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);
69 DEFINE_PER_CPU(int, cpu_number);
70 EXPORT_PER_CPU_SYMBOL(cpu_number);
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();
97 current_thread_info()->status |= TS_POLLING;
99 /* endless idle loop with no priority at all */
101 tick_nohz_stop_sched_tick(1);
102 while (!need_resched()) {
107 if (cpu_is_offline(cpu))
111 __get_cpu_var(irq_stat).idle_timestamp = jiffies;
112 /* Don't trace irqs off for idle */
113 stop_critical_timings();
115 start_critical_timings();
117 tick_nohz_restart_sched_tick();
118 preempt_enable_no_resched();
124 void __show_regs(struct pt_regs *regs, int all)
126 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L;
127 unsigned long d0, d1, d2, d3, d6, d7;
129 unsigned short ss, gs;
132 if (user_mode_vm(regs)) {
134 ss = regs->ss & 0xffff;
137 sp = (unsigned long) (®s->sp);
144 board = dmi_get_system_info(DMI_PRODUCT_NAME);
147 printk("Pid: %d, comm: %s %s (%s %.*s) %s\n",
148 task_pid_nr(current), current->comm,
149 print_tainted(), init_utsname()->release,
150 (int)strcspn(init_utsname()->version, " "),
151 init_utsname()->version, board);
153 printk("EIP: %04x:[<%08lx>] EFLAGS: %08lx CPU: %d\n",
154 (u16)regs->cs, regs->ip, regs->flags,
156 print_symbol("EIP is at %s\n", regs->ip);
158 printk("EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
159 regs->ax, regs->bx, regs->cx, regs->dx);
160 printk("ESI: %08lx EDI: %08lx EBP: %08lx ESP: %08lx\n",
161 regs->si, regs->di, regs->bp, sp);
162 printk(" DS: %04x ES: %04x FS: %04x GS: %04x SS: %04x\n",
163 (u16)regs->ds, (u16)regs->es, (u16)regs->fs, gs, ss);
171 cr4 = read_cr4_safe();
172 printk("CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n",
179 printk("DR0: %08lx DR1: %08lx DR2: %08lx DR3: %08lx\n",
184 printk("DR6: %08lx DR7: %08lx\n",
188 void show_regs(struct pt_regs *regs)
190 __show_regs(regs, 1);
191 show_trace(NULL, regs, ®s->sp, regs->bp);
195 * This gets run with %bx containing the
196 * function to call, and %dx containing
199 extern void kernel_thread_helper(void);
202 * Create a kernel thread
204 int kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
208 memset(®s, 0, sizeof(regs));
210 regs.bx = (unsigned long) fn;
211 regs.dx = (unsigned long) arg;
215 regs.fs = __KERNEL_PERCPU;
217 regs.ip = (unsigned long) kernel_thread_helper;
218 regs.cs = __KERNEL_CS | get_kernel_rpl();
219 regs.flags = X86_EFLAGS_IF | X86_EFLAGS_SF | X86_EFLAGS_PF | 0x2;
221 /* Ok, create the new process.. */
222 return do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0, ®s, 0, NULL, NULL);
224 EXPORT_SYMBOL(kernel_thread);
227 * Free current thread data structures etc..
229 void exit_thread(void)
231 /* The process may have allocated an io port bitmap... nuke it. */
232 if (unlikely(test_thread_flag(TIF_IO_BITMAP))) {
233 struct task_struct *tsk = current;
234 struct thread_struct *t = &tsk->thread;
236 struct tss_struct *tss = &per_cpu(init_tss, cpu);
238 kfree(t->io_bitmap_ptr);
239 t->io_bitmap_ptr = NULL;
240 clear_thread_flag(TIF_IO_BITMAP);
242 * Careful, clear this in the TSS too:
244 memset(tss->io_bitmap, 0xff, tss->io_bitmap_max);
245 t->io_bitmap_max = 0;
246 tss->io_bitmap_owner = NULL;
247 tss->io_bitmap_max = 0;
248 tss->x86_tss.io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
252 ds_exit_thread(current);
255 void flush_thread(void)
257 struct task_struct *tsk = current;
259 tsk->thread.debugreg0 = 0;
260 tsk->thread.debugreg1 = 0;
261 tsk->thread.debugreg2 = 0;
262 tsk->thread.debugreg3 = 0;
263 tsk->thread.debugreg6 = 0;
264 tsk->thread.debugreg7 = 0;
265 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
266 clear_tsk_thread_flag(tsk, TIF_DEBUG);
268 * Forget coprocessor state..
