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>
40 #include <linux/uaccess.h>
43 #include <asm/pgtable.h>
44 #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>
54 #include <asm/syscalls.h>
56 asmlinkage extern void ret_from_fork(void);
58 unsigned long kernel_thread_flags = CLONE_VM | CLONE_UNTRACED;
60 static ATOMIC_NOTIFIER_HEAD(idle_notifier);
62 void idle_notifier_register(struct notifier_block *n)
64 atomic_notifier_chain_register(&idle_notifier, n);
70 atomic_notifier_call_chain(&idle_notifier, IDLE_START, NULL);
73 static void __exit_idle(void)
75 if (test_and_clear_bit_pda(0, isidle) == 0)
77 atomic_notifier_call_chain(&idle_notifier, IDLE_END, NULL);
80 /* Called from interrupts to signify idle end */
83 /* idle loop has pid 0 */
90 static inline void play_dead(void)
97 * The idle thread. There's no useful work to be
98 * done, so just try to conserve power and have a
99 * low exit latency (ie sit in a loop waiting for
100 * somebody to say that they'd like to reschedule)
104 current_thread_info()->status |= TS_POLLING;
105 /* endless idle loop with no priority at all */
107 tick_nohz_stop_sched_tick(1);
108 while (!need_resched()) {
112 if (cpu_is_offline(smp_processor_id()))
115 * Idle routines should keep interrupts disabled
116 * from here on, until they go to idle.
117 * Otherwise, idle callbacks can misfire.
121 /* Don't trace irqs off for idle */
122 stop_critical_timings();
124 start_critical_timings();
125 /* In many cases the interrupt that ended idle
126 has already called exit_idle. But some idle
127 loops can be woken up without interrupt. */
131 tick_nohz_restart_sched_tick();
132 preempt_enable_no_resched();
138 /* Prints also some state that isn't saved in the pt_regs */
139 void __show_regs(struct pt_regs *regs, int all)
141 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L, fs, gs, shadowgs;
142 unsigned long d0, d1, d2, d3, d6, d7;
143 unsigned int fsindex, gsindex;
144 unsigned int ds, cs, es;
148 printk(KERN_INFO "Pid: %d, comm: %.20s %s %s %.*s\n",
149 current->pid, current->comm, print_tainted(),
150 init_utsname()->release,
151 (int)strcspn(init_utsname()->version, " "),
152 init_utsname()->version);
153 printk(KERN_INFO "RIP: %04lx:[<%016lx>] ", regs->cs & 0xffff, regs->ip);
154 printk_address(regs->ip, 1);
155 printk(KERN_INFO "RSP: %04lx:%016lx EFLAGS: %08lx\n", regs->ss,
156 regs->sp, regs->flags);
157 printk(KERN_INFO "RAX: %016lx RBX: %016lx RCX: %016lx\n",
158 regs->ax, regs->bx, regs->cx);
159 printk(KERN_INFO "RDX: %016lx RSI: %016lx RDI: %016lx\n",
160 regs->dx, regs->si, regs->di);
161 printk(KERN_INFO "RBP: %016lx R08: %016lx R09: %016lx\n",
162 regs->bp, regs->r8, regs->r9);
163 printk(KERN_INFO "R10: %016lx R11: %016lx R12: %016lx\n",
164 regs->r10, regs->r11, regs->r12);
165 printk(KERN_INFO "R13: %016lx R14: %016lx R15: %016lx\n",
166 regs->r13, regs->r14, regs->r15);
168 asm("movl %%ds,%0" : "=r" (ds));
169 asm("movl %%cs,%0" : "=r" (cs));
170 asm("movl %%es,%0" : "=r" (es));
171 asm("movl %%fs,%0" : "=r" (fsindex));
172 asm("movl %%gs,%0" : "=r" (gsindex));
174 rdmsrl(MSR_FS_BASE, fs);
175 rdmsrl(MSR_GS_BASE, gs);
176 rdmsrl(MSR_KERNEL_GS_BASE, shadowgs);
186 printk(KERN_INFO "FS: %016lx(%04x) GS:%016lx(%04x) knlGS:%016lx\n",
187 fs, fsindex, gs, gsindex, shadowgs);
188 printk(KERN_INFO "CS: %04x DS: %04x ES: %04x CR0: %016lx\n", cs, ds,
190 printk(KERN_INFO "CR2: %016lx CR3: %016lx CR4: %016lx\n", cr2, cr3,
196 printk(KERN_INFO "DR0: %016lx DR1: %016lx DR2: %016lx\n", d0, d1, d2);
200 printk(KERN_INFO "DR3: %016lx DR6: %016lx DR7: %016lx\n", d3, d6, d7);
203 void show_regs(struct pt_regs *regs)
205 printk(KERN_INFO "CPU %d:", smp_processor_id());
206 __show_regs(regs, 1);
207 show_trace(NULL, regs, (void *)(regs + 1), regs->bp);
211 * Free current thread data structures etc..
