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 */
100 __get_cpu_var(cpu_state) = CPU_DEAD;
103 * With physical CPU hotplug, we should halt the cpu
106 /* mask all interrupts, flush any and all caches, and halt */
110 static inline void play_dead(void)
114 #endif /* CONFIG_HOTPLUG_CPU */
117 * The idle thread. There's no useful work to be
118 * done, so just try to conserve power and have a
119 * low exit latency (ie sit in a loop waiting for
120 * somebody to say that they'd like to reschedule)
124 int cpu = smp_processor_id();
126 current_thread_info()->status |= TS_POLLING;
128 /* endless idle loop with no priority at all */
130 tick_nohz_stop_sched_tick(1);
131 while (!need_resched()) {
136 if (rcu_pending(cpu))
137 rcu_check_callbacks(cpu, 0);
139 if (cpu_is_offline(cpu))
143 __get_cpu_var(irq_stat).idle_timestamp = jiffies;
144 /* Don't trace irqs off for idle */
145 stop_critical_timings();
147 start_critical_timings();
149 tick_nohz_restart_sched_tick();
150 preempt_enable_no_resched();
156 void __show_registers(struct pt_regs *regs, int all)
158 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L;
159 unsigned long d0, d1, d2, d3, d6, d7;
161 unsigned short ss, gs;
163 if (user_mode_vm(regs)) {
165 ss = regs->ss & 0xffff;
168 sp = (unsigned long) (®s->sp);
174 printk("Pid: %d, comm: %s %s (%s %.*s)\n",
175 task_pid_nr(current), current->comm,
176 print_tainted(), init_utsname()->release,
177 (int)strcspn(init_utsname()->version, " "),
178 init_utsname()->version);
180 printk("EIP: %04x:[<%08lx>] EFLAGS: %08lx CPU: %d\n",
181 (u16)regs->cs, regs->ip, regs->flags,
183 print_symbol("EIP is at %s\n", regs->ip);
185 printk("EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
186 regs->ax, regs->bx, regs->cx, regs->dx);
187 printk("ESI: %08lx EDI: %08lx EBP: %08lx ESP: %08lx\n",
188 regs->si, regs->di, regs->bp, sp);
189 printk(" DS: %04x ES: %04x FS: %04x GS: %04x SS: %04x\n",
190 (u16)regs->ds, (u16)regs->es, (u16)regs->fs, gs, ss);
198 cr4 = read_cr4_safe();
199 printk("CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n",
206 printk("DR0: %08lx DR1: %08lx DR2: %08lx DR3: %08lx\n",
211 printk("DR6: %08lx DR7: %08lx\n",
215 void show_regs(struct pt_regs *regs)
217 __show_registers(regs, 1);
218 show_trace(NULL, regs, ®s->sp, regs->bp);
222 * This gets run with %bx containing the
223 * function to call, and %dx containing
226 extern void kernel_thread_helper(void);
229 * Create a kernel thread
231 int kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
235 memset(®s, 0, sizeof(regs));
237 regs.bx = (unsigned long) fn;
238 regs.dx = (unsigned long) arg;
242 regs.fs = __KERNEL_PERCPU;
244 regs.ip = (unsigned long) kernel_thread_helper;
245 regs.cs = __KERNEL_CS | get_kernel_rpl();
246 regs.flags = X86_EFLAGS_IF | X86_EFLAGS_SF | X86_EFLAGS_PF | 0x2;
248 /* Ok, create the new process.. */
249 return do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0, ®s, 0, NULL, NULL);
251 EXPORT_SYMBOL(kernel_thread);
254 * Free current thread data structures etc..
