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/stackprotector.h>
15 #include <linux/cpu.h>
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/stddef.h>
24 #include <linux/slab.h>
25 #include <linux/vmalloc.h>
26 #include <linux/user.h>
27 #include <linux/interrupt.h>
28 #include <linux/utsname.h>
29 #include <linux/delay.h>
30 #include <linux/reboot.h>
31 #include <linux/init.h>
32 #include <linux/mc146818rtc.h>
33 #include <linux/module.h>
34 #include <linux/kallsyms.h>
35 #include <linux/ptrace.h>
36 #include <linux/random.h>
37 #include <linux/personality.h>
38 #include <linux/tick.h>
39 #include <linux/percpu.h>
40 #include <linux/prctl.h>
41 #include <linux/dmi.h>
42 #include <linux/ftrace.h>
43 #include <linux/uaccess.h>
45 #include <linux/kdebug.h>
47 #include <asm/pgtable.h>
48 #include <asm/system.h>
50 #include <asm/processor.h>
53 #ifdef CONFIG_MATH_EMULATION
54 #include <asm/math_emu.h>
57 #include <linux/err.h>
59 #include <asm/tlbflush.h>
62 #include <asm/syscalls.h>
64 #include <asm/debugreg.h>
65 #include <asm/hw_breakpoint.h>
67 asmlinkage void ret_from_fork(void) __asm__("ret_from_fork");
69 DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task;
70 EXPORT_PER_CPU_SYMBOL(current_task);
73 * Return saved PC of a blocked thread.
75 unsigned long thread_saved_pc(struct task_struct *tsk)
77 return ((unsigned long *)tsk->thread.sp)[3];
81 static inline void play_dead(void)
88 * The idle thread. There's no useful work to be
89 * done, so just try to conserve power and have a
90 * low exit latency (ie sit in a loop waiting for
91 * somebody to say that they'd like to reschedule)
95 int cpu = smp_processor_id();
98 * If we're the non-boot CPU, nothing set the stack canary up
99 * for us. CPU0 already has it initialized but no harm in
100 * doing it again. This is a good place for updating it, as
101 * we wont ever return from this function (so the invalid
102 * canaries already on the stack wont ever trigger).
104 boot_init_stack_canary();
106 current_thread_info()->status |= TS_POLLING;
108 /* endless idle loop with no priority at all */
110 tick_nohz_stop_sched_tick(1);
111 while (!need_resched()) {
116 if (cpu_is_offline(cpu))
120 /* Don't trace irqs off for idle */
121 stop_critical_timings();
123 start_critical_timings();
125 tick_nohz_restart_sched_tick();
126 preempt_enable_no_resched();
132 void __show_regs(struct pt_regs *regs, int all)
134 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L;
135 unsigned long d0, d1, d2, d3, d6, d7;
137 unsigned short ss, gs;
140 if (user_mode_vm(regs)) {
142 ss = regs->ss & 0xffff;
143 gs = get_user_gs(regs);
145 sp = (unsigned long) (®s->sp);
152 board = dmi_get_system_info(DMI_PRODUCT_NAME);
155 printk("Pid: %d, comm: %s %s (%s %.*s) %s\n",
156 task_pid_nr(current), current->comm,
157 print_tainted(), init_utsname()->release,
158 (int)strcspn(init_utsname()->version, " "),
159 init_utsname()->version, board);
161 printk("EIP: %04x:[<%08lx>] EFLAGS: %08lx CPU: %d\n",
162 (u16)regs->cs, regs->ip, regs->flags,
164 print_symbol("EIP is at %s\n", regs->ip);
166 printk("EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
167 regs->ax, regs->bx, regs->cx, regs->dx);
168 printk("ESI: %08lx EDI: %08lx EBP: %08lx ESP: %08lx\n",
169 regs->si, regs->di, regs->bp, sp);
170 printk(" DS: %04x ES: %04x FS: %04x GS: %04x SS: %04x\n",
171 (u16)regs->ds, (u16)regs->es, (u16)regs->fs, gs, ss);
179 cr4 = read_cr4_safe();
180 printk("CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n",
187 printk("DR0: %08lx DR1: %08lx DR2: %08lx DR3: %08lx\n",
192 printk("DR6: %08lx DR7: %08lx\n",
196 void show_regs(struct pt_regs *regs)
198 __show_regs(regs, 1);
199 show_trace(NULL, regs, ®s->sp, regs->bp);
203 * This gets run with %bx containing the
204 * function to call, and %dx containing
207 extern void kernel_thread_helper(void);
210 * Create a kernel thread
212 int kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
216 memset(®s, 0, sizeof(regs));
218 regs.bx = (unsigned long) fn;
219 regs.dx = (unsigned long) arg;
223 regs.fs = __KERNEL_PERCPU;
224 regs.gs = __KERNEL_STACK_CANARY;
226 regs.ip = (unsigned long) kernel_thread_helper;
227 regs.cs = __KERNEL_CS | get_kernel_rpl();
228 regs.flags = X86_EFLAGS_IF | X86_EFLAGS_SF | X86_EFLAGS_PF | 0x2;
230 /* Ok, create the new process.. */
231 return do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0, ®s, 0, NULL, NULL);
233 EXPORT_SYMBOL(kernel_thread);
235 void release_thread(struct task_struct *dead_task)
237 BUG_ON(dead_task->mm);
238 release_vm86_irqs(dead_task);
242 * This gets called before we allocate a new thread and copy
243 * the current task into it.