270 tsk->fpu_counter = 0;
275 void release_thread(struct task_struct *dead_task)
277 BUG_ON(dead_task->mm);
278 release_vm86_irqs(dead_task);
282 * This gets called before we allocate a new thread and copy
283 * the current task into it.
285 void prepare_to_copy(struct task_struct *tsk)
290 int copy_thread(int nr, unsigned long clone_flags, unsigned long sp,
291 unsigned long unused,
292 struct task_struct *p, struct pt_regs *regs)
294 struct pt_regs *childregs;
295 struct task_struct *tsk;
298 childregs = task_pt_regs(p);
303 p->thread.sp = (unsigned long) childregs;
304 p->thread.sp0 = (unsigned long) (childregs+1);
306 p->thread.ip = (unsigned long) ret_from_fork;
308 savesegment(gs, p->thread.gs);
311 if (unlikely(test_tsk_thread_flag(tsk, TIF_IO_BITMAP))) {
312 p->thread.io_bitmap_ptr = kmemdup(tsk->thread.io_bitmap_ptr,
313 IO_BITMAP_BYTES, GFP_KERNEL);
314 if (!p->thread.io_bitmap_ptr) {
315 p->thread.io_bitmap_max = 0;
318 set_tsk_thread_flag(p, TIF_IO_BITMAP);
324 * Set a new TLS for the child thread?
326 if (clone_flags & CLONE_SETTLS)
327 err = do_set_thread_area(p, -1,
328 (struct user_desc __user *)childregs->si, 0);
330 if (err && p->thread.io_bitmap_ptr) {
331 kfree(p->thread.io_bitmap_ptr);
332 p->thread.io_bitmap_max = 0;
335 ds_copy_thread(p, current);
337 clear_tsk_thread_flag(p, TIF_DEBUGCTLMSR);
338 p->thread.debugctlmsr = 0;
344 start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp)
346 __asm__("movl %0, %%gs" : : "r"(0));
349 regs->ds = __USER_DS;
350 regs->es = __USER_DS;
351 regs->ss = __USER_DS;
352 regs->cs = __USER_CS;
356 * Free the old FP and other extended state
358 free_thread_xstate(current);
360 EXPORT_SYMBOL_GPL(start_thread);
362 static void hard_disable_TSC(void)
364 write_cr4(read_cr4() | X86_CR4_TSD);
367 void disable_TSC(void)
370 if (!test_and_set_thread_flag(TIF_NOTSC))
372 * Must flip the CPU state synchronously with
373 * TIF_NOTSC in the current running context.
379 static void hard_enable_TSC(void)
381 write_cr4(read_cr4() & ~X86_CR4_TSD);
384 static void enable_TSC(void)
387 if (test_and_clear_thread_flag(TIF_NOTSC))
389 * Must flip the CPU state synchronously with
390 * TIF_NOTSC in the current running context.
