2 * Copyright (C) 1995 Linus Torvalds
4 * Pentium III FXSR, SSE support
5 * Gareth Hughes <gareth@valinux.com>, May 2000
10 * CPU hotplug support - ashok.raj@intel.com
14 * This file handles the architecture-dependent parts of process handling..
19 #include <linux/cpu.h>
20 #include <linux/errno.h>
21 #include <linux/sched.h>
23 #include <linux/kernel.h>
25 #include <linux/elfcore.h>
26 #include <linux/smp.h>
27 #include <linux/slab.h>
28 #include <linux/user.h>
29 #include <linux/interrupt.h>
30 #include <linux/utsname.h>
31 #include <linux/delay.h>
32 #include <linux/module.h>
33 #include <linux/ptrace.h>
34 #include <linux/random.h>
35 #include <linux/notifier.h>
36 #include <linux/kprobes.h>
37 #include <linux/kdebug.h>
38 #include <linux/tick.h>
39 #include <linux/prctl.h>
41 #include <asm/uaccess.h>
42 #include <asm/pgtable.h>
43 #include <asm/system.h>
45 #include <asm/processor.h>
47 #include <asm/mmu_context.h>
49 #include <asm/prctl.h>
51 #include <asm/proto.h>
55 asmlinkage extern void ret_from_fork(void);
57 unsigned long kernel_thread_flags = CLONE_VM | CLONE_UNTRACED;
59 unsigned long boot_option_idle_override = 0;
60 EXPORT_SYMBOL(boot_option_idle_override);
63 * Powermanagement idle function, if any..
65 void (*pm_idle)(void);
66 EXPORT_SYMBOL(pm_idle);
68 static ATOMIC_NOTIFIER_HEAD(idle_notifier);
70 void idle_notifier_register(struct notifier_block *n)
72 atomic_notifier_chain_register(&idle_notifier, n);
78 atomic_notifier_call_chain(&idle_notifier, IDLE_START, NULL);
81 static void __exit_idle(void)
83 if (test_and_clear_bit_pda(0, isidle) == 0)
85 atomic_notifier_call_chain(&idle_notifier, IDLE_END, NULL);
88 /* Called from interrupts to signify idle end */
91 /* idle loop has pid 0 */
98 * We use this if we don't have any better
101 void default_idle(void)
103 current_thread_info()->status &= ~TS_POLLING;
105 * TS_POLLING-cleared state must be visible before we
110 safe_halt(); /* enables interrupts racelessly */
113 current_thread_info()->status |= TS_POLLING;
116 #ifdef CONFIG_HOTPLUG_CPU
117 DECLARE_PER_CPU(int, cpu_state);
120 /* We halt the CPU with physical CPU hotplug */
121 static inline void play_dead(void)
127 __get_cpu_var(cpu_state) = CPU_DEAD;
134 static inline void play_dead(void)
138 #endif /* CONFIG_HOTPLUG_CPU */
141 * The idle thread. There's no useful work to be
142 * done, so just try to conserve power and have a
143 * low exit latency (ie sit in a loop waiting for
144 * somebody to say that they'd like to reschedule)
148 current_thread_info()->status |= TS_POLLING;
150 #ifdef CONFIG_CC_STACKPROTECTOR
152 * If we're the non-boot CPU, nothing set the PDA stack
153 * canary up for us. This is as good a place as any for
156 write_pda(stack_canary, current->stack_canary);
158 /* endless idle loop with no priority at all */
160 tick_nohz_stop_sched_tick();
161 while (!need_resched()) {
168 if (cpu_is_offline(smp_processor_id()))
171 * Idle routines should keep interrupts disabled
172 * from here on, until they go to idle.
173 * Otherwise, idle callbacks can misfire.