245 void prepare_to_copy(struct task_struct *tsk)
250 int copy_thread(unsigned long clone_flags, unsigned long sp,
251 unsigned long unused,
252 struct task_struct *p, struct pt_regs *regs)
254 struct pt_regs *childregs;
255 struct task_struct *tsk;
258 childregs = task_pt_regs(p);
263 p->thread.sp = (unsigned long) childregs;
264 p->thread.sp0 = (unsigned long) (childregs+1);
266 p->thread.ip = (unsigned long) ret_from_fork;
268 task_user_gs(p) = get_user_gs(regs);
270 p->thread.io_bitmap_ptr = NULL;
273 if (unlikely(test_tsk_thread_flag(tsk, TIF_DEBUG)))
274 if (copy_thread_hw_breakpoint(tsk, p, clone_flags))
277 if (unlikely(test_tsk_thread_flag(tsk, TIF_IO_BITMAP))) {
278 p->thread.io_bitmap_ptr = kmemdup(tsk->thread.io_bitmap_ptr,
279 IO_BITMAP_BYTES, GFP_KERNEL);
280 if (!p->thread.io_bitmap_ptr) {
281 p->thread.io_bitmap_max = 0;
284 set_tsk_thread_flag(p, TIF_IO_BITMAP);
290 * Set a new TLS for the child thread?
292 if (clone_flags & CLONE_SETTLS)
293 err = do_set_thread_area(p, -1,
294 (struct user_desc __user *)childregs->si, 0);
297 if (err && p->thread.io_bitmap_ptr) {
298 kfree(p->thread.io_bitmap_ptr);
299 p->thread.io_bitmap_max = 0;
302 flush_thread_hw_breakpoint(p);
304 clear_tsk_thread_flag(p, TIF_DS_AREA_MSR);
305 p->thread.ds_ctx = NULL;
307 clear_tsk_thread_flag(p, TIF_DEBUGCTLMSR);
308 p->thread.debugctlmsr = 0;
314 start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp)
316 set_user_gs(regs, 0);
319 regs->ds = __USER_DS;
320 regs->es = __USER_DS;
321 regs->ss = __USER_DS;
322 regs->cs = __USER_CS;
326 * Free the old FP and other extended state
328 free_thread_xstate(current);
330 EXPORT_SYMBOL_GPL(start_thread);
334 * switch_to(x,yn) should switch tasks from x to y.
336 * We fsave/fwait so that an exception goes off at the right time
337 * (as a call from the fsave or fwait in effect) rather than to
338 * the wrong process. Lazy FP saving no longer makes any sense
339 * with modern CPU's, and this simplifies a lot of things (SMP
340 * and UP become the same).
342 * NOTE! We used to use the x86 hardware context switching. The
343 * reason for not using it any more becomes apparent when you
344 * try to recover gracefully from saved state that is no longer
345 * valid (stale segment register values in particular). With the
346 * hardware task-switch, there is no way to fix up bad state in
347 * a reasonable manner.
349 * The fact that Intel documents the hardware task-switching to
350 * be slow is a fairly red herring - this code is not noticeably
351 * faster. However, there _is_ some room for improvement here,
352 * so the performance issues may eventually be a valid point.
353 * More important, however, is the fact that this allows us much
356 * The return value (in %ax) will be the "prev" task after
357 * the task-switch, and shows up in ret_from_fork in entry.S,
360 __notrace_funcgraph struct task_struct *
361 __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
363 struct thread_struct *prev = &prev_p->thread,
364 *next = &next_p->thread;
365 int cpu = smp_processor_id();
366 struct tss_struct *tss = &per_cpu(init_tss, cpu);
368 /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
370 __unlazy_fpu(prev_p);
373 /* we're going to use this soon, after a few expensive things */
374 if (next_p->fpu_counter > 5)
375 prefetch(next->xstate);
383 * Save away %gs. No need to save %fs, as it was saved on the
384 * stack on entry. No need to save %es and %ds, as those are
385 * always kernel segments while inside the kernel. Doing this
386 * before setting the new TLS descriptors avoids the situation
387 * where we temporarily have non-reloadable segments in %fs
388 * and %gs. This could be an issue if the NMI handler ever
389 * used %fs or %gs (it does not today), or if the kernel is
390 * running inside of a hypervisor layer.
