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 asmlinkage void ret_from_fork(void) __asm__("ret_from_fork");
61 DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task;
62 EXPORT_PER_CPU_SYMBOL(current_task);
64 DEFINE_PER_CPU(int, cpu_number);
65 EXPORT_PER_CPU_SYMBOL(cpu_number);
68 * Return saved PC of a blocked thread.
70 unsigned long thread_saved_pc(struct task_struct *tsk)
72 return ((unsigned long *)tsk->thread.sp)[3];
75 #ifdef CONFIG_HOTPLUG_CPU
78 static void cpu_exit_clear(void)
80 int cpu = raw_smp_processor_id();
87 cpu_clear(cpu, cpu_callout_map);
88 cpu_clear(cpu, cpu_callin_map);
93 /* We don't actually take CPU down, just spin without interrupts. */
94 static inline void play_dead(void)
96 /* This must be done before dead CPU ack */
101 __get_cpu_var(cpu_state) = CPU_DEAD;
104 * With physical CPU hotplug, we should halt the cpu
111 static inline void play_dead(void)
115 #endif /* CONFIG_HOTPLUG_CPU */
118 * The idle thread. There's no useful work to be
119 * done, so just try to conserve power and have a
120 * low exit latency (ie sit in a loop waiting for
121 * somebody to say that they'd like to reschedule)
125 int cpu = smp_processor_id();
127 current_thread_info()->status |= TS_POLLING;
129 /* endless idle loop with no priority at all */
131 tick_nohz_stop_sched_tick();
132 while (!need_resched()) {
137 if (rcu_pending(cpu))
138 rcu_check_callbacks(cpu, 0);
140 if (cpu_is_offline(cpu))
144 __get_cpu_var(irq_stat).idle_timestamp = jiffies;
145 /* Don't trace irqs off for idle */
146 stop_critical_timings();
148 start_critical_timings();
150 tick_nohz_restart_sched_tick();
151 preempt_enable_no_resched();
157 void __show_registers(struct pt_regs *regs, int all)
159 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L;
160 unsigned long d0, d1, d2, d3, d6, d7;
162 unsigned short ss, gs;
164 if (user_mode_vm(regs)) {
166 ss = regs->ss & 0xffff;
169 sp = (unsigned long) (®s->sp);
175 printk("Pid: %d, comm: %s %s (%s %.*s)\n",
176 task_pid_nr(current), current->comm,
177 print_tainted(), init_utsname()->release,
178 (int)strcspn(init_utsname()->version, " "),
179 init_utsname()->version);
181 printk("EIP: %04x:[<%08lx>] EFLAGS: %08lx CPU: %d\n",
182 (u16)regs->cs, regs->ip, regs->flags,
184 print_symbol("EIP is at %s\n", regs->ip);
186 printk("EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
187 regs->ax, regs->bx, regs->cx, regs->dx);
188 printk("ESI: %08lx EDI: %08lx EBP: %08lx ESP: %08lx\n",
189 regs->si, regs->di, regs->bp, sp);
190 printk(" DS: %04x ES: %04x FS: %04x GS: %04x SS: %04x\n",
191 (u16)regs->ds, (u16)regs->es, (u16)regs->fs, gs, ss);
199 cr4 = read_cr4_safe();
200 printk("CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n",
207 printk("DR0: %08lx DR1: %08lx DR2: %08lx DR3: %08lx\n",
212 printk("DR6: %08lx DR7: %08lx\n",
216 void show_regs(struct pt_regs *regs)
218 __show_registers(regs, 1);
219 show_trace(NULL, regs, ®s->sp, regs->bp);
223 * This gets run with %bx containing the
224 * function to call, and %dx containing
227 extern void kernel_thread_helper(void);
230 * Create a kernel thread
232 int kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
236 memset(®s, 0, sizeof(regs));
238 regs.bx = (unsigned long) fn;
239 regs.dx = (unsigned long) arg;
243 regs.fs = __KERNEL_PERCPU;
245 regs.ip = (unsigned long) kernel_thread_helper;
246 regs.cs = __KERNEL_CS | get_kernel_rpl();
247 regs.flags = X86_EFLAGS_IF | X86_EFLAGS_SF | X86_EFLAGS_PF | 0x2;
249 /* Ok, create the new process.. */
250 return do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0, ®s, 0, NULL, NULL);
252 EXPORT_SYMBOL(kernel_thread);
255 * Free current thread data structures etc..
