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>
40 #include <linux/dmi.h>
41 #include <linux/ftrace.h>
42 #include <linux/uaccess.h>
44 #include <linux/kdebug.h>
46 #include <asm/pgtable.h>
47 #include <asm/system.h>
49 #include <asm/processor.h>
52 #ifdef CONFIG_MATH_EMULATION
53 #include <asm/math_emu.h>
56 #include <linux/err.h>
58 #include <asm/tlbflush.h>
61 #include <asm/syscalls.h>
64 asmlinkage void ret_from_fork(void) __asm__("ret_from_fork");
66 DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task;
67 EXPORT_PER_CPU_SYMBOL(current_task);
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];
78 static inline void play_dead(void)
85 * The idle thread. There's no useful work to be
86 * done, so just try to conserve power and have a
87 * low exit latency (ie sit in a loop waiting for
88 * somebody to say that they'd like to reschedule)
92 int cpu = smp_processor_id();
94 current_thread_info()->status |= TS_POLLING;
96 /* endless idle loop with no priority at all */
98 tick_nohz_stop_sched_tick(1);
99 while (!need_resched()) {
104 if (rcu_pending(cpu))
105 rcu_check_callbacks(cpu, 0);
107 if (cpu_is_offline(cpu))
111 /* Don't trace irqs off for idle */
112 stop_critical_timings();
114 start_critical_timings();
116 tick_nohz_restart_sched_tick();
117 preempt_enable_no_resched();
123 void __show_regs(struct pt_regs *regs, int all)
125 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L;
126 unsigned long d0, d1, d2, d3, d6, d7;
128 unsigned short ss, gs;
131 if (user_mode_vm(regs)) {
133 ss = regs->ss & 0xffff;
136 sp = (unsigned long) (®s->sp);
143 board = dmi_get_system_info(DMI_PRODUCT_NAME);
146 printk("Pid: %d, comm: %s %s (%s %.*s) %s\n",
147 task_pid_nr(current), current->comm,
148 print_tainted(), init_utsname()->release,
149 (int)strcspn(init_utsname()->version, " "),
150 init_utsname()->version, board);
152 printk("EIP: %04x:[<%08lx>] EFLAGS: %08lx CPU: %d\n",
153 (u16)regs->cs, regs->ip, regs->flags,
155 print_symbol("EIP is at %s\n", regs->ip);
157 printk("EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
158 regs->ax, regs->bx, regs->cx, regs->dx);
159 printk("ESI: %08lx EDI: %08lx EBP: %08lx ESP: %08lx\n",
160 regs->si, regs->di, regs->bp, sp);
161 printk(" DS: %04x ES: %04x FS: %04x GS: %04x SS: %04x\n",
162 (u16)regs->ds, (u16)regs->es, (u16)regs->fs, gs, ss);
170 cr4 = read_cr4_safe();
171 printk("CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n",
178 printk("DR0: %08lx DR1: %08lx DR2: %08lx DR3: %08lx\n",
183 printk("DR6: %08lx DR7: %08lx\n",
187 void show_regs(struct pt_regs *regs)
189 __show_regs(regs, 1);
190 show_trace(NULL, regs, ®s->sp, regs->bp);
194 * This gets run with %bx containing the
195 * function to call, and %dx containing
198 extern void kernel_thread_helper(void);
201 * Create a kernel thread
203 int kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
207 memset(®s, 0, sizeof(regs));
209 regs.bx = (unsigned long) fn;
210 regs.dx = (unsigned long) arg;
214 regs.fs = __KERNEL_PERCPU;
216 regs.ip = (unsigned long) kernel_thread_helper;
217 regs.cs = __KERNEL_CS | get_kernel_rpl();
218 regs.flags = X86_EFLAGS_IF | X86_EFLAGS_SF | X86_EFLAGS_PF | 0x2;
220 /* Ok, create the new process.. */
221 return do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0, ®s, 0, NULL, NULL);
223 EXPORT_SYMBOL(kernel_thread);
226 * Free current thread data structures etc..