178 /* In many cases the interrupt that ended idle
179 has already called exit_idle. But some idle
180 loops can be woken up without interrupt. */
184 tick_nohz_restart_sched_tick();
185 preempt_enable_no_resched();
191 /* Prints also some state that isn't saved in the pt_regs */
192 void __show_regs(struct pt_regs * regs)
194 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L, fs, gs, shadowgs;
195 unsigned long d0, d1, d2, d3, d6, d7;
196 unsigned int fsindex, gsindex;
197 unsigned int ds, cs, es;
201 printk("Pid: %d, comm: %.20s %s %s %.*s\n",
202 current->pid, current->comm, print_tainted(),
203 init_utsname()->release,
204 (int)strcspn(init_utsname()->version, " "),
205 init_utsname()->version);
206 printk("RIP: %04lx:[<%016lx>] ", regs->cs & 0xffff, regs->ip);
207 printk_address(regs->ip, 1);
208 printk("RSP: %04lx:%016lx EFLAGS: %08lx\n", regs->ss, regs->sp,
210 printk("RAX: %016lx RBX: %016lx RCX: %016lx\n",
211 regs->ax, regs->bx, regs->cx);
212 printk("RDX: %016lx RSI: %016lx RDI: %016lx\n",
213 regs->dx, regs->si, regs->di);
214 printk("RBP: %016lx R08: %016lx R09: %016lx\n",
215 regs->bp, regs->r8, regs->r9);
216 printk("R10: %016lx R11: %016lx R12: %016lx\n",
217 regs->r10, regs->r11, regs->r12);
218 printk("R13: %016lx R14: %016lx R15: %016lx\n",
219 regs->r13, regs->r14, regs->r15);
221 asm("movl %%ds,%0" : "=r" (ds));
222 asm("movl %%cs,%0" : "=r" (cs));
223 asm("movl %%es,%0" : "=r" (es));
224 asm("movl %%fs,%0" : "=r" (fsindex));
225 asm("movl %%gs,%0" : "=r" (gsindex));
227 rdmsrl(MSR_FS_BASE, fs);
228 rdmsrl(MSR_GS_BASE, gs);
229 rdmsrl(MSR_KERNEL_GS_BASE, shadowgs);
236 printk("FS: %016lx(%04x) GS:%016lx(%04x) knlGS:%016lx\n",
237 fs,fsindex,gs,gsindex,shadowgs);
238 printk("CS: %04x DS: %04x ES: %04x CR0: %016lx\n", cs, ds, es, cr0);
239 printk("CR2: %016lx CR3: %016lx CR4: %016lx\n", cr2, cr3, cr4);
244 printk("DR0: %016lx DR1: %016lx DR2: %016lx\n", d0, d1, d2);
248 printk("DR3: %016lx DR6: %016lx DR7: %016lx\n", d3, d6, d7);
251 void show_regs(struct pt_regs *regs)
253 printk("CPU %d:", smp_processor_id());
255 show_trace(NULL, regs, (void *)(regs + 1), regs->bp);
259 * Free current thread data structures etc..
261 void exit_thread(void)
263 struct task_struct *me = current;
264 struct thread_struct *t = &me->thread;
266 if (me->thread.io_bitmap_ptr) {
267 struct tss_struct *tss = &per_cpu(init_tss, get_cpu());
269 kfree(t->io_bitmap_ptr);
270 t->io_bitmap_ptr = NULL;
271 clear_thread_flag(TIF_IO_BITMAP);
273 * Careful, clear this in the TSS too:
275 memset(tss->io_bitmap, 0xff, t->io_bitmap_max);
276 t->io_bitmap_max = 0;
281 void flush_thread(void)
283 struct task_struct *tsk = current;
285 if (test_tsk_thread_flag(tsk, TIF_ABI_PENDING)) {
286 clear_tsk_thread_flag(tsk, TIF_ABI_PENDING);
287 if (test_tsk_thread_flag(tsk, TIF_IA32)) {
288 clear_tsk_thread_flag(tsk, TIF_IA32);
290 set_tsk_thread_flag(tsk, TIF_IA32);
291 current_thread_info()->status |= TS_COMPAT;
294 clear_tsk_thread_flag(tsk, TIF_DEBUG);
296 tsk->thread.debugreg0 = 0;
297 tsk->thread.debugreg1 = 0;
298 tsk->thread.debugreg2 = 0;
299 tsk->thread.debugreg3 = 0;
300 tsk->thread.debugreg6 = 0;
301 tsk->thread.debugreg7 = 0;
302 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
304 * Forget coprocessor state..
310 void release_thread(struct task_struct *dead_task)
313 if (dead_task->mm->context.size) {
314 printk("WARNING: dead process %8s still has LDT? <%p/%d>\n",
316 dead_task->mm->context.ldt,
317 dead_task->mm->context.size);
323 static inline void set_32bit_tls(struct task_struct *t, int tls, u32 addr)
325 struct user_desc ud = {
332 struct desc_struct *desc = t->thread.tls_array;
337 static inline u32 read_32bit_tls(struct task_struct *t, int tls)
339 return get_desc_base(&t->thread.tls_array[tls]);
343 * This gets called before we allocate a new thread and copy
344 * the current task into it.