392 lazy_save_gs(prev->gs);
395 * Load the per-thread Thread-Local Storage descriptor.
400 * Restore IOPL if needed. In normal use, the flags restore
401 * in the switch assembly will handle this. But if the kernel
402 * is running virtualized at a non-zero CPL, the popf will
403 * not restore flags, so it must be done in a separate step.
405 if (get_kernel_rpl() && unlikely(prev->iopl != next->iopl))
406 set_iopl_mask(next->iopl);
409 * Now maybe handle debug registers and/or IO bitmaps
411 if (unlikely(task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV ||
412 task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT))
413 __switch_to_xtra(prev_p, next_p, tss);
416 * Leave lazy mode, flushing any hypercalls made here.
417 * This must be done before restoring TLS segments so
418 * the GDT and LDT are properly updated, and must be
419 * done before math_state_restore, so the TS bit is up
422 arch_leave_lazy_cpu_mode();
424 /* If the task has used fpu the last 5 timeslices, just do a full
425 * restore of the math state immediately to avoid the trap; the
426 * chances of needing FPU soon are obviously high now
428 * tsk_used_math() checks prevent calling math_state_restore(),
429 * which can sleep in the case of !tsk_used_math()
431 if (tsk_used_math(next_p) && next_p->fpu_counter > 5)
432 math_state_restore();
435 * Restore %gs if needed (which is common)
437 if (prev->gs | next->gs)
438 lazy_load_gs(next->gs);
440 percpu_write(current_task, next_p);
442 * There's a problem with moving the arch_install_thread_hw_breakpoint()
443 * call before current is updated. Suppose a kernel breakpoint is
444 * triggered in between the two, the hw-breakpoint handler will see that
445 * the 'current' task does not have TIF_DEBUG flag set and will think it
446 * is leftover from an old task (lazy switching) and will erase it. Then
447 * until the next context switch, no user-breakpoints will be installed.
449 * The real problem is that it's impossible to update both current and
450 * physical debug registers at the same instant, so there will always be
451 * a window in which they disagree and a breakpoint might get triggered.
452 * Since we use lazy switching, we are forced to assume that a
453 * disagreement means that current is correct and the exception is due
454 * to lazy debug register switching.
456 if (unlikely(test_tsk_thread_flag(next_p, TIF_DEBUG)))
457 arch_install_thread_hw_breakpoint(next_p);
462 int sys_clone(struct pt_regs *regs)
464 unsigned long clone_flags;
466 int __user *parent_tidptr, *child_tidptr;
468 clone_flags = regs->bx;
470 parent_tidptr = (int __user *)regs->dx;
471 child_tidptr = (int __user *)regs->di;
474 return do_fork(clone_flags, newsp, regs, 0, parent_tidptr, child_tidptr);
478 * sys_execve() executes a new program.
480 int sys_execve(struct pt_regs *regs)
485 filename = getname((char __user *) regs->bx);
486 error = PTR_ERR(filename);
487 if (IS_ERR(filename))
489 error = do_execve(filename,
490 (char __user * __user *) regs->cx,
491 (char __user * __user *) regs->dx,
494 /* Make sure we don't return using sysenter.. */
495 set_thread_flag(TIF_IRET);
502 #define top_esp (THREAD_SIZE - sizeof(unsigned long))
503 #define top_ebp (THREAD_SIZE - 2*sizeof(unsigned long))
505 unsigned long get_wchan(struct task_struct *p)
507 unsigned long bp, sp, ip;
508 unsigned long stack_page;
510 if (!p || p == current || p->state == TASK_RUNNING)
512 stack_page = (unsigned long)task_stack_page(p);
514 if (!stack_page || sp < stack_page || sp > top_esp+stack_page)
516 /* include/asm-i386/system.h:switch_to() pushes bp last. */
517 bp = *(unsigned long *) sp;
519 if (bp < stack_page || bp > top_ebp+stack_page)
521 ip = *(unsigned long *) (bp+4);
522 if (!in_sched_functions(ip))
524 bp = *(unsigned long *) bp;
525 } while (count++ < 16);
529 unsigned long arch_align_stack(unsigned long sp)
531 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
532 sp -= get_random_int() % 8192;
536 unsigned long arch_randomize_brk(struct mm_struct *mm)
538 unsigned long range_end = mm->brk + 0x02000000;
539 return randomize_range(mm->brk, range_end, 0) ? : mm->brk;