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>
42 #include <asm/uaccess.h>
43 #include <asm/pgtable.h>
44 #include <asm/system.h>
47 #include <asm/processor.h>
50 #ifdef CONFIG_MATH_EMULATION
51 #include <asm/math_emu.h>
54 #include <linux/err.h>
56 #include <asm/tlbflush.h>
58 #include <asm/kdebug.h>
60 #include <asm/syscalls.h>
63 asmlinkage void ret_from_fork(void) __asm__("ret_from_fork");
65 DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task;
66 EXPORT_PER_CPU_SYMBOL(current_task);
68 DEFINE_PER_CPU(int, cpu_number);
69 EXPORT_PER_CPU_SYMBOL(cpu_number);
72 * Return saved PC of a blocked thread.
74 unsigned long thread_saved_pc(struct task_struct *tsk)
76 return ((unsigned long *)tsk->thread.sp)[3];
80 static inline void play_dead(void)
87 * The idle thread. There's no useful work to be
88 * done, so just try to conserve power and have a
89 * low exit latency (ie sit in a loop waiting for
90 * somebody to say that they'd like to reschedule)
94 int cpu = smp_processor_id();
96 current_thread_info()->status |= TS_POLLING;
98 /* endless idle loop with no priority at all */
100 tick_nohz_stop_sched_tick(1);
101 while (!need_resched()) {
106 if (rcu_pending(cpu))
107 rcu_check_callbacks(cpu, 0);
109 if (cpu_is_offline(cpu))
113 __get_cpu_var(irq_stat).idle_timestamp = jiffies;
114 /* Don't trace irqs off for idle */
115 stop_critical_timings();
117 start_critical_timings();
119 tick_nohz_restart_sched_tick();
120 preempt_enable_no_resched();
126 void __show_regs(struct pt_regs *regs, int all)
128 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L;
129 unsigned long d0, d1, d2, d3, d6, d7;
131 unsigned short ss, gs;
134 if (user_mode_vm(regs)) {
136 ss = regs->ss & 0xffff;
139 sp = (unsigned long) (®s->sp);
146 board = dmi_get_system_info(DMI_PRODUCT_NAME);
149 printk("Pid: %d, comm: %s %s (%s %.*s) %s\n",
150 task_pid_nr(current), current->comm,
151 print_tainted(), init_utsname()->release,
152 (int)strcspn(init_utsname()->version, " "),
153 init_utsname()->version, board);
155 printk("EIP: %04x:[<%08lx>] EFLAGS: %08lx CPU: %d\n",
156 (u16)regs->cs, regs->ip, regs->flags,
158 print_symbol("EIP is at %s\n", regs->ip);
160 printk("EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
161 regs->ax, regs->bx, regs->cx, regs->dx);
162 printk("ESI: %08lx EDI: %08lx EBP: %08lx ESP: %08lx\n",
163 regs->si, regs->di, regs->bp, sp);
164 printk(" DS: %04x ES: %04x FS: %04x GS: %04x SS: %04x\n",
165 (u16)regs->ds, (u16)regs->es, (u16)regs->fs, gs, ss);
173 cr4 = read_cr4_safe();
174 printk("CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n",
181 printk("DR0: %08lx DR1: %08lx DR2: %08lx DR3: %08lx\n",
186 printk("DR6: %08lx DR7: %08lx\n",
190 void show_regs(struct pt_regs *regs)
192 __show_regs(regs, 1);
193 show_trace(NULL, regs, ®s->sp, regs->bp);
197 * This gets run with %bx containing the
198 * function to call, and %dx containing
201 extern void kernel_thread_helper(void);
204 * Create a kernel thread
206 int kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
210 memset(®s, 0, sizeof(regs));
212 regs.bx = (unsigned long) fn;
213 regs.dx = (unsigned long) arg;
217 regs.fs = __KERNEL_PERCPU;
219 regs.ip = (unsigned long) kernel_thread_helper;
220 regs.cs = __KERNEL_CS | get_kernel_rpl();
221 regs.flags = X86_EFLAGS_IF | X86_EFLAGS_SF | X86_EFLAGS_PF | 0x2;
223 /* Ok, create the new process.. */
224 return do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0, ®s, 0, NULL, NULL);
226 EXPORT_SYMBOL(kernel_thread);
229 * Free current thread data structures etc..