228 void exit_thread(void)
230 /* The process may have allocated an io port bitmap... nuke it. */
231 if (unlikely(test_thread_flag(TIF_IO_BITMAP))) {
232 struct task_struct *tsk = current;
233 struct thread_struct *t = &tsk->thread;
235 struct tss_struct *tss = &per_cpu(init_tss, cpu);
237 kfree(t->io_bitmap_ptr);
238 t->io_bitmap_ptr = NULL;
239 clear_thread_flag(TIF_IO_BITMAP);
241 * Careful, clear this in the TSS too:
243 memset(tss->io_bitmap, 0xff, tss->io_bitmap_max);
244 t->io_bitmap_max = 0;
245 tss->io_bitmap_owner = NULL;
246 tss->io_bitmap_max = 0;
247 tss->x86_tss.io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
251 ds_exit_thread(current);
254 void flush_thread(void)
256 struct task_struct *tsk = current;
258 tsk->thread.debugreg0 = 0;
259 tsk->thread.debugreg1 = 0;
260 tsk->thread.debugreg2 = 0;
261 tsk->thread.debugreg3 = 0;
262 tsk->thread.debugreg6 = 0;
263 tsk->thread.debugreg7 = 0;
264 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
265 clear_tsk_thread_flag(tsk, TIF_DEBUG);
267 * Forget coprocessor state..
269 tsk->fpu_counter = 0;
274 void release_thread(struct task_struct *dead_task)
276 BUG_ON(dead_task->mm);
277 release_vm86_irqs(dead_task);
281 * This gets called before we allocate a new thread and copy
282 * the current task into it.
284 void prepare_to_copy(struct task_struct *tsk)
289 int copy_thread(int nr, unsigned long clone_flags, unsigned long sp,
290 unsigned long unused,
291 struct task_struct *p, struct pt_regs *regs)
293 struct pt_regs *childregs;
294 struct task_struct *tsk;
297 childregs = task_pt_regs(p);
302 p->thread.sp = (unsigned long) childregs;
303 p->thread.sp0 = (unsigned long) (childregs+1);
305 p->thread.ip = (unsigned long) ret_from_fork;
307 savesegment(gs, p->thread.gs);
310 if (unlikely(test_tsk_thread_flag(tsk, TIF_IO_BITMAP))) {
311 p->thread.io_bitmap_ptr = kmemdup(tsk->thread.io_bitmap_ptr,
312 IO_BITMAP_BYTES, GFP_KERNEL);
313 if (!p->thread.io_bitmap_ptr) {
314 p->thread.io_bitmap_max = 0;
317 set_tsk_thread_flag(p, TIF_IO_BITMAP);
323 * Set a new TLS for the child thread?
325 if (clone_flags & CLONE_SETTLS)
326 err = do_set_thread_area(p, -1,
327 (struct user_desc __user *)childregs->si, 0);
329 if (err && p->thread.io_bitmap_ptr) {
330 kfree(p->thread.io_bitmap_ptr);
331 p->thread.io_bitmap_max = 0;
334 ds_copy_thread(p, current);
336 clear_tsk_thread_flag(p, TIF_DEBUGCTLMSR);
337 p->thread.debugctlmsr = 0;
343 start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp)
345 __asm__("movl %0, %%gs" : : "r"(0));
348 regs->ds = __USER_DS;
349 regs->es = __USER_DS;
350 regs->ss = __USER_DS;
351 regs->cs = __USER_CS;
355 * Free the old FP and other extended state
357 free_thread_xstate(current);
359 EXPORT_SYMBOL_GPL(start_thread);
361 static void hard_disable_TSC(void)
363 write_cr4(read_cr4() | X86_CR4_TSD);
366 void disable_TSC(void)
369 if (!test_and_set_thread_flag(TIF_NOTSC))
371 * Must flip the CPU state synchronously with
372 * TIF_NOTSC in the current running context.
378 static void hard_enable_TSC(void)
380 write_cr4(read_cr4() & ~X86_CR4_TSD);
383 static void enable_TSC(void)
386 if (test_and_clear_thread_flag(TIF_NOTSC))
388 * Must flip the CPU state synchronously with
389 * TIF_NOTSC in the current running context.