346 void prepare_to_copy(struct task_struct *tsk)
351 int copy_thread(int nr, unsigned long clone_flags, unsigned long sp,
352 unsigned long unused,
353 struct task_struct * p, struct pt_regs * regs)
356 struct pt_regs * childregs;
357 struct task_struct *me = current;
359 childregs = ((struct pt_regs *)
360 (THREAD_SIZE + task_stack_page(p))) - 1;
366 childregs->sp = (unsigned long)childregs;
368 p->thread.sp = (unsigned long) childregs;
369 p->thread.sp0 = (unsigned long) (childregs+1);
370 p->thread.usersp = me->thread.usersp;
372 set_tsk_thread_flag(p, TIF_FORK);
374 p->thread.fs = me->thread.fs;
375 p->thread.gs = me->thread.gs;
377 asm("mov %%gs,%0" : "=m" (p->thread.gsindex));
378 asm("mov %%fs,%0" : "=m" (p->thread.fsindex));
379 asm("mov %%es,%0" : "=m" (p->thread.es));
380 asm("mov %%ds,%0" : "=m" (p->thread.ds));
382 if (unlikely(test_tsk_thread_flag(me, TIF_IO_BITMAP))) {
383 p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL);
384 if (!p->thread.io_bitmap_ptr) {
385 p->thread.io_bitmap_max = 0;
388 memcpy(p->thread.io_bitmap_ptr, me->thread.io_bitmap_ptr,
390 set_tsk_thread_flag(p, TIF_IO_BITMAP);
394 * Set a new TLS for the child thread?
396 if (clone_flags & CLONE_SETTLS) {
397 #ifdef CONFIG_IA32_EMULATION
398 if (test_thread_flag(TIF_IA32))
399 err = do_set_thread_area(p, -1,
400 (struct user_desc __user *)childregs->si, 0);
403 err = do_arch_prctl(p, ARCH_SET_FS, childregs->r8);
409 if (err && p->thread.io_bitmap_ptr) {
410 kfree(p->thread.io_bitmap_ptr);
411 p->thread.io_bitmap_max = 0;
417 start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp)
419 asm volatile("movl %0, %%fs; movl %0, %%es; movl %0, %%ds" :: "r"(0));
423 write_pda(oldrsp, new_sp);
424 regs->cs = __USER_CS;
425 regs->ss = __USER_DS;
429 * Free the old FP and other extended state
431 free_thread_xstate(current);
433 EXPORT_SYMBOL_GPL(start_thread);
435 static void hard_disable_TSC(void)
437 write_cr4(read_cr4() | X86_CR4_TSD);
440 void disable_TSC(void)
443 if (!test_and_set_thread_flag(TIF_NOTSC))
445 * Must flip the CPU state synchronously with
446 * TIF_NOTSC in the current running context.
452 static void hard_enable_TSC(void)
454 write_cr4(read_cr4() & ~X86_CR4_TSD);
457 static void enable_TSC(void)
460 if (test_and_clear_thread_flag(TIF_NOTSC))
462 * Must flip the CPU state synchronously with
463 * TIF_NOTSC in the current running context.
469 int get_tsc_mode(unsigned long adr)
473 if (test_thread_flag(TIF_NOTSC))
474 val = PR_TSC_SIGSEGV;
478 return put_user(val, (unsigned int __user *)adr);
481 int set_tsc_mode(unsigned int val)
483 if (val == PR_TSC_SIGSEGV)
485 else if (val == PR_TSC_ENABLE)
494 * This special macro can be used to load a debugging register
496 #define loaddebug(thread, r) set_debugreg(thread->debugreg ## r, r)
498 static inline void __switch_to_xtra(struct task_struct *prev_p,
499 struct task_struct *next_p,
500 struct tss_struct *tss)
502 struct thread_struct *prev, *next;
503 unsigned long debugctl;
505 prev = &prev_p->thread,
506 next = &next_p->thread;
508 debugctl = prev->debugctlmsr;
509 if (next->ds_area_msr != prev->ds_area_msr) {
510 /* we clear debugctl to make sure DS
511 * is not in use when we change it */
513 update_debugctlmsr(0);
514 wrmsrl(MSR_IA32_DS_AREA, next->ds_area_msr);
517 if (next->debugctlmsr != debugctl)
518 update_debugctlmsr(next->debugctlmsr);
520 if (test_tsk_thread_flag(next_p, TIF_DEBUG)) {
530 if (test_tsk_thread_flag(prev_p, TIF_NOTSC) ^
531 test_tsk_thread_flag(next_p, TIF_NOTSC)) {
532 /* prev and next are different */
533 if (test_tsk_thread_flag(next_p, TIF_NOTSC))
539 if (test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
541 * Copy the relevant range of the IO bitmap.