257 void exit_thread(void)
259 /* The process may have allocated an io port bitmap... nuke it. */
260 if (unlikely(test_thread_flag(TIF_IO_BITMAP))) {
261 struct task_struct *tsk = current;
262 struct thread_struct *t = &tsk->thread;
264 struct tss_struct *tss = &per_cpu(init_tss, cpu);
266 kfree(t->io_bitmap_ptr);
267 t->io_bitmap_ptr = NULL;
268 clear_thread_flag(TIF_IO_BITMAP);
270 * Careful, clear this in the TSS too:
272 memset(tss->io_bitmap, 0xff, tss->io_bitmap_max);
273 t->io_bitmap_max = 0;
274 tss->io_bitmap_owner = NULL;
275 tss->io_bitmap_max = 0;
276 tss->x86_tss.io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
281 void flush_thread(void)
283 struct task_struct *tsk = current;
285 tsk->thread.debugreg0 = 0;
286 tsk->thread.debugreg1 = 0;
287 tsk->thread.debugreg2 = 0;
288 tsk->thread.debugreg3 = 0;
289 tsk->thread.debugreg6 = 0;
290 tsk->thread.debugreg7 = 0;
291 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
292 clear_tsk_thread_flag(tsk, TIF_DEBUG);
294 * Forget coprocessor state..
296 tsk->fpu_counter = 0;
301 void release_thread(struct task_struct *dead_task)
303 BUG_ON(dead_task->mm);
304 release_vm86_irqs(dead_task);
308 * This gets called before we allocate a new thread and copy
309 * the current task into it.
311 void prepare_to_copy(struct task_struct *tsk)
316 int copy_thread(int nr, unsigned long clone_flags, unsigned long sp,
317 unsigned long unused,
318 struct task_struct * p, struct pt_regs * regs)
320 struct pt_regs * childregs;
321 struct task_struct *tsk;
324 childregs = task_pt_regs(p);
329 p->thread.sp = (unsigned long) childregs;
330 p->thread.sp0 = (unsigned long) (childregs+1);
332 p->thread.ip = (unsigned long) ret_from_fork;
334 savesegment(gs, p->thread.gs);
337 if (unlikely(test_tsk_thread_flag(tsk, TIF_IO_BITMAP))) {
338 p->thread.io_bitmap_ptr = kmemdup(tsk->thread.io_bitmap_ptr,
339 IO_BITMAP_BYTES, GFP_KERNEL);
340 if (!p->thread.io_bitmap_ptr) {
341 p->thread.io_bitmap_max = 0;
344 set_tsk_thread_flag(p, TIF_IO_BITMAP);
350 * Set a new TLS for the child thread?
352 if (clone_flags & CLONE_SETTLS)
353 err = do_set_thread_area(p, -1,
354 (struct user_desc __user *)childregs->si, 0);
356 if (err && p->thread.io_bitmap_ptr) {
357 kfree(p->thread.io_bitmap_ptr);
358 p->thread.io_bitmap_max = 0;
364 start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp)
366 __asm__("movl %0, %%gs" :: "r"(0));
369 regs->ds = __USER_DS;
370 regs->es = __USER_DS;
371 regs->ss = __USER_DS;
372 regs->cs = __USER_CS;
376 * Free the old FP and other extended state
378 free_thread_xstate(current);
380 EXPORT_SYMBOL_GPL(start_thread);
382 static void hard_disable_TSC(void)
384 write_cr4(read_cr4() | X86_CR4_TSD);
387 void disable_TSC(void)
390 if (!test_and_set_thread_flag(TIF_NOTSC))
392 * Must flip the CPU state synchronously with
393 * TIF_NOTSC in the current running context.
399 static void hard_enable_TSC(void)
401 write_cr4(read_cr4() & ~X86_CR4_TSD);
404 static void enable_TSC(void)
407 if (test_and_clear_thread_flag(TIF_NOTSC))
409 * Must flip the CPU state synchronously with
410 * TIF_NOTSC in the current running context.