256 void exit_thread(void)
258 /* The process may have allocated an io port bitmap... nuke it. */
259 if (unlikely(test_thread_flag(TIF_IO_BITMAP))) {
260 struct task_struct *tsk = current;
261 struct thread_struct *t = &tsk->thread;
263 struct tss_struct *tss = &per_cpu(init_tss, cpu);
265 kfree(t->io_bitmap_ptr);
266 t->io_bitmap_ptr = NULL;
267 clear_thread_flag(TIF_IO_BITMAP);
269 * Careful, clear this in the TSS too:
271 memset(tss->io_bitmap, 0xff, tss->io_bitmap_max);
272 t->io_bitmap_max = 0;
273 tss->io_bitmap_owner = NULL;
274 tss->io_bitmap_max = 0;
275 tss->x86_tss.io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
280 void flush_thread(void)
282 struct task_struct *tsk = current;
284 tsk->thread.debugreg0 = 0;
285 tsk->thread.debugreg1 = 0;
286 tsk->thread.debugreg2 = 0;
287 tsk->thread.debugreg3 = 0;
288 tsk->thread.debugreg6 = 0;
289 tsk->thread.debugreg7 = 0;
290 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
291 clear_tsk_thread_flag(tsk, TIF_DEBUG);
293 * Forget coprocessor state..
295 tsk->fpu_counter = 0;
300 void release_thread(struct task_struct *dead_task)
302 BUG_ON(dead_task->mm);
303 release_vm86_irqs(dead_task);
307 * This gets called before we allocate a new thread and copy
308 * the current task into it.
310 void prepare_to_copy(struct task_struct *tsk)
315 int copy_thread(int nr, unsigned long clone_flags, unsigned long sp,
316 unsigned long unused,
317 struct task_struct * p, struct pt_regs * regs)
319 struct pt_regs * childregs;
320 struct task_struct *tsk;
323 childregs = task_pt_regs(p);
328 p->thread.sp = (unsigned long) childregs;
329 p->thread.sp0 = (unsigned long) (childregs+1);
331 p->thread.ip = (unsigned long) ret_from_fork;
333 savesegment(gs, p->thread.gs);
336 if (unlikely(test_tsk_thread_flag(tsk, TIF_IO_BITMAP))) {
337 p->thread.io_bitmap_ptr = kmemdup(tsk->thread.io_bitmap_ptr,
338 IO_BITMAP_BYTES, GFP_KERNEL);
339 if (!p->thread.io_bitmap_ptr) {
340 p->thread.io_bitmap_max = 0;
343 set_tsk_thread_flag(p, TIF_IO_BITMAP);
349 * Set a new TLS for the child thread?
351 if (clone_flags & CLONE_SETTLS)
352 err = do_set_thread_area(p, -1,
353 (struct user_desc __user *)childregs->si, 0);
355 if (err && p->thread.io_bitmap_ptr) {
356 kfree(p->thread.io_bitmap_ptr);
357 p->thread.io_bitmap_max = 0;
363 start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp)
365 __asm__("movl %0, %%gs" :: "r"(0));
368 regs->ds = __USER_DS;
369 regs->es = __USER_DS;
370 regs->ss = __USER_DS;
371 regs->cs = __USER_CS;
375 * Free the old FP and other extended state
377 free_thread_xstate(current);
379 EXPORT_SYMBOL_GPL(start_thread);
381 static void hard_disable_TSC(void)
383 write_cr4(read_cr4() | X86_CR4_TSD);
386 void disable_TSC(void)
389 if (!test_and_set_thread_flag(TIF_NOTSC))
391 * Must flip the CPU state synchronously with
392 * TIF_NOTSC in the current running context.
398 static void hard_enable_TSC(void)
400 write_cr4(read_cr4() & ~X86_CR4_TSD);
403 static void enable_TSC(void)
406 if (test_and_clear_thread_flag(TIF_NOTSC))
408 * Must flip the CPU state synchronously with
409 * TIF_NOTSC in the current running context.