213 void exit_thread(void)
215 struct task_struct *me = current;
216 struct thread_struct *t = &me->thread;
218 if (me->thread.io_bitmap_ptr) {
219 struct tss_struct *tss = &per_cpu(init_tss, get_cpu());
221 kfree(t->io_bitmap_ptr);
222 t->io_bitmap_ptr = NULL;
223 clear_thread_flag(TIF_IO_BITMAP);
225 * Careful, clear this in the TSS too:
227 memset(tss->io_bitmap, 0xff, t->io_bitmap_max);
228 t->io_bitmap_max = 0;
232 /* Free any DS contexts that have not been properly released. */
233 if (unlikely(t->ds_ctx)) {
234 /* we clear debugctl to make sure DS is not used. */
235 update_debugctlmsr(0);
238 #endif /* CONFIG_X86_DS */
241 void flush_thread(void)
243 struct task_struct *tsk = current;
245 if (test_tsk_thread_flag(tsk, TIF_ABI_PENDING)) {
246 clear_tsk_thread_flag(tsk, TIF_ABI_PENDING);
247 if (test_tsk_thread_flag(tsk, TIF_IA32)) {
248 clear_tsk_thread_flag(tsk, TIF_IA32);
250 set_tsk_thread_flag(tsk, TIF_IA32);
251 current_thread_info()->status |= TS_COMPAT;
254 clear_tsk_thread_flag(tsk, TIF_DEBUG);
256 tsk->thread.debugreg0 = 0;
257 tsk->thread.debugreg1 = 0;
258 tsk->thread.debugreg2 = 0;
259 tsk->thread.debugreg3 = 0;
260 tsk->thread.debugreg6 = 0;
261 tsk->thread.debugreg7 = 0;
262 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
264 * Forget coprocessor state..
266 tsk->fpu_counter = 0;
271 void release_thread(struct task_struct *dead_task)
274 if (dead_task->mm->context.size) {
275 printk("WARNING: dead process %8s still has LDT? <%p/%d>\n",
277 dead_task->mm->context.ldt,
278 dead_task->mm->context.size);
284 static inline void set_32bit_tls(struct task_struct *t, int tls, u32 addr)
286 struct user_desc ud = {
293 struct desc_struct *desc = t->thread.tls_array;
298 static inline u32 read_32bit_tls(struct task_struct *t, int tls)
300 return get_desc_base(&t->thread.tls_array[tls]);
304 * This gets called before we allocate a new thread and copy
305 * the current task into it.
307 void prepare_to_copy(struct task_struct *tsk)
312 int copy_thread(int nr, unsigned long clone_flags, unsigned long sp,
313 unsigned long unused,
314 struct task_struct *p, struct pt_regs *regs)
317 struct pt_regs *childregs;
318 struct task_struct *me = current;
320 childregs = ((struct pt_regs *)
321 (THREAD_SIZE + task_stack_page(p))) - 1;
327 childregs->sp = (unsigned long)childregs;
329 p->thread.sp = (unsigned long) childregs;
330 p->thread.sp0 = (unsigned long) (childregs+1);
331 p->thread.usersp = me->thread.usersp;
333 set_tsk_thread_flag(p, TIF_FORK);
335 p->thread.fs = me->thread.fs;
336 p->thread.gs = me->thread.gs;
338 savesegment(gs, p->thread.gsindex);
339 savesegment(fs, p->thread.fsindex);
340 savesegment(es, p->thread.es);
341 savesegment(ds, p->thread.ds);
343 if (unlikely(test_tsk_thread_flag(me, TIF_IO_BITMAP))) {
344 p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL);
345 if (!p->thread.io_bitmap_ptr) {
346 p->thread.io_bitmap_max = 0;
349 memcpy(p->thread.io_bitmap_ptr, me->thread.io_bitmap_ptr,
351 set_tsk_thread_flag(p, TIF_IO_BITMAP);
355 * Set a new TLS for the child thread?