231 void exit_thread(void)
233 /* The process may have allocated an io port bitmap... nuke it. */
234 if (unlikely(test_thread_flag(TIF_IO_BITMAP))) {
235 struct task_struct *tsk = current;
236 struct thread_struct *t = &tsk->thread;
238 struct tss_struct *tss = &per_cpu(init_tss, cpu);
240 kfree(t->io_bitmap_ptr);
241 t->io_bitmap_ptr = NULL;
242 clear_thread_flag(TIF_IO_BITMAP);
244 * Careful, clear this in the TSS too:
246 memset(tss->io_bitmap, 0xff, tss->io_bitmap_max);
247 t->io_bitmap_max = 0;
248 tss->io_bitmap_owner = NULL;
249 tss->io_bitmap_max = 0;
250 tss->x86_tss.io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
254 /* Free any DS contexts that have not been properly released. */
255 if (unlikely(current->thread.ds_ctx)) {
256 /* we clear debugctl to make sure DS is not used. */
257 update_debugctlmsr(0);
258 ds_free(current->thread.ds_ctx);
260 #endif /* CONFIG_X86_DS */
263 void flush_thread(void)
265 struct task_struct *tsk = current;
267 tsk->thread.debugreg0 = 0;
268 tsk->thread.debugreg1 = 0;
269 tsk->thread.debugreg2 = 0;
270 tsk->thread.debugreg3 = 0;
271 tsk->thread.debugreg6 = 0;
272 tsk->thread.debugreg7 = 0;
273 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
274 clear_tsk_thread_flag(tsk, TIF_DEBUG);
276 * Forget coprocessor state..
278 tsk->fpu_counter = 0;
283 void release_thread(struct task_struct *dead_task)
285 BUG_ON(dead_task->mm);
286 release_vm86_irqs(dead_task);
290 * This gets called before we allocate a new thread and copy
291 * the current task into it.
293 void prepare_to_copy(struct task_struct *tsk)
298 int copy_thread(int nr, unsigned long clone_flags, unsigned long sp,
299 unsigned long unused,
300 struct task_struct * p, struct pt_regs * regs)
302 struct pt_regs * childregs;
303 struct task_struct *tsk;
306 childregs = task_pt_regs(p);
311 p->thread.sp = (unsigned long) childregs;
312 p->thread.sp0 = (unsigned long) (childregs+1);
314 p->thread.ip = (unsigned long) ret_from_fork;
316 savesegment(gs, p->thread.gs);
319 if (unlikely(test_tsk_thread_flag(tsk, TIF_IO_BITMAP))) {
320 p->thread.io_bitmap_ptr = kmemdup(tsk->thread.io_bitmap_ptr,
321 IO_BITMAP_BYTES, GFP_KERNEL);
322 if (!p->thread.io_bitmap_ptr) {
323 p->thread.io_bitmap_max = 0;
326 set_tsk_thread_flag(p, TIF_IO_BITMAP);
332 * Set a new TLS for the child thread?
334 if (clone_flags & CLONE_SETTLS)
335 err = do_set_thread_area(p, -1,
336 (struct user_desc __user *)childregs->si, 0);
338 if (err && p->thread.io_bitmap_ptr) {
339 kfree(p->thread.io_bitmap_ptr);
340 p->thread.io_bitmap_max = 0;
346 start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp)
348 __asm__("movl %0, %%gs" :: "r"(0));
351 regs->ds = __USER_DS;
352 regs->es = __USER_DS;
353 regs->ss = __USER_DS;
354 regs->cs = __USER_CS;
358 * Free the old FP and other extended state
360 free_thread_xstate(current);
362 EXPORT_SYMBOL_GPL(start_thread);
364 static void hard_disable_TSC(void)
366 write_cr4(read_cr4() | X86_CR4_TSD);
369 void disable_TSC(void)
372 if (!test_and_set_thread_flag(TIF_NOTSC))
374 * Must flip the CPU state synchronously with
375 * TIF_NOTSC in the current running context.
381 static void hard_enable_TSC(void)
383 write_cr4(read_cr4() & ~X86_CR4_TSD);
386 static void enable_TSC(void)
389 if (test_and_clear_thread_flag(TIF_NOTSC))
391 * Must flip the CPU state synchronously with
392 * TIF_NOTSC in the current running context.
