (next->debugctlmsr != debugctl)
493 update_debugctlmsr(next->debugctlmsr);
495 if (test_tsk_thread_flag(next_p, TIF_DEBUG)) {
496 set_debugreg(next->debugreg0, 0);
497 set_debugreg(next->debugreg1, 1);
498 set_debugreg(next->debugreg2, 2);
499 set_debugreg(next->debugreg3, 3);
501 set_debugreg(next->debugreg6, 6);
502 set_debugreg(next->debugreg7, 7);
505 if (test_tsk_thread_flag(prev_p, TIF_NOTSC) ^
506 test_tsk_thread_flag(next_p, TIF_NOTSC)) {
507 /* prev and next are different */
508 if (test_tsk_thread_flag(next_p, TIF_NOTSC))
514 #ifdef CONFIG_X86_PTRACE_BTS
515 if (test_tsk_thread_flag(prev_p, TIF_BTS_TRACE_TS))
516 ptrace_bts_take_timestamp(prev_p, BTS_TASK_DEPARTS);
518 if (test_tsk_thread_flag(next_p, TIF_BTS_TRACE_TS))
519 ptrace_bts_take_timestamp(next_p, BTS_TASK_ARRIVES);
520 #endif /* CONFIG_X86_PTRACE_BTS */
523 if (!test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
525 * Disable the bitmap via an invalid offset. We still cache
526 * the previous bitmap owner and the IO bitmap contents:
528 tss->x86_tss.io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
532 if (likely(next == tss->io_bitmap_owner)) {
534 * Previous owner of the bitmap (hence the bitmap content)
535 * matches the next task, we dont have to do anything but
536 * to set a valid offset in the TSS:
538 tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET;
542 * Lazy TSS's I/O bitmap copy. We set an invalid offset here
543 * and we let the task to get a GPF in case an I/O instruction
544 * is performed. The handler of the GPF will verify that the
545 * faulting task has a valid I/O bitmap and, it true, does the
546 * real copy and restart the instruction. This will save us
547 * redundant copies when the currently switched task does not
548 * perform any I/O during its timeslice.
550 tss->x86_tss.io_bitmap_base = INVALID_IO_BITMAP_OFFSET_LAZY;
554 * switch_to(x,yn) should switch tasks from x to y.
556 * We fsave/fwait so that an exception goes off at the right time
557 * (as a call from the fsave or fwait in effect) rather than to
558 * the wrong process. Lazy FP saving no longer makes any sense
559 * with modern CPU's, and this simplifies a lot of things (SMP
560 * and UP become the same).
562 * NOTE! We used to use the x86 hardware context switching. The
563 * reason for not using it any more becomes apparent when you
564 * try to recover gracefully from saved state that is no longer
565 * valid (stale segment register values in particular). With the
566 * hardware task-switch, there is no way to fix up bad state in
567 * a reasonable manner.
569 * The fact that Intel documents the hardware task-switching to
570 * be slow is a fairly red herring - this code is not noticeably
571 * faster. However, there _is_ some room for improvement here,
572 * so the performance issues may eventually be a valid point.
573 * More important, however, is the fact that this allows us much
576 * The return value (in %ax) will be the "prev" task after
577 * the task-switch, and shows up in ret_from_fork in entry.S,
580 struct task_struct * __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
582 struct thread_struct *prev = &prev_p->thread,
583 *next = &next_p->thread;
584 int cpu = smp_processor_id();
585 struct tss_struct *tss = &per_cpu(init_tss, cpu);
587 /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
589 __unlazy_fpu(prev_p);
592 /* we're going to use this soon, after a few expensive things */
593 if (next_p->fpu_counter > 5)
594 prefetch(next->xstate);
602 * Save away %gs. No need to save %fs, as it was saved on the
603 * stack on entry. No need to save %es and %ds, as those are
604 * always kernel segments while inside the kernel. Doing this
605 * before setting the new TLS descriptors avoids the situation
606 * where we temporarily have non-reloadable segments in %fs
607 * and %gs. This could be an issue if the NMI handler ever
608 * used %fs or %gs (it does not today), or if the kernel is
609 * running inside of a hypervisor layer.