357 if (clone_flags & CLONE_SETTLS) {
358 #ifdef CONFIG_IA32_EMULATION
359 if (test_thread_flag(TIF_IA32))
360 err = do_set_thread_area(p, -1,
361 (struct user_desc __user *)childregs->si, 0);
364 err = do_arch_prctl(p, ARCH_SET_FS, childregs->r8);
370 if (err && p->thread.io_bitmap_ptr) {
371 kfree(p->thread.io_bitmap_ptr);
372 p->thread.io_bitmap_max = 0;
378 start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp)
386 write_pda(oldrsp, new_sp);
387 regs->cs = __USER_CS;
388 regs->ss = __USER_DS;
392 * Free the old FP and other extended state
394 free_thread_xstate(current);
396 EXPORT_SYMBOL_GPL(start_thread);
398 static void hard_disable_TSC(void)
400 write_cr4(read_cr4() | X86_CR4_TSD);
403 void disable_TSC(void)
406 if (!test_and_set_thread_flag(TIF_NOTSC))
408 * Must flip the CPU state synchronously with
409 * TIF_NOTSC in the current running context.
415 static void hard_enable_TSC(void)
417 write_cr4(read_cr4() & ~X86_CR4_TSD);
420 static void enable_TSC(void)
423 if (test_and_clear_thread_flag(TIF_NOTSC))
425 * Must flip the CPU state synchronously with
426 * TIF_NOTSC in the current running context.
432 int get_tsc_mode(unsigned long adr)
436 if (test_thread_flag(TIF_NOTSC))
437 val = PR_TSC_SIGSEGV;
441 return put_user(val, (unsigned int __user *)adr);
444 int set_tsc_mode(unsigned int val)
446 if (val == PR_TSC_SIGSEGV)
448 else if (val == PR_TSC_ENABLE)
457 * This special macro can be used to load a debugging register
459 #define loaddebug(thread, r) set_debugreg(thread->debugreg ## r, r)
461 static inline void __switch_to_xtra(struct task_struct *prev_p,
462 struct task_struct *next_p,
463 struct tss_struct *tss)
465 struct thread_struct *prev, *next;
466 unsigned long debugctl;
468 prev = &prev_p->thread,
469 next = &next_p->thread;
471 debugctl = prev->debugctlmsr;
475 unsigned long ds_prev = 0, ds_next = 0;
478 ds_prev = (unsigned long)prev->ds_ctx->ds;
480 ds_next = (unsigned long)next->ds_ctx->ds;
482 if (ds_next != ds_prev) {
484 * We clear debugctl to make sure DS
485 * is not in use when we change it:
488 update_debugctlmsr(0);
489 wrmsrl(MSR_IA32_DS_AREA, ds_next);
492 #endif /* CONFIG_X86_DS */
494 if (next->debugctlmsr != debugctl)
495 update_debugctlmsr(next->debugctlmsr);
497 if (test_tsk_thread_flag(next_p, TIF_DEBUG)) {
507 if (test_tsk_thread_flag(prev_p, TIF_NOTSC) ^
508 test_tsk_thread_flag(next_p, TIF_NOTSC)) {
509 /* prev and next are different */
510 if (test_tsk_thread_flag(next_p, TIF_NOTSC))
516 if (test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
518 * Copy the relevant range of the IO bitmap.
519 * Normally this is 128 bytes or less:
521 memcpy(tss->io_bitmap, next->io_bitmap_ptr,
522 max(prev->io_bitmap_max, next->io_bitmap_max));
523 } else if (test_tsk_thread_flag(prev_p, TIF_IO_BITMAP)) {
525 * Clear any possible leftover bits:
527 memset(tss->io_bitmap, 0xff, prev->io_bitmap_max);
530 #ifdef CONFIG_X86_PTRACE_BTS
531 if (test_tsk_thread_flag(prev_p, TIF_BTS_TRACE_TS))
532 ptrace_bts_take_timestamp(prev_p, BTS_TASK_DEPARTS);
534 if (test_tsk_thread_flag(next_p, TIF_BTS_TRACE_TS))
535 ptrace_bts_take_timestamp(next_p, BTS_TASK_ARRIVES);
536 #endif /* CONFIG_X86_PTRACE_BTS */
540 * switch_to(x,y) should switch tasks from x to y.