398 int get_tsc_mode(unsigned long adr)
402 if (test_thread_flag(TIF_NOTSC))
403 val = PR_TSC_SIGSEGV;
407 return put_user(val, (unsigned int __user *)adr);
410 int set_tsc_mode(unsigned int val)
412 if (val == PR_TSC_SIGSEGV)
414 else if (val == PR_TSC_ENABLE)
423 static int update_debugctl(struct thread_struct *prev,
424 struct thread_struct *next, unsigned long debugctl)
426 unsigned long ds_prev = 0;
427 unsigned long ds_next = 0;
430 ds_prev = (unsigned long)prev->ds_ctx->ds;
432 ds_next = (unsigned long)next->ds_ctx->ds;
434 if (ds_next != ds_prev) {
435 /* we clear debugctl to make sure DS
436 * is not in use when we change it */
438 update_debugctlmsr(0);
439 wrmsr(MSR_IA32_DS_AREA, ds_next, 0);
444 static int update_debugctl(struct thread_struct *prev,
445 struct thread_struct *next, unsigned long debugctl)
449 #endif /* CONFIG_X86_DS */
452 __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p,
453 struct tss_struct *tss)
455 struct thread_struct *prev, *next;
456 unsigned long debugctl;
458 prev = &prev_p->thread;
459 next = &next_p->thread;
461 debugctl = update_debugctl(prev, next, prev->debugctlmsr);
463 if (next->debugctlmsr != debugctl)
464 update_debugctlmsr(next->debugctlmsr);
466 if (test_tsk_thread_flag(next_p, TIF_DEBUG)) {
467 set_debugreg(next->debugreg0, 0);
468 set_debugreg(next->debugreg1, 1);
469 set_debugreg(next->debugreg2, 2);
470 set_debugreg(next->debugreg3, 3);
472 set_debugreg(next->debugreg6, 6);
473 set_debugreg(next->debugreg7, 7);
476 if (test_tsk_thread_flag(prev_p, TIF_NOTSC) ^
477 test_tsk_thread_flag(next_p, TIF_NOTSC)) {
478 /* prev and next are different */
479 if (test_tsk_thread_flag(next_p, TIF_NOTSC))
485 #ifdef CONFIG_X86_PTRACE_BTS
486 if (test_tsk_thread_flag(prev_p, TIF_BTS_TRACE_TS))
487 ptrace_bts_take_timestamp(prev_p, BTS_TASK_DEPARTS);
489 if (test_tsk_thread_flag(next_p, TIF_BTS_TRACE_TS))
490 ptrace_bts_take_timestamp(next_p, BTS_TASK_ARRIVES);
491 #endif /* CONFIG_X86_PTRACE_BTS */
494 if (!test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
496 * Disable the bitmap via an invalid offset. We still cache
497 * the previous bitmap owner and the IO bitmap contents:
499 tss->x86_tss.io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
503 if (likely(next == tss->io_bitmap_owner)) {
505 * Previous owner of the bitmap (hence the bitmap content)
506 * matches the next task, we dont have to do anything but
507 * to set a valid offset in the TSS:
509 tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET;
513 * Lazy TSS's I/O bitmap copy. We set an invalid offset here
514 * and we let the task to get a GPF in case an I/O instruction
515 * is performed. The handler of the GPF will verify that the
516 * faulting task has a valid I/O bitmap and, it true, does the
517 * real copy and restart the instruction. This will save us
518 * redundant copies when the currently switched task does not
519 * perform any I/O during its timeslice.
521 tss->x86_tss.io_bitmap_base = INVALID_IO_BITMAP_OFFSET_LAZY;
525 * switch_to(x,yn) should switch tasks from x to y.
527 * We fsave/fwait so that an exception goes off at the right time
528 * (as a call from the fsave or fwait in effect) rather than to
529 * the wrong process. Lazy FP saving no longer makes any sense
530 * with modern CPU's, and this simplifies a lot of things (SMP
531 * and UP become the same).
533 * NOTE! We used to use the x86 hardware context switching. The
534 * reason for not using it any more becomes apparent when you
535 * try to recover gracefully from saved state that is no longer
536 * valid (stale segment register values in particular). With the
537 * hardware task-switch, there is no way to fix up bad state in
538 * a reasonable manner.
540 * The fact that Intel documents the hardware task-switching to
541 * be slow is a fairly red herring - this code is not noticeably
542 * faster. However, there _is_ some room for improvement here,
543 * so the performance issues may eventually be a valid point.
544 * More important, however, is the fact that this allows us much
547 * The return value (in %ax) will be the "prev" task after
548 * the task-switch, and shows up in ret_from_fork in entry.S,
551 struct task_struct * __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
553 struct thread_struct *prev = &prev_p->thread,
554 *next = &next_p->thread;
555 int cpu = smp_processor_id();
556 struct tss_struct *tss = &per_cpu(init_tss, cpu);
558 /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
560 __unlazy_fpu(prev_p);
563 /* we're going to use this soon, after a few expensive things */
564 if (next_p->fpu_counter > 5)
565 prefetch(next->xstate);
573 * Save away %gs. No need to save %fs, as it was saved on the
574 * stack on entry. No need to save %es and %ds, as those are
575 * always kernel segments while inside the kernel. Doing this
576 * before setting the new TLS descriptors avoids the situation
577 * where we temporarily have non-reloadable segments in %fs
578 * and %gs. This could be an issue if the NMI handler ever
579 * used %fs or %gs (it does not today), or if the kernel is
580 * running inside of a hypervisor layer.