416 int get_tsc_mode(unsigned long adr)
420 if (test_thread_flag(TIF_NOTSC))
421 val = PR_TSC_SIGSEGV;
425 return put_user(val, (unsigned int __user *)adr);
428 int set_tsc_mode(unsigned int val)
430 if (val == PR_TSC_SIGSEGV)
432 else if (val == PR_TSC_ENABLE)
441 __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p,
442 struct tss_struct *tss)
444 struct thread_struct *prev, *next;
445 unsigned long debugctl;
447 prev = &prev_p->thread;
448 next = &next_p->thread;
450 debugctl = prev->debugctlmsr;
451 if (next->ds_area_msr != prev->ds_area_msr) {
452 /* we clear debugctl to make sure DS
453 * is not in use when we change it */
455 update_debugctlmsr(0);
456 wrmsr(MSR_IA32_DS_AREA, next->ds_area_msr, 0);
459 if (next->debugctlmsr != debugctl)
460 update_debugctlmsr(next->debugctlmsr);
462 if (test_tsk_thread_flag(next_p, TIF_DEBUG)) {
463 set_debugreg(next->debugreg0, 0);
464 set_debugreg(next->debugreg1, 1);
465 set_debugreg(next->debugreg2, 2);
466 set_debugreg(next->debugreg3, 3);
468 set_debugreg(next->debugreg6, 6);
469 set_debugreg(next->debugreg7, 7);
472 if (test_tsk_thread_flag(prev_p, TIF_NOTSC) ^
473 test_tsk_thread_flag(next_p, TIF_NOTSC)) {
474 /* prev and next are different */
475 if (test_tsk_thread_flag(next_p, TIF_NOTSC))
482 if (test_tsk_thread_flag(prev_p, TIF_BTS_TRACE_TS))
483 ptrace_bts_take_timestamp(prev_p, BTS_TASK_DEPARTS);
485 if (test_tsk_thread_flag(next_p, TIF_BTS_TRACE_TS))
486 ptrace_bts_take_timestamp(next_p, BTS_TASK_ARRIVES);
490 if (!test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
492 * Disable the bitmap via an invalid offset. We still cache
493 * the previous bitmap owner and the IO bitmap contents:
495 tss->x86_tss.io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
499 if (likely(next == tss->io_bitmap_owner)) {
501 * Previous owner of the bitmap (hence the bitmap content)
502 * matches the next task, we dont have to do anything but
503 * to set a valid offset in the TSS:
505 tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET;
509 * Lazy TSS's I/O bitmap copy. We set an invalid offset here
510 * and we let the task to get a GPF in case an I/O instruction
511 * is performed. The handler of the GPF will verify that the
512 * faulting task has a valid I/O bitmap and, it true, does the
513 * real copy and restart the instruction. This will save us
514 * redundant copies when the currently switched task does not
515 * perform any I/O during its timeslice.
517 tss->x86_tss.io_bitmap_base = INVALID_IO_BITMAP_OFFSET_LAZY;
521 * switch_to(x,yn) should switch tasks from x to y.
523 * We fsave/fwait so that an exception goes off at the right time
524 * (as a call from the fsave or fwait in effect) rather than to
525 * the wrong process. Lazy FP saving no longer makes any sense
526 * with modern CPU's, and this simplifies a lot of things (SMP
527 * and UP become the same).
529 * NOTE! We used to use the x86 hardware context switching. The
530 * reason for not using it any more becomes apparent when you
531 * try to recover gracefully from saved state that is no longer
532 * valid (stale segment register values in particular). With the
533 * hardware task-switch, there is no way to fix up bad state in
534 * a reasonable manner.
536 * The fact that Intel documents the hardware task-switching to
537 * be slow is a fairly red herring - this code is not noticeably
538 * faster. However, there _is_ some room for improvement here,
539 * so the performance issues may eventually be a valid point.