542 * Normally this is 128 bytes or less:
544 memcpy(tss->io_bitmap, next->io_bitmap_ptr,
545 max(prev->io_bitmap_max, next->io_bitmap_max));
546 } else if (test_tsk_thread_flag(prev_p, TIF_IO_BITMAP)) {
548 * Clear any possible leftover bits:
550 memset(tss->io_bitmap, 0xff, prev->io_bitmap_max);
554 if (test_tsk_thread_flag(prev_p, TIF_BTS_TRACE_TS))
555 ptrace_bts_take_timestamp(prev_p, BTS_TASK_DEPARTS);
557 if (test_tsk_thread_flag(next_p, TIF_BTS_TRACE_TS))
558 ptrace_bts_take_timestamp(next_p, BTS_TASK_ARRIVES);
563 * switch_to(x,y) should switch tasks from x to y.
565 * This could still be optimized:
566 * - fold all the options into a flag word and test it with a single test.
567 * - could test fs/gs bitsliced
569 * Kprobes not supported here. Set the probe on schedule instead.
572 __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
574 struct thread_struct *prev = &prev_p->thread,
575 *next = &next_p->thread;
576 int cpu = smp_processor_id();
577 struct tss_struct *tss = &per_cpu(init_tss, cpu);
579 /* we're going to use this soon, after a few expensive things */
580 if (next_p->fpu_counter>5)
581 prefetch(next->xstate);
584 * Reload esp0, LDT and the page table pointer:
590 * This won't pick up thread selector changes, but I guess that is ok.
592 asm volatile("mov %%es,%0" : "=m" (prev->es));
593 if (unlikely(next->es | prev->es))
594 loadsegment(es, next->es);
596 asm volatile ("mov %%ds,%0" : "=m" (prev->ds));
597 if (unlikely(next->ds | prev->ds))
598 loadsegment(ds, next->ds);
607 asm volatile("movl %%fs,%0" : "=r" (fsindex));
608 /* segment register != 0 always requires a reload.
609 also reload when it has changed.
610 when prev process used 64bit base always reload
611 to avoid an information leak. */
612 if (unlikely(fsindex | next->fsindex | prev->fs)) {
613 loadsegment(fs, next->fsindex);
614 /* check if the user used a selector != 0
615 * if yes clear 64bit base, since overloaded base
616 * is always mapped to the Null selector
621 /* when next process has a 64bit base use it */
623 wrmsrl(MSR_FS_BASE, next->fs);
624 prev->fsindex = fsindex;
628 asm volatile("movl %%gs,%0" : "=r" (gsindex));
629 if (unlikely(gsindex | next->gsindex | prev->gs)) {
630 load_gs_index(next->gsindex);
635 wrmsrl(MSR_KERNEL_GS_BASE, next->gs);
636 prev->gsindex = gsindex;
639 /* Must be after DS reload */
643 * Switch the PDA and FPU contexts.
645 prev->usersp = read_pda(oldrsp);
646 write_pda(oldrsp, next->usersp);
647 write_pda(pcurrent, next_p);
649 write_pda(kernelstack,
650 (unsigned long)task_stack_page(next_p) + THREAD_SIZE - PDA_STACKOFFSET);
651 #ifdef CONFIG_CC_STACKPROTECTOR
653 * Build time only check to make sure the stack_canary is at
654 * offset 40 in the pda; this is a gcc ABI requirement
656 BUILD_BUG_ON(offsetof(struct x8664_pda, stack_canary) != 40);
660 * Now maybe reload the debug registers and handle I/O bitmaps
662 if (unlikely(task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT ||
663 task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV))
664 __switch_to_xtra(prev_p, next_p, tss);
666 /* If the task has used fpu the last 5 timeslices, just do a full
667 * restore of the math state immediately to avoid the trap; the
668 * chances of needing FPU soon are obviously high now
670 if (next_p->fpu_counter>5)
671 math_state_restore();
676 * sys_execve() executes a new program.