415 int get_tsc_mode(unsigned long adr)
419 if (test_thread_flag(TIF_NOTSC))
420 val = PR_TSC_SIGSEGV;
424 return put_user(val, (unsigned int __user *)adr);
427 int set_tsc_mode(unsigned int val)
429 if (val == PR_TSC_SIGSEGV)
431 else if (val == PR_TSC_ENABLE)
440 __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p,
441 struct tss_struct *tss)
443 struct thread_struct *prev, *next;
444 unsigned long debugctl;
446 prev = &prev_p->thread;
447 next = &next_p->thread;
449 debugctl = prev->debugctlmsr;
450 if (next->ds_area_msr != prev->ds_area_msr) {
451 /* we clear debugctl to make sure DS
452 * is not in use when we change it */
454 update_debugctlmsr(0);
455 wrmsr(MSR_IA32_DS_AREA, next->ds_area_msr, 0);
458 if (next->debugctlmsr != debugctl)
459 update_debugctlmsr(next->debugctlmsr);
461 if (test_tsk_thread_flag(next_p, TIF_DEBUG)) {
462 set_debugreg(next->debugreg0, 0);
463 set_debugreg(next->debugreg1, 1);
464 set_debugreg(next->debugreg2, 2);
465 set_debugreg(next->debugreg3, 3);
467 set_debugreg(next->debugreg6, 6);
468 set_debugreg(next->debugreg7, 7);
471 if (test_tsk_thread_flag(prev_p, TIF_NOTSC) ^
472 test_tsk_thread_flag(next_p, TIF_NOTSC)) {
473 /* prev and next are different */
474 if (test_tsk_thread_flag(next_p, TIF_NOTSC))
481 if (test_tsk_thread_flag(prev_p, TIF_BTS_TRACE_TS))
482 ptrace_bts_take_timestamp(prev_p, BTS_TASK_DEPARTS);
484 if (test_tsk_thread_flag(next_p, TIF_BTS_TRACE_TS))
485 ptrace_bts_take_timestamp(next_p, BTS_TASK_ARRIVES);
489 if (!test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
491 * Disable the bitmap via an invalid offset. We still cache
492 * the previous bitmap owner and the IO bitmap contents:
494 tss->x86_tss.io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
498 if (likely(next == tss->io_bitmap_owner)) {
500 * Previous owner of the bitmap (hence the bitmap content)
501 * matches the next task, we dont have to do anything but
502 * to set a valid offset in the TSS:
504 tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET;
508 * Lazy TSS's I/O bitmap copy. We set an invalid offset here
509 * and we let the task to get a GPF in case an I/O instruction
510 * is performed. The handler of the GPF will verify that the
511 * faulting task has a valid I/O bitmap and, it true, does the
512 * real copy and restart the instruction. This will save us
513 * redundant copies when the currently switched task does not
514 * perform any I/O during its timeslice.
516 tss->x86_tss.io_bitmap_base = INVALID_IO_BITMAP_OFFSET_LAZY;
520 * switch_to(x,yn) should switch tasks from x to y.
522 * We fsave/fwait so that an exception goes off at the right time
523 * (as a call from the fsave or fwait in effect) rather than to
524 * the wrong process. Lazy FP saving no longer makes any sense
525 * with modern CPU's, and this simplifies a lot of things (SMP
526 * and UP become the same).
528 * NOTE! We used to use the x86 hardware context switching. The
529 * reason for not using it any more becomes apparent when you
530 * try to recover gracefully from saved state that is no longer
531 * valid (stale segment register values in particular). With the
532 * hardware task-switch, there is no way to fix up bad state in
533 * a reasonable manner.
535 * The fact that Intel documents the hardware task-switching to
536 * be slow is a fairly red herring - this code is not noticeably
537 * faster. However, there _is_ some room for improvement here,
538 * so the performance issues may eventually be a valid point.
539 * More important, however, is the fact that this allows us much
542 * The return value (in %ax) will be the "prev" task after
543 * the task-switch, and shows up in ret_from_fork in entry.S,
546 struct task_struct * __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
548 struct thread_struct *prev = &prev_p->thread,
549 *next = &next_p->thread;
550 int cpu = smp_processor_id();
551 struct tss_struct *tss = &per_cpu(init_tss, cpu);
553 /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
555 __unlazy_fpu(prev_p);
558 /* we're going to use this soon, after a few expensive things */
559 if (next_p->fpu_counter > 5)
560 prefetch(next->xstate);
568 * Save away %gs. No need to save %fs, as it was saved on the
569 * stack on entry. No need to save %es and %ds, as those are
570 * always kernel segments while inside the kernel. Doing this
571 * before setting the new TLS descriptors avoids the situation
572 * where we temporarily have non-reloadable segments in %fs
573 * and %gs. This could be an issue if the NMI handler ever
574 * used %fs or %gs (it does not today), or if the kernel is
575 * running inside of a hypervisor layer.