540 * More important, however, is the fact that this allows us much
543 * The return value (in %ax) will be the "prev" task after
544 * the task-switch, and shows up in ret_from_fork in entry.S,
547 struct task_struct * __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
549 struct thread_struct *prev = &prev_p->thread,
550 *next = &next_p->thread;
551 int cpu = smp_processor_id();
552 struct tss_struct *tss = &per_cpu(init_tss, cpu);
554 /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
556 __unlazy_fpu(prev_p);
559 /* we're going to use this soon, after a few expensive things */
560 if (next_p->fpu_counter > 5)
561 prefetch(next->xstate);
569 * Save away %gs. No need to save %fs, as it was saved on the
570 * stack on entry. No need to save %es and %ds, as those are
571 * always kernel segments while inside the kernel. Doing this
572 * before setting the new TLS descriptors avoids the situation
573 * where we temporarily have non-reloadable segments in %fs
574 * and %gs. This could be an issue if the NMI handler ever
575 * used %fs or %gs (it does not today), or if the kernel is
576 * running inside of a hypervisor layer.
578 savesegment(gs, prev->gs);
581 * Load the per-thread Thread-Local Storage descriptor.
586 * Restore IOPL if needed. In normal use, the flags restore
587 * in the switch assembly will handle this. But if the kernel
588 * is running virtualized at a non-zero CPL, the popf will
589 * not restore flags, so it must be done in a separate step.
591 if (get_kernel_rpl() && unlikely(prev->iopl != next->iopl))
592 set_iopl_mask(next->iopl);
595 * Now maybe handle debug registers and/or IO bitmaps
597 if (unlikely(task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV ||
598 task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT))
599 __switch_to_xtra(prev_p, next_p, tss);
602 * Leave lazy mode, flushing any hypercalls made here.
603 * This must be done before restoring TLS segments so
604 * the GDT and LDT are properly updated, and must be
605 * done before math_state_restore, so the TS bit is up
608 arch_leave_lazy_cpu_mode();
610 /* If the task has used fpu the last 5 timeslices, just do a full
611 * restore of the math state immediately to avoid the trap; the
612 * chances of needing FPU soon are obviously high now
614 * tsk_used_math() checks prevent calling math_state_restore(),
615 * which can sleep in the case of !tsk_used_math()
617 if (tsk_used_math(next_p) && next_p->fpu_counter > 5)
618 math_state_restore();
621 * Restore %gs if needed (which is common)
623 if (prev->gs | next->gs)
624 loadsegment(gs, next->gs);
626 x86_write_percpu(current_task, next_p);
631 asmlinkage int sys_fork(struct pt_regs regs)
633 return do_fork(SIGCHLD, regs.sp, ®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.bx;
644 parent_tidptr = (int __user *)regs.dx;
645 child_tidptr = (int __user *)regs.di;
648 return do_fork(clone_flags, 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.sp, ®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.bx);
675 error = PTR_ERR(filename);
676 if (IS_ERR(filename))
678 error = do_execve(filename,
679 (char __user * __user *) regs.cx,
680 (char __user * __user *) regs.dx,
683 /* Make sure we don't return using sysenter.. */
684 set_thread_flag(TIF_IRET);
691 #define top_esp (THREAD_SIZE - sizeof(unsigned long))
692 #define top_ebp (THREAD_SIZE - 2*sizeof(unsigned long))
694 unsigned long get_wchan(struct task_struct *p)
696 unsigned long bp, sp, ip;
697 unsigned long stack_page;
699 if (!p || p == current || p->state == TASK_RUNNING)
701 stack_page = (unsigned long)task_stack_page(p);
703 if (!stack_page || sp < stack_page || sp > top_esp+stack_page)
705 /* include/asm-i386/system.h:switch_to() pushes bp last. */
706 bp = *(unsigned long *) sp;
708 if (bp < stack_page || bp > top_ebp+stack_page)
710 ip = *(unsigned long *) (bp+4);
711 if (!in_sched_functions(ip))
713 bp = *(unsigned long *) bp;
714 } while (count++ < 16);
718 unsigned long arch_align_stack(unsigned long sp)
720 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
721 sp -= get_random_int() % 8192;
725 unsigned long arch_randomize_brk(struct mm_struct *mm)
727 unsigned long range_end = mm->brk + 0x02000000;
728 return randomize_range(mm->brk, range_end, 0) ? : mm->brk;