396 int get_tsc_mode(unsigned long adr)
400 if (test_thread_flag(TIF_NOTSC))
401 val = PR_TSC_SIGSEGV;
405 return put_user(val, (unsigned int __user *)adr);
408 int set_tsc_mode(unsigned int val)
410 if (val == PR_TSC_SIGSEGV)
412 else if (val == PR_TSC_ENABLE)
421 __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p,
422 struct tss_struct *tss)
424 struct thread_struct *prev, *next;
426 prev = &prev_p->thread;
427 next = &next_p->thread;
429 if (test_tsk_thread_flag(next_p, TIF_DS_AREA_MSR) ||
430 test_tsk_thread_flag(prev_p, TIF_DS_AREA_MSR))
431 ds_switch_to(prev_p, next_p);
432 else if (next->debugctlmsr != prev->debugctlmsr)
433 update_debugctlmsr(next->debugctlmsr);
435 if (test_tsk_thread_flag(next_p, TIF_DEBUG)) {
436 set_debugreg(next->debugreg0, 0);
437 set_debugreg(next->debugreg1, 1);
438 set_debugreg(next->debugreg2, 2);
439 set_debugreg(next->debugreg3, 3);
441 set_debugreg(next->debugreg6, 6);
442 set_debugreg(next->debugreg7, 7);
445 if (test_tsk_thread_flag(prev_p, TIF_NOTSC) ^
446 test_tsk_thread_flag(next_p, TIF_NOTSC)) {
447 /* prev and next are different */
448 if (test_tsk_thread_flag(next_p, TIF_NOTSC))
454 if (!test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
456 * Disable the bitmap via an invalid offset. We still cache
457 * the previous bitmap owner and the IO bitmap contents:
459 tss->x86_tss.io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
463 if (likely(next == tss->io_bitmap_owner)) {
465 * Previous owner of the bitmap (hence the bitmap content)
466 * matches the next task, we dont have to do anything but
467 * to set a valid offset in the TSS:
469 tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET;
473 * Lazy TSS's I/O bitmap copy. We set an invalid offset here
474 * and we let the task to get a GPF in case an I/O instruction
475 * is performed. The handler of the GPF will verify that the
476 * faulting task has a valid I/O bitmap and, it true, does the
477 * real copy and restart the instruction. This will save us
478 * redundant copies when the currently switched task does not
479 * perform any I/O during its timeslice.
481 tss->x86_tss.io_bitmap_base = INVALID_IO_BITMAP_OFFSET_LAZY;
485 * switch_to(x,yn) should switch tasks from x to y.
487 * We fsave/fwait so that an exception goes off at the right time
488 * (as a call from the fsave or fwait in effect) rather than to
489 * the wrong process. Lazy FP saving no longer makes any sense
490 * with modern CPU's, and this simplifies a lot of things (SMP
491 * and UP become the same).
493 * NOTE! We used to use the x86 hardware context switching. The
494 * reason for not using it any more becomes apparent when you
495 * try to recover gracefully from saved state that is no longer
496 * valid (stale segment register values in particular). With the
497 * hardware task-switch, there is no way to fix up bad state in
498 * a reasonable manner.
500 * The fact that Intel documents the hardware task-switching to
501 * be slow is a fairly red herring - this code is not noticeably
502 * faster. However, there _is_ some room for improvement here,
503 * so the performance issues may eventually be a valid point.
504 * More important, however, is the fact that this allows us much
507 * The return value (in %ax) will be the "prev" task after
508 * the task-switch, and shows up in ret_from_fork in entry.S,
511 __notrace_funcgraph struct task_struct *
512 __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
514 struct thread_struct *prev = &prev_p->thread,
515 *next = &next_p->thread;
516 int cpu = smp_processor_id();
517 struct tss_struct *tss = &per_cpu(init_tss, cpu);
519 /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
521 __unlazy_fpu(prev_p);
524 /* we're going to use this soon, after a few expensive things */
525 if (next_p->fpu_counter > 5)
526 prefetch(next->xstate);
534 * Save away %gs. No need to save %fs, as it was saved on the
535 * stack on entry. No need to save %es and %ds, as those are
536 * always kernel segments while inside the kernel. Doing this
537 * before setting the new TLS descriptors avoids the situation
538 * where we temporarily have non-reloadable segments in %fs
539 * and %gs. This could be an issue if the NMI handler ever
540 * used %fs or %gs (it does not today), or if the kernel is
541 * running inside of a hypervisor layer.
543 savesegment(gs, prev->gs);
546 * Load the per-thread Thread-Local Storage descriptor.