395 int get_tsc_mode(unsigned long adr)
399 if (test_thread_flag(TIF_NOTSC))
400 val = PR_TSC_SIGSEGV;
404 return put_user(val, (unsigned int __user *)adr);
407 int set_tsc_mode(unsigned int val)
409 if (val == PR_TSC_SIGSEGV)
411 else if (val == PR_TSC_ENABLE)
420 __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p,
421 struct tss_struct *tss)
423 struct thread_struct *prev, *next;
425 prev = &prev_p->thread;
426 next = &next_p->thread;
428 if (test_tsk_thread_flag(next_p, TIF_DS_AREA_MSR) ||
429 test_tsk_thread_flag(prev_p, TIF_DS_AREA_MSR))
430 ds_switch_to(prev_p, next_p);
431 else if (next->debugctlmsr != prev->debugctlmsr)
432 update_debugctlmsr(next->debugctlmsr);
434 if (test_tsk_thread_flag(next_p, TIF_DEBUG)) {
435 set_debugreg(next->debugreg0, 0);
436 set_debugreg(next->debugreg1, 1);
437 set_debugreg(next->debugreg2, 2);
438 set_debugreg(next->debugreg3, 3);
440 set_debugreg(next->debugreg6, 6);
441 set_debugreg(next->debugreg7, 7);
444 if (test_tsk_thread_flag(prev_p, TIF_NOTSC) ^
445 test_tsk_thread_flag(next_p, TIF_NOTSC)) {
446 /* prev and next are different */
447 if (test_tsk_thread_flag(next_p, TIF_NOTSC))
453 if (!test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
455 * Disable the bitmap via an invalid offset. We still cache
456 * the previous bitmap owner and the IO bitmap contents:
458 tss->x86_tss.io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
462 if (likely(next == tss->io_bitmap_owner)) {
464 * Previous owner of the bitmap (hence the bitmap content)
465 * matches the next task, we dont have to do anything but
466 * to set a valid offset in the TSS:
468 tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET;
472 * Lazy TSS's I/O bitmap copy. We set an invalid offset here
473 * and we let the task to get a GPF in case an I/O instruction
474 * is performed. The handler of the GPF will verify that the
475 * faulting task has a valid I/O bitmap and, it true, does the
476 * real copy and restart the instruction. This will save us
477 * redundant copies when the currently switched task does not
478 * perform any I/O during its timeslice.
480 tss->x86_tss.io_bitmap_base = INVALID_IO_BITMAP_OFFSET_LAZY;
484 * switch_to(x,yn) should switch tasks from x to y.
486 * We fsave/fwait so that an exception goes off at the right time
487 * (as a call from the fsave or fwait in effect) rather than to
488 * the wrong process. Lazy FP saving no longer makes any sense
489 * with modern CPU's, and this simplifies a lot of things (SMP
490 * and UP become the same).
492 * NOTE! We used to use the x86 hardware context switching. The
493 * reason for not using it any more becomes apparent when you
494 * try to recover gracefully from saved state that is no longer
495 * valid (stale segment register values in particular). With the
496 * hardware task-switch, there is no way to fix up bad state in
497 * a reasonable manner.
499 * The fact that Intel documents the hardware task-switching to
500 * be slow is a fairly red herring - this code is not noticeably
501 * faster. However, there _is_ some room for improvement here,
502 * so the performance issues may eventually be a valid point.
503 * More important, however, is the fact that this allows us much
506 * The return value (in %ax) will be the "prev" task after
507 * the task-switch, and shows up in ret_from_fork in entry.S,
510 __notrace_funcgraph struct task_struct *
511 __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
513 struct thread_struct *prev = &prev_p->thread,
514 *next = &next_p->thread;
515 int cpu = smp_processor_id();
516 struct tss_struct *tss = &per_cpu(init_tss, cpu);
518 /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
520 __unlazy_fpu(prev_p);
523 /* we're going to use this soon, after a few expensive things */
524 if (next_p->fpu_counter > 5)
525 prefetch(next->xstate);
533 * Save away %gs. No need to save %fs, as it was saved on the
534 * stack on entry. No need to save %es and %ds, as those are
535 * always kernel segments while inside the kernel. Doing this
536 * before setting the new TLS descriptors avoids the situation
537 * where we temporarily have non-reloadable segments in %fs
538 * and %gs. This could be an issue if the NMI handler ever
539 * used %fs or %gs (it does not today), or if the kernel is
540 * running inside of a hypervisor layer.