679 long sys_execve(char __user *name, char __user * __user *argv,
680 char __user * __user *envp, struct pt_regs *regs)
685 filename = getname(name);
686 error = PTR_ERR(filename);
687 if (IS_ERR(filename))
689 error = do_execve(filename, argv, envp, regs);
694 void set_personality_64bit(void)
696 /* inherit personality from parent */
698 /* Make sure to be in 64bit mode */
699 clear_thread_flag(TIF_IA32);
701 /* TBD: overwrites user setup. Should have two bits.
702 But 64bit processes have always behaved this way,
703 so it's not too bad. The main problem is just that
704 32bit childs are affected again. */
705 current->personality &= ~READ_IMPLIES_EXEC;
708 asmlinkage long sys_fork(struct pt_regs *regs)
710 return do_fork(SIGCHLD, regs->sp, regs, 0, NULL, NULL);
714 sys_clone(unsigned long clone_flags, unsigned long newsp,
715 void __user *parent_tid, void __user *child_tid, struct pt_regs *regs)
719 return do_fork(clone_flags, newsp, regs, 0, parent_tid, child_tid);
723 * This is trivial, and on the face of it looks like it
724 * could equally well be done in user mode.
726 * Not so, for quite unobvious reasons - register pressure.
727 * In user mode vfork() cannot have a stack frame, and if
728 * done by calling the "clone()" system call directly, you
729 * do not have enough call-clobbered registers to hold all
730 * the information you need.
732 asmlinkage long sys_vfork(struct pt_regs *regs)
734 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->sp, regs, 0,
738 unsigned long get_wchan(struct task_struct *p)
744 if (!p || p == current || p->state==TASK_RUNNING)
746 stack = (unsigned long)task_stack_page(p);
747 if (p->thread.sp < stack || p->thread.sp > stack+THREAD_SIZE)
749 fp = *(u64 *)(p->thread.sp);
751 if (fp < (unsigned long)stack ||
752 fp > (unsigned long)stack+THREAD_SIZE)
755 if (!in_sched_functions(ip))
758 } while (count++ < 16);
762 long do_arch_prctl(struct task_struct *task, int code, unsigned long addr)
765 int doit = task == current;
770 if (addr >= TASK_SIZE_OF(task))
773 /* handle small bases via the GDT because that's faster to
775 if (addr <= 0xffffffff) {
776 set_32bit_tls(task, GS_TLS, addr);
778 load_TLS(&task->thread, cpu);
779 load_gs_index(GS_TLS_SEL);
781 task->thread.gsindex = GS_TLS_SEL;
784 task->thread.gsindex = 0;
785 task->thread.gs = addr;
788 ret = checking_wrmsrl(MSR_KERNEL_GS_BASE, addr);
794 /* Not strictly needed for fs, but do it for symmetry
796 if (addr >= TASK_SIZE_OF(task))
799 /* handle small bases via the GDT because that's faster to
801 if (addr <= 0xffffffff) {
802 set_32bit_tls(task, FS_TLS, addr);
804 load_TLS(&task->thread, cpu);
805 asm volatile("movl %0,%%fs" :: "r"(FS_TLS_SEL));
807 task->thread.fsindex = FS_TLS_SEL;
810 task->thread.fsindex = 0;
811 task->thread.fs = addr;
813 /* set the selector to 0 to not confuse
815 asm volatile("movl %0,%%fs" :: "r" (0));
816 ret = checking_wrmsrl(MSR_FS_BASE, addr);
823 if (task->thread.fsindex == FS_TLS_SEL)
824 base = read_32bit_tls(task, FS_TLS);
826 rdmsrl(MSR_FS_BASE, base);
828 base = task->thread.fs;
829 ret = put_user(base, (unsigned long __user *)addr);
835 if (task->thread.gsindex == GS_TLS_SEL)
836 base = read_32bit_tls(task, GS_TLS);
838 asm("movl %%gs,%0" : "=r" (gsindex));
840 rdmsrl(MSR_KERNEL_GS_BASE, base);
842 base = task->thread.gs;
845 base = task->thread.gs;
846 ret = put_user(base, (unsigned long __user *)addr);
858 long sys_arch_prctl(int code, unsigned long addr)
860 return do_arch_prctl(current, code, addr);
863 unsigned long arch_align_stack(unsigned long sp)
865 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
866 sp -= get_random_int() % 8192;
870 unsigned long arch_randomize_brk(struct mm_struct *mm)
872 unsigned long range_end = mm->brk + 0x02000000;
873 return randomize_range(mm->brk, range_end, 0) ? : mm->brk;