542 * This could still be optimized:
543 * - fold all the options into a flag word and test it with a single test.
544 * - could test fs/gs bitsliced
546 * Kprobes not supported here. Set the probe on schedule instead.
549 __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
551 struct thread_struct *prev = &prev_p->thread;
552 struct thread_struct *next = &next_p->thread;
553 int cpu = smp_processor_id();
554 struct tss_struct *tss = &per_cpu(init_tss, cpu);
555 unsigned fsindex, gsindex;
557 /* we're going to use this soon, after a few expensive things */
558 if (next_p->fpu_counter > 5)
559 prefetch(next->xstate);
562 * Reload esp0, LDT and the page table pointer:
568 * This won't pick up thread selector changes, but I guess that is ok.
570 savesegment(es, prev->es);
571 if (unlikely(next->es | prev->es))
572 loadsegment(es, next->es);
574 savesegment(ds, prev->ds);
575 if (unlikely(next->ds | prev->ds))
576 loadsegment(ds, next->ds);
579 /* We must save %fs and %gs before load_TLS() because
580 * %fs and %gs may be cleared by load_TLS().
582 * (e.g. xen_load_tls())
584 savesegment(fs, fsindex);
585 savesegment(gs, gsindex);
590 * Leave lazy mode, flushing any hypercalls made here.
591 * This must be done before restoring TLS segments so
592 * the GDT and LDT are properly updated, and must be
593 * done before math_state_restore, so the TS bit is up
596 arch_leave_lazy_cpu_mode();
601 * Segment register != 0 always requires a reload. Also
602 * reload when it has changed. When prev process used 64bit
603 * base always reload to avoid an information leak.
605 if (unlikely(fsindex | next->fsindex | prev->fs)) {
606 loadsegment(fs, next->fsindex);
608 * Check if the user used a selector != 0; if yes
609 * clear 64bit base, since overloaded base is always
610 * mapped to the Null selector
615 /* when next process has a 64bit base use it */
617 wrmsrl(MSR_FS_BASE, next->fs);
618 prev->fsindex = fsindex;
620 if (unlikely(gsindex | next->gsindex | prev->gs)) {
621 load_gs_index(next->gsindex);
626 wrmsrl(MSR_KERNEL_GS_BASE, next->gs);
627 prev->gsindex = gsindex;
629 /* Must be after DS reload */
633 * Switch the PDA and FPU contexts.
635 prev->usersp = read_pda(oldrsp);
636 write_pda(oldrsp, next->usersp);
637 write_pda(pcurrent, next_p);
639 write_pda(kernelstack,
640 (unsigned long)task_stack_page(next_p) +
641 THREAD_SIZE - PDA_STACKOFFSET);
642 #ifdef CONFIG_CC_STACKPROTECTOR
643 write_pda(stack_canary, next_p->stack_canary);
645 * Build time only check to make sure the stack_canary is at
646 * offset 40 in the pda; this is a gcc ABI requirement
648 BUILD_BUG_ON(offsetof(struct x8664_pda, stack_canary) != 40);
652 * Now maybe reload the debug registers and handle I/O bitmaps
654 if (unlikely(task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT ||
655 task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV))
656 __switch_to_xtra(prev_p, next_p, tss);
658 /* If the task has used fpu the last 5 timeslices, just do a full
659 * restore of the math state immediately to avoid the trap; the
660 * chances of needing FPU soon are obviously high now
662 * tsk_used_math() checks prevent calling math_state_restore(),
663 * which can sleep in the case of !tsk_used_math()
665 if (tsk_used_math(next_p) && next_p->fpu_counter > 5)
666 math_state_restore();
671 * sys_execve() executes a new program.