542 savesegment(gs, prev->gs);
545 * Load the per-thread Thread-Local Storage descriptor.
550 * Restore IOPL if needed. In normal use, the flags restore
551 * in the switch assembly will handle this. But if the kernel
552 * is running virtualized at a non-zero CPL, the popf will
553 * not restore flags, so it must be done in a separate step.
555 if (get_kernel_rpl() && unlikely(prev->iopl != next->iopl))
556 set_iopl_mask(next->iopl);
559 * Now maybe handle debug registers and/or IO bitmaps
561 if (unlikely(task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV ||
562 task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT))
563 __switch_to_xtra(prev_p, next_p, tss);
566 * Leave lazy mode, flushing any hypercalls made here.
567 * This must be done before restoring TLS segments so
568 * the GDT and LDT are properly updated, and must be
569 * done before math_state_restore, so the TS bit is up
572 arch_leave_lazy_cpu_mode();
574 /* If the task has used fpu the last 5 timeslices, just do a full
575 * restore of the math state immediately to avoid the trap; the
576 * chances of needing FPU soon are obviously high now
578 * tsk_used_math() checks prevent calling math_state_restore(),
579 * which can sleep in the case of !tsk_used_math()
581 if (tsk_used_math(next_p) && next_p->fpu_counter > 5)
582 math_state_restore();
585 * Restore %gs if needed (which is common)
587 if (prev->gs | next->gs)
588 loadsegment(gs, next->gs);
590 percpu_write(current_task, next_p);
595 asmlinkage int sys_fork(struct pt_regs regs)
597 return do_fork(SIGCHLD, regs.sp, ®s, 0, NULL, NULL);
600 asmlinkage int sys_clone(struct pt_regs regs)
602 unsigned long clone_flags;
604 int __user *parent_tidptr, *child_tidptr;
606 clone_flags = regs.bx;
608 parent_tidptr = (int __user *)regs.dx;
609 child_tidptr = (int __user *)regs.di;
612 return do_fork(clone_flags, newsp, ®s, 0, parent_tidptr, child_tidptr);
616 * This is trivial, and on the face of it looks like it
617 * could equally well be done in user mode.
619 * Not so, for quite unobvious reasons - register pressure.
620 * In user mode vfork() cannot have a stack frame, and if
621 * done by calling the "clone()" system call directly, you
622 * do not have enough call-clobbered registers to hold all
623 * the information you need.
625 asmlinkage int sys_vfork(struct pt_regs regs)
627 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs.sp, ®s, 0, NULL, NULL);
631 * sys_execve() executes a new program.
633 asmlinkage int sys_execve(struct pt_regs regs)
638 filename = getname((char __user *) regs.bx);
639 error = PTR_ERR(filename);
640 if (IS_ERR(filename))
642 error = do_execve(filename,
643 (char __user * __user *) regs.cx,
644 (char __user * __user *) regs.dx,
647 /* Make sure we don't return using sysenter.. */
648 set_thread_flag(TIF_IRET);
655 #define top_esp (THREAD_SIZE - sizeof(unsigned long))
656 #define top_ebp (THREAD_SIZE - 2*sizeof(unsigned long))
658 unsigned long get_wchan(struct task_struct *p)
660 unsigned long bp, sp, ip;
661 unsigned long stack_page;
663 if (!p || p == current || p->state == TASK_RUNNING)
665 stack_page = (unsigned long)task_stack_page(p);
667 if (!stack_page || sp < stack_page || sp > top_esp+stack_page)
669 /* include/asm-i386/system.h:switch_to() pushes bp last. */
670 bp = *(unsigned long *) sp;
672 if (bp < stack_page || bp > top_ebp+stack_page)
674 ip = *(unsigned long *) (bp+4);
675 if (!in_sched_functions(ip))
677 bp = *(unsigned long *) bp;
678 } while (count++ < 16);
682 unsigned long arch_align_stack(unsigned long sp)
684 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
685 sp -= get_random_int() % 8192;
689 unsigned long arch_randomize_brk(struct mm_struct *mm)
691 unsigned long range_end = mm->brk + 0x02000000;
692 return randomize_range(mm->brk, range_end, 0) ? : mm->brk;