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 static ATOMIC_NOTIFIER_HEAD(idle_notifier);
61 void idle_notifier_register(struct notifier_block *n)
63 atomic_notifier_chain_register(&idle_notifier, n);
69 atomic_notifier_call_chain(&idle_notifier, IDLE_START, NULL);
72 static void __exit_idle(void)
74 if (test_and_clear_bit_pda(0, isidle) == 0)
76 atomic_notifier_call_chain(&idle_notifier, IDLE_END, NULL);
79 /* Called from interrupts to signify idle end */
82 /* idle loop has pid 0 */
88 #ifdef CONFIG_HOTPLUG_CPU
89 DECLARE_PER_CPU(int, cpu_state);
92 /* We halt the CPU with physical CPU hotplug */
93 static inline void play_dead(void)
99 __get_cpu_var(cpu_state) = CPU_DEAD;
106 static inline void play_dead(void)
110 #endif /* CONFIG_HOTPLUG_CPU */
113 * The idle thread. There's no useful work to be
114 * done, so just try to conserve power and have a
115 * low exit latency (ie sit in a loop waiting for
116 * somebody to say that they'd like to reschedule)
120 current_thread_info()->status |= TS_POLLING;
121 /* endless idle loop with no priority at all */
123 tick_nohz_stop_sched_tick();
124 while (!need_resched()) {
128 if (cpu_is_offline(smp_processor_id()))
131 * Idle routines should keep interrupts disabled
132 * from here on, until they go to idle.
133 * Otherwise, idle callbacks can misfire.
138 /* In many cases the interrupt that ended idle
139 has already called exit_idle. But some idle
140 loops can be woken up without interrupt. */
144 tick_nohz_restart_sched_tick();
145 preempt_enable_no_resched();
151 /* Prints also some state that isn't saved in the pt_regs */
152 void __show_regs(struct pt_regs * regs)
154 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L, fs, gs, shadowgs;
155 unsigned long d0, d1, d2, d3, d6, d7;
156 unsigned int fsindex, gsindex;
157 unsigned int ds, cs, es;
161 printk("Pid: %d, comm: %.20s %s %s %.*s\n",
162 current->pid, current->comm, print_tainted(),
163 init_utsname()->release,
164 (int)strcspn(init_utsname()->version, " "),
165 init_utsname()->version);
166 printk("RIP: %04lx:[<%016lx>] ", regs->cs & 0xffff, regs->ip);
167 printk_address(regs->ip, 1);
168 printk("RSP: %04lx:%016lx EFLAGS: %08lx\n", regs->ss, regs->sp,
170 printk("RAX: %016lx RBX: %016lx RCX: %016lx\n",
171 regs->ax, regs->bx, regs->cx);
172 printk("RDX: %016lx RSI: %016lx RDI: %016lx\n",
173 regs->dx, regs->si, regs->di);
174 printk("RBP: %016lx R08: %016lx R09: %016lx\n",
175 regs->bp, regs->r8, regs->r9);
176 printk("R10: %016lx R11: %016lx R12: %016lx\n",
177 regs->r10, regs->r11, regs->r12);
178 printk("R13: %016lx R14: %016lx R15: %016lx\n",
179 regs->r13, regs->r14, regs->r15);
181 asm("movl %%ds,%0" : "=r" (ds));
182 asm("movl %%cs,%0" : "=r" (cs));
183 asm("movl %%es,%0" : "=r" (es));
184 asm("movl %%fs,%0" : "=r" (fsindex));
185 asm("movl %%gs,%0" : "=r" (gsindex));
187 rdmsrl(MSR_FS_BASE, fs);
188 rdmsrl(MSR_GS_BASE, gs);
189 rdmsrl(MSR_KERNEL_GS_BASE, shadowgs);
196 printk("FS: %016lx(%04x) GS:%016lx(%04x) knlGS:%016lx\n",
197 fs,fsindex,gs,gsindex,shadowgs);
198 printk("CS: %04x DS: %04x ES: %04x CR0: %016lx\n", cs, ds, es, cr0);
199 printk("CR2: %016lx CR3: %016lx CR4: %016lx\n", cr2, cr3, cr4);
204 printk("DR0: %016lx DR1: %016lx DR2: %016lx\n", d0, d1, d2);
208 printk("DR3: %016lx DR6: %016lx DR7: %016lx\n", d3, d6, d7);
211 void show_regs(struct pt_regs *regs)
213 printk("CPU %d:", smp_processor_id());
215 show_trace(NULL, regs, (void *)(regs + 1), regs->bp);
219 * Free current thread data structures etc..