674 long sys_execve(char __user *name, char __user * __user *argv,
675 char __user * __user *envp, struct pt_regs *regs)
680 filename = getname(name);
681 error = PTR_ERR(filename);
682 if (IS_ERR(filename))
684 error = do_execve(filename, argv, envp, regs);
689 void set_personality_64bit(void)
691 /* inherit personality from parent */
693 /* Make sure to be in 64bit mode */
694 clear_thread_flag(TIF_IA32);
696 /* TBD: overwrites user setup. Should have two bits.
697 But 64bit processes have always behaved this way,
698 so it's not too bad. The main problem is just that
699 32bit childs are affected again. */
700 current->personality &= ~READ_IMPLIES_EXEC;
703 asmlinkage long sys_fork(struct pt_regs *regs)
705 return do_fork(SIGCHLD, regs->sp, regs, 0, NULL, NULL);
709 sys_clone(unsigned long clone_flags, unsigned long newsp,
710 void __user *parent_tid, void __user *child_tid, struct pt_regs *regs)
714 return do_fork(clone_flags, newsp, regs, 0, parent_tid, child_tid);
718 * This is trivial, and on the face of it looks like it
719 * could equally well be done in user mode.
721 * Not so, for quite unobvious reasons - register pressure.
722 * In user mode vfork() cannot have a stack frame, and if
723 * done by calling the "clone()" system call directly, you
724 * do not have enough call-clobbered registers to hold all
725 * the information you need.
727 asmlinkage long sys_vfork(struct pt_regs *regs)
729 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->sp, regs, 0,
733 unsigned long get_wchan(struct task_struct *p)
739 if (!p || p == current || p->state == TASK_RUNNING)
741 stack = (unsigned long)task_stack_page(p);
742 if (p->thread.sp < stack || p->thread.sp >= stack+THREAD_SIZE)
744 fp = *(u64 *)(p->thread.sp);
746 if (fp < (unsigned long)stack ||
747 fp >= (unsigned long)stack+THREAD_SIZE)
750 if (!in_sched_functions(ip))
753 } while (count++ < 16);
757 long do_arch_prctl(struct task_struct *task, int code, unsigned long addr)
760 int doit = task == current;
765 if (addr >= TASK_SIZE_OF(task))
768 /* handle small bases via the GDT because that's faster to
770 if (addr <= 0xffffffff) {
771 set_32bit_tls(task, GS_TLS, addr);
773 load_TLS(&task->thread, cpu);
774 load_gs_index(GS_TLS_SEL);
776 task->thread.gsindex = GS_TLS_SEL;
779 task->thread.gsindex = 0;
780 task->thread.gs = addr;
783 ret = checking_wrmsrl(MSR_KERNEL_GS_BASE, addr);
789 /* Not strictly needed for fs, but do it for symmetry
791 if (addr >= TASK_SIZE_OF(task))
794 /* handle small bases via the GDT because that's faster to
796 if (addr <= 0xffffffff) {
797 set_32bit_tls(task, FS_TLS, addr);
799 load_TLS(&task->thread, cpu);
800 loadsegment(fs, FS_TLS_SEL);
802 task->thread.fsindex = FS_TLS_SEL;
805 task->thread.fsindex = 0;
806 task->thread.fs = addr;
808 /* set the selector to 0 to not confuse
811 ret = checking_wrmsrl(MSR_FS_BASE, addr);
818 if (task->thread.fsindex == FS_TLS_SEL)
819 base = read_32bit_tls(task, FS_TLS);
821 rdmsrl(MSR_FS_BASE, base);
823 base = task->thread.fs;
824 ret = put_user(base, (unsigned long __user *)addr);
830 if (task->thread.gsindex == GS_TLS_SEL)
831 base = read_32bit_tls(task, GS_TLS);
833 savesegment(gs, gsindex);
835 rdmsrl(MSR_KERNEL_GS_BASE, base);
837 base = task->thread.gs;
839 base = task->thread.gs;
840 ret = put_user(base, (unsigned long __user *)addr);
852 long sys_arch_prctl(int code, unsigned long addr)
854 return do_arch_prctl(current, code, addr);
857 unsigned long arch_align_stack(unsigned long sp)
859 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
860 sp -= get_random_int() % 8192;
864 unsigned long arch_randomize_brk(struct mm_struct *mm)
866 unsigned long range_end = mm->brk + 0x02000000;
867 return randomize_range(mm->brk, range_end, 0) ? : mm->brk;