611 savesegment(gs, prev->gs);
614 * Load the per-thread Thread-Local Storage descriptor.
619 * Restore IOPL if needed. In normal use, the flags restore
620 * in the switch assembly will handle this. But if the kernel
621 * is running virtualized at a non-zero CPL, the popf will
622 * not restore flags, so it must be done in a separate step.
624 if (get_kernel_rpl() && unlikely(prev->iopl != next->iopl))
625 set_iopl_mask(next->iopl);
628 * Now maybe handle debug registers and/or IO bitmaps
630 if (unlikely(task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV ||
631 task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT))
632 __switch_to_xtra(prev_p, next_p, tss);
635 * Leave lazy mode, flushing any hypercalls made here.
636 * This must be done before restoring TLS segments so
637 * the GDT and LDT are properly updated, and must be
638 * done before math_state_restore, so the TS bit is up
641 arch_leave_lazy_cpu_mode();
643 /* If the task has used fpu the last 5 timeslices, just do a full
644 * restore of the math state immediately to avoid the trap; the
645 * chances of needing FPU soon are obviously high now
647 * tsk_used_math() checks prevent calling math_state_restore(),
648 * which can sleep in the case of !tsk_used_math()
650 if (tsk_used_math(next_p) && next_p->fpu_counter > 5)
651 math_state_restore();
654 * Restore %gs if needed (which is common)
656 if (prev->gs | next->gs)
657 loadsegment(gs, next->gs);
659 x86_write_percpu(current_task, next_p);
664 asmlinkage int sys_fork(struct pt_regs regs)
666 return do_fork(SIGCHLD, regs.sp, ®s, 0, NULL, NULL);
669 asmlinkage int sys_clone(struct pt_regs regs)
671 unsigned long clone_flags;
673 int __user *parent_tidptr, *child_tidptr;
675 clone_flags = regs.bx;
677 parent_tidptr = (int __user *)regs.dx;
678 child_tidptr = (int __user *)regs.di;
681 return do_fork(clone_flags, newsp, ®s, 0, parent_tidptr, child_tidptr);
685 * This is trivial, and on the face of it looks like it
686 * could equally well be done in user mode.
688 * Not so, for quite unobvious reasons - register pressure.
689 * In user mode vfork() cannot have a stack frame, and if
690 * done by calling the "clone()" system call directly, you
691 * do not have enough call-clobbered registers to hold all
692 * the information you need.
694 asmlinkage int sys_vfork(struct pt_regs regs)
696 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs.sp, ®s, 0, NULL, NULL);
700 * sys_execve() executes a new program.
702 asmlinkage int sys_execve(struct pt_regs regs)
707 filename = getname((char __user *) regs.bx);
708 error = PTR_ERR(filename);
709 if (IS_ERR(filename))
711 error = do_execve(filename,
712 (char __user * __user *) regs.cx,
713 (char __user * __user *) regs.dx,
716 /* Make sure we don't return using sysenter.. */
717 set_thread_flag(TIF_IRET);
724 #define top_esp (THREAD_SIZE - sizeof(unsigned long))
725 #define top_ebp (THREAD_SIZE - 2*sizeof(unsigned long))
727 unsigned long get_wchan(struct task_struct *p)
729 unsigned long bp, sp, ip;
730 unsigned long stack_page;
732 if (!p || p == current || p->state == TASK_RUNNING)
734 stack_page = (unsigned long)task_stack_page(p);
736 if (!stack_page || sp < stack_page || sp > top_esp+stack_page)
738 /* include/asm-i386/system.h:switch_to() pushes bp last. */
739 bp = *(unsigned long *) sp;
741 if (bp < stack_page || bp > top_ebp+stack_page)
743 ip = *(unsigned long *) (bp+4);
744 if (!in_sched_functions(ip))
746 bp = *(unsigned long *) bp;
747 } while (count++ < 16);
751 unsigned long arch_align_stack(unsigned long sp)
753 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
754 sp -= get_random_int() % 8192;
758 unsigned long arch_randomize_brk(struct mm_struct *mm)
760 unsigned long range_end = mm->brk + 0x02000000;
761 return randomize_range(mm->brk, range_end, 0) ? : mm->brk;