221 void exit_thread(void)
223 struct task_struct *me = current;
224 struct thread_struct *t = &me->thread;
226 if (me->thread.io_bitmap_ptr) {
227 struct tss_struct *tss = &per_cpu(init_tss, get_cpu());
229 kfree(t->io_bitmap_ptr);
230 t->io_bitmap_ptr = NULL;
231 clear_thread_flag(TIF_IO_BITMAP);
233 * Careful, clear this in the TSS too:
235 memset(tss->io_bitmap, 0xff, t->io_bitmap_max);
236 t->io_bitmap_max = 0;
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..
270 void release_thread(struct task_struct *dead_task)
273 if (dead_task->mm->context.size) {
274 printk("WARNING: dead process %8s still has LDT? <%p/%d>\n",
276 dead_task->mm->context.ldt,
277 dead_task->mm->context.size);
283 static inline void set_32bit_tls(struct task_struct *t, int tls, u32 addr)
285 struct user_desc ud = {
292 struct desc_struct *desc = t->thread.tls_array;
297 static inline u32 read_32bit_tls(struct task_struct *t, int tls)
299 return get_desc_base(&t->thread.tls_array[tls]);
303 * This gets called before we allocate a new thread and copy
304 * the current task into it.
306 void prepare_to_copy(struct task_struct *tsk)
311 int copy_thread(int nr, unsigned long clone_flags, unsigned long sp,
312 unsigned long unused,
313 struct task_struct * p, struct pt_regs * regs)
316 struct pt_regs * childregs;
317 struct task_struct *me = current;
319 childregs = ((struct pt_regs *)
320 (THREAD_SIZE + task_stack_page(p))) - 1;
326 childregs->sp = (unsigned long)childregs;
328 p->thread.sp = (unsigned long) childregs;
329 p->thread.sp0 = (unsigned long) (childregs+1);
330 p->thread.usersp = me->thread.usersp;
332 set_tsk_thread_flag(p, TIF_FORK);
334 p->thread.fs = me->thread.fs;
335 p->thread.gs = me->thread.gs;
337 asm("mov %%gs,%0" : "=m" (p->thread.gsindex));
338 asm("mov %%fs,%0" : "=m" (p->thread.fsindex));
339 asm("mov %%es,%0" : "=m" (p->thread.es));
340 asm("mov %%ds,%0" : "=m" (p->thread.ds));
342 if (unlikely(test_tsk_thread_flag(me, TIF_IO_BITMAP))) {
343 p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL);
344 if (!p->thread.io_bitmap_ptr) {
345 p->thread.io_bitmap_max = 0;
348 memcpy(p->thread.io_bitmap_ptr, me->thread.io_bitmap_ptr,
350 set_tsk_thread_flag(p, TIF_IO_BITMAP);
354 * Set a new TLS for the child thread?
356 if (clone_flags & CLONE_SETTLS) {
357 #ifdef CONFIG_IA32_EMULATION
358 if (test_thread_flag(TIF_IA32))
359 err = do_set_thread_area(p, -1,
360 (struct user_desc __user *)childregs->si, 0);
363 err = do_arch_prctl(p, ARCH_SET_FS, childregs->r8);
369 if (err && p->thread.io_bitmap_ptr) {
370 kfree(p->thread.io_bitmap_ptr);
371 p->thread.io_bitmap_max = 0;
377 start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp)
379 asm volatile("movl %0, %%fs; movl %0, %%es; movl %0, %%ds" :: "r"(0));
383 write_pda(oldrsp, new_sp);
384 regs->cs = __USER_CS;
385 regs->ss = __USER_DS;
389 * Free the old FP and other extended state
391 free_thread_xstate(current);
393 EXPORT_SYMBOL_GPL(start_thread);
395 static void hard_disable_TSC(void)
397 write_cr4(read_cr4() | X86_CR4_TSD);
400 void disable_TSC(void)
403 if (!test_and_set_thread_flag(TIF_NOTSC))
405 * Must flip the CPU state synchronously with
406 * TIF_NOTSC in the current running context.
412 static void hard_enable_TSC(void)
414 write_cr4(read_cr4() & ~X86_CR4_TSD);
417 static void enable_TSC(void)
420 if (test_and_clear_thread_flag(TIF_NOTSC))
422 * Must flip the CPU state synchronously with
423 * TIF_NOTSC in the current running context.