577 savesegment(gs, prev->gs);
580 * Load the per-thread Thread-Local Storage descriptor.
585 * Restore IOPL if needed. In normal use, the flags restore
586 * in the switch assembly will handle this. But if the kernel
587 * is running virtualized at a non-zero CPL, the popf will
588 * not restore flags, so it must be done in a separate step.
590 if (get_kernel_rpl() && unlikely(prev->iopl != next->iopl))
591 set_iopl_mask(next->iopl);
594 * Now maybe handle debug registers and/or IO bitmaps
596 if (unlikely(task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV ||
597 task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT))
598 __switch_to_xtra(prev_p, next_p, tss);
601 * Leave lazy mode, flushing any hypercalls made here.
602 * This must be done before restoring TLS segments so
603 * the GDT and LDT are properly updated, and must be
604 * done before math_state_restore, so the TS bit is up
607 arch_leave_lazy_cpu_mode();
609 /* If the task has used fpu the last 5 timeslices, just do a full
610 * restore of the math state immediately to avoid the trap; the
611 * chances of needing FPU soon are obviously high now
613 * tsk_used_math() checks prevent calling math_state_restore(),
614 * which can sleep in the case of !tsk_used_math()
616 if (tsk_used_math(next_p) && next_p->fpu_counter > 5)
617 math_state_restore();
620 * Restore %gs if needed (which is common)
622 if (prev->gs | next->gs)
623 loadsegment(gs, next->gs);
625 x86_write_percpu(current_task, next_p);
630 asmlinkage int sys_fork(struct pt_regs regs)
632 return do_fork(SIGCHLD, regs.sp, ®s, 0, NULL, NULL);
635 asmlinkage int sys_clone(struct pt_regs regs)
637 unsigned long clone_flags;
639 int __user *parent_tidptr, *child_tidptr;
641 clone_flags = regs.bx;
643 parent_tidptr = (int __user *)regs.dx;
644 child_tidptr = (int __user *)regs.di;
647 return do_fork(clone_flags, newsp, ®s, 0, parent_tidptr, child_tidptr);
651 * This is trivial, and on the face of it looks like it
652 * could equally well be done in user mode.
654 * Not so, for quite unobvious reasons - register pressure.
655 * In user mode vfork() cannot have a stack frame, and if
656 * done by calling the "clone()" system call directly, you
657 * do not have enough call-clobbered registers to hold all
658 * the information you need.
660 asmlinkage int sys_vfork(struct pt_regs regs)
662 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs.sp, ®s, 0, NULL, NULL);
666 * sys_execve() executes a new program.
668 asmlinkage int sys_execve(struct pt_regs regs)
673 filename = getname((char __user *) regs.bx);
674 error = PTR_ERR(filename);
675 if (IS_ERR(filename))
677 error = do_execve(filename,
678 (char __user * __user *) regs.cx,
679 (char __user * __user *) regs.dx,
682 /* Make sure we don't return using sysenter.. */
683 set_thread_flag(TIF_IRET);
690 #define top_esp (THREAD_SIZE - sizeof(unsigned long))
691 #define top_ebp (THREAD_SIZE - 2*sizeof(unsigned long))
693 unsigned long get_wchan(struct task_struct *p)
695 unsigned long bp, sp, ip;
696 unsigned long stack_page;
698 if (!p || p == current || p->state == TASK_RUNNING)
700 stack_page = (unsigned long)task_stack_page(p);
702 if (!stack_page || sp < stack_page || sp > top_esp+stack_page)
704 /* include/asm-i386/system.h:switch_to() pushes bp last. */
705 bp = *(unsigned long *) sp;
707 if (bp < stack_page || bp > top_ebp+stack_page)
709 ip = *(unsigned long *) (bp+4);
710 if (!in_sched_functions(ip))
712 bp = *(unsigned long *) bp;
713 } while (count++ < 16);
717 unsigned long arch_align_stack(unsigned long sp)
719 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
720 sp -= get_random_int() % 8192;
724 unsigned long arch_randomize_brk(struct mm_struct *mm)
726 unsigned long range_end = mm->brk + 0x02000000;
727 return randomize_range(mm->brk, range_end, 0) ? : mm->brk;