551 * Restore IOPL if needed. In normal use, the flags restore
552 * in the switch assembly will handle this. But if the kernel
553 * is running virtualized at a non-zero CPL, the popf will
554 * not restore flags, so it must be done in a separate step.
556 if (get_kernel_rpl() && unlikely(prev->iopl != next->iopl))
557 set_iopl_mask(next->iopl);
560 * Now maybe handle debug registers and/or IO bitmaps
562 if (unlikely(task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV ||
563 task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT))
564 __switch_to_xtra(prev_p, next_p, tss);
567 * Leave lazy mode, flushing any hypercalls made here.
568 * This must be done before restoring TLS segments so
569 * the GDT and LDT are properly updated, and must be
570 * done before math_state_restore, so the TS bit is up
573 arch_leave_lazy_cpu_mode();
575 /* If the task has used fpu the last 5 timeslices, just do a full
576 * restore of the math state immediately to avoid the trap; the
577 * chances of needing FPU soon are obviously high now
579 * tsk_used_math() checks prevent calling math_state_restore(),
580 * which can sleep in the case of !tsk_used_math()
582 if (tsk_used_math(next_p) && next_p->fpu_counter > 5)
583 math_state_restore();
586 * Restore %gs if needed (which is common)
588 if (prev->gs | next->gs)
589 loadsegment(gs, next->gs);
591 x86_write_percpu(current_task, next_p);
596 asmlinkage int sys_fork(struct pt_regs regs)
598 return do_fork(SIGCHLD, regs.sp, ®s, 0, NULL, NULL);
601 asmlinkage int sys_clone(struct pt_regs regs)
603 unsigned long clone_flags;
605 int __user *parent_tidptr, *child_tidptr;
607 clone_flags = regs.bx;
609 parent_tidptr = (int __user *)regs.dx;
610 child_tidptr = (int __user *)regs.di;
613 return do_fork(clone_flags, newsp, ®s, 0, parent_tidptr, child_tidptr);
617 * This is trivial, and on the face of it looks like it
618 * could equally well be done in user mode.
620 * Not so, for quite unobvious reasons - register pressure.
621 * In user mode vfork() cannot have a stack frame, and if
622 * done by calling the "clone()" system call directly, you
623 * do not have enough call-clobbered registers to hold all
624 * the information you need.
626 asmlinkage int sys_vfork(struct pt_regs regs)
628 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs.sp, ®s, 0, NULL, NULL);
632 * sys_execve() executes a new program.
634 asmlinkage int sys_execve(struct pt_regs regs)
639 filename = getname((char __user *) regs.bx);
640 error = PTR_ERR(filename);
641 if (IS_ERR(filename))
643 error = do_execve(filename,
644 (char __user * __user *) regs.cx,
645 (char __user * __user *) regs.dx,
648 /* Make sure we don't return using sysenter.. */
649 set_thread_flag(TIF_IRET);
656 #define top_esp (THREAD_SIZE - sizeof(unsigned long))
657 #define top_ebp (THREAD_SIZE - 2*sizeof(unsigned long))
659 unsigned long get_wchan(struct task_struct *p)
661 unsigned long bp, sp, ip;
662 unsigned long stack_page;
664 if (!p || p == current || p->state == TASK_RUNNING)
666 stack_page = (unsigned long)task_stack_page(p);
668 if (!stack_page || sp < stack_page || sp > top_esp+stack_page)
670 /* include/asm-i386/system.h:switch_to() pushes bp last. */
671 bp = *(unsigned long *) sp;
673 if (bp < stack_page || bp > top_ebp+stack_page)
675 ip = *(unsigned long *) (bp+4);
676 if (!in_sched_functions(ip))
678 bp = *(unsigned long *) bp;
679 } while (count++ < 16);
683 unsigned long arch_align_stack(unsigned long sp)
685 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
686 sp -= get_random_int() % 8192;
690 unsigned long arch_randomize_brk(struct mm_struct *mm)
692 unsigned long range_end = mm->brk + 0x02000000;
693 return randomize_range(mm->brk, range_end, 0) ? : mm->brk;