582 savesegment(gs, prev->gs);
585 * Load the per-thread Thread-Local Storage descriptor.
590 * Restore IOPL if needed. In normal use, the flags restore
591 * in the switch assembly will handle this. But if the kernel
592 * is running virtualized at a non-zero CPL, the popf will
593 * not restore flags, so it must be done in a separate step.
595 if (get_kernel_rpl() && unlikely(prev->iopl != next->iopl))
596 set_iopl_mask(next->iopl);
599 * Now maybe handle debug registers and/or IO bitmaps
601 if (unlikely(task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV ||
602 task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT))
603 __switch_to_xtra(prev_p, next_p, tss);
606 * Leave lazy mode, flushing any hypercalls made here.
607 * This must be done before restoring TLS segments so
608 * the GDT and LDT are properly updated, and must be
609 * done before math_state_restore, so the TS bit is up
612 arch_leave_lazy_cpu_mode();
614 /* If the task has used fpu the last 5 timeslices, just do a full
615 * restore of the math state immediately to avoid the trap; the
616 * chances of needing FPU soon are obviously high now
618 * tsk_used_math() checks prevent calling math_state_restore(),
619 * which can sleep in the case of !tsk_used_math()
621 if (tsk_used_math(next_p) && next_p->fpu_counter > 5)
622 math_state_restore();
625 * Restore %gs if needed (which is common)
627 if (prev->gs | next->gs)
628 loadsegment(gs, next->gs);
630 x86_write_percpu(current_task, next_p);
635 asmlinkage int sys_fork(struct pt_regs regs)
637 return do_fork(SIGCHLD, regs.sp, ®s, 0, NULL, NULL);
640 asmlinkage int sys_clone(struct pt_regs regs)
642 unsigned long clone_flags;
644 int __user *parent_tidptr, *child_tidptr;
646 clone_flags = regs.bx;
648 parent_tidptr = (int __user *)regs.dx;
649 child_tidptr = (int __user *)regs.di;
652 return do_fork(clone_flags, newsp, ®s, 0, parent_tidptr, child_tidptr);
656 * This is trivial, and on the face of it looks like it
657 * could equally well be done in user mode.
659 * Not so, for quite unobvious reasons - register pressure.
660 * In user mode vfork() cannot have a stack frame, and if
661 * done by calling the "clone()" system call directly, you
662 * do not have enough call-clobbered registers to hold all
663 * the information you need.
665 asmlinkage int sys_vfork(struct pt_regs regs)
667 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs.sp, ®s, 0, NULL, NULL);
671 * sys_execve() executes a new program.
673 asmlinkage int sys_execve(struct pt_regs regs)
678 filename = getname((char __user *) regs.bx);
679 error = PTR_ERR(filename);
680 if (IS_ERR(filename))
682 error = do_execve(filename,
683 (char __user * __user *) regs.cx,
684 (char __user * __user *) regs.dx,
687 /* Make sure we don't return using sysenter.. */
688 set_thread_flag(TIF_IRET);
695 #define top_esp (THREAD_SIZE - sizeof(unsigned long))
696 #define top_ebp (THREAD_SIZE - 2*sizeof(unsigned long))
698 unsigned long get_wchan(struct task_struct *p)
700 unsigned long bp, sp, ip;
701 unsigned long stack_page;
703 if (!p || p == current || p->state == TASK_RUNNING)
705 stack_page = (unsigned long)task_stack_page(p);
707 if (!stack_page || sp < stack_page || sp > top_esp+stack_page)
709 /* include/asm-i386/system.h:switch_to() pushes bp last. */
710 bp = *(unsigned long *) sp;
712 if (bp < stack_page || bp > top_ebp+stack_page)
714 ip = *(unsigned long *) (bp+4);
715 if (!in_sched_functions(ip))
717 bp = *(unsigned long *) bp;
718 } while (count++ < 16);
722 unsigned long arch_align_stack(unsigned long sp)
724 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
725 sp -= get_random_int() % 8192;
729 unsigned long arch_randomize_brk(struct mm_struct *mm)
731 unsigned long range_end = mm->brk + 0x02000000;
732 return randomize_range(mm->brk, range_end, 0) ? : mm->brk;