429 int get_tsc_mode(unsigned long adr)
433 if (test_thread_flag(TIF_NOTSC))
434 val = PR_TSC_SIGSEGV;
438 return put_user(val, (unsigned int __user *)adr);
441 int set_tsc_mode(unsigned int val)
443 if (val == PR_TSC_SIGSEGV)
445 else if (val == PR_TSC_ENABLE)
454 * This special macro can be used to load a debugging register
456 #define loaddebug(thread, r) set_debugreg(thread->debugreg ## r, r)
458 static inline void __switch_to_xtra(struct task_struct *prev_p,
459 struct task_struct *next_p,
460 struct tss_struct *tss)
462 struct thread_struct *prev, *next;
463 unsigned long debugctl;
465 prev = &prev_p->thread,
466 next = &next_p->thread;
468 debugctl = prev->debugctlmsr;
469 if (next->ds_area_msr != prev->ds_area_msr) {
470 /* we clear debugctl to make sure DS
471 * is not in use when we change it */
473 update_debugctlmsr(0);
474 wrmsrl(MSR_IA32_DS_AREA, next->ds_area_msr);
477 if (next->debugctlmsr != debugctl)
478 update_debugctlmsr(next->debugctlmsr);
480 if (test_tsk_thread_flag(next_p, TIF_DEBUG)) {
490 if (test_tsk_thread_flag(prev_p, TIF_NOTSC) ^
491 test_tsk_thread_flag(next_p, TIF_NOTSC)) {
492 /* prev and next are different */
493 if (test_tsk_thread_flag(next_p, TIF_NOTSC))
499 if (test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
501 * Copy the relevant range of the IO bitmap.
502 * Normally this is 128 bytes or less:
504 memcpy(tss->io_bitmap, next->io_bitmap_ptr,
505 max(prev->io_bitmap_max, next->io_bitmap_max));
506 } else if (test_tsk_thread_flag(prev_p, TIF_IO_BITMAP)) {
508 * Clear any possible leftover bits:
510 memset(tss->io_bitmap, 0xff, prev->io_bitmap_max);
514 if (test_tsk_thread_flag(prev_p, TIF_BTS_TRACE_TS))
515 ptrace_bts_take_timestamp(prev_p, BTS_TASK_DEPARTS);
517 if (test_tsk_thread_flag(next_p, TIF_BTS_TRACE_TS))
518 ptrace_bts_take_timestamp(next_p, BTS_TASK_ARRIVES);
523 * switch_to(x,y) should switch tasks from x to y.
525 * This could still be optimized:
526 * - fold all the options into a flag word and test it with a single test.
527 * - could test fs/gs bitsliced
529 * Kprobes not supported here. Set the probe on schedule instead.
532 __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
534 struct thread_struct *prev = &prev_p->thread,
535 *next = &next_p->thread;
536 int cpu = smp_processor_id();
537 struct tss_struct *tss = &per_cpu(init_tss, cpu);
539 /* we're going to use this soon, after a few expensive things */
540 if (next_p->fpu_counter>5)
541 prefetch(next->xstate);
544 * Reload esp0, LDT and the page table pointer:
550 * This won't pick up thread selector changes, but I guess that is ok.
552 asm volatile("mov %%es,%0" : "=m" (prev->es));
553 if (unlikely(next->es | prev->es))
554 loadsegment(es, next->es);
556 asm volatile ("mov %%ds,%0" : "=m" (prev->ds));
557 if (unlikely(next->ds | prev->ds))
558 loadsegment(ds, next->ds);
567 asm volatile("movl %%fs,%0" : "=r" (fsindex));
568 /* segment register != 0 always requires a reload.
569 also reload when it has changed.
570 when prev process used 64bit base always reload
571 to avoid an information leak. */
572 if (unlikely(fsindex | next->fsindex | prev->fs)) {
573 loadsegment(fs, next->fsindex);
574 /* check if the user used a selector != 0
575 * if yes clear 64bit base, since overloaded base
576 * is always mapped to the Null selector
581 /* when next process has a 64bit base use it */
583 wrmsrl(MSR_FS_BASE, next->fs);
584 prev->fsindex = fsindex;
588 asm volatile("movl %%gs,%0" : "=r" (gsindex));
589 if (unlikely(gsindex | next->gsindex | prev->gs)) {
590 load_gs_index(next->gsindex);
595 wrmsrl(MSR_KERNEL_GS_BASE, next->gs);
596 prev->gsindex = gsindex;
599 /* Must be after DS reload */
603 * Switch the PDA and FPU contexts.
605 prev->usersp = read_pda(oldrsp);
606 write_pda(oldrsp, next->usersp);
607 write_pda(pcurrent, next_p);
609 write_pda(kernelstack,
610 (unsigned long)task_stack_page(next_p) + THREAD_SIZE - PDA_STACKOFFSET);
611 #ifdef CONFIG_CC_STACKPROTECTOR
612 write_pda(stack_canary, next_p->stack_canary);
614 * Build time only check to make sure the stack_canary is at
615 * offset 40 in the pda; this is a gcc ABI requirement
617 BUILD_BUG_ON(offsetof(struct x8664_pda, stack_canary) != 40);
621 * Now maybe reload the debug registers and handle I/O bitmaps
623 if (unlikely(task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT ||
624 task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV))
625 __switch_to_xtra(prev_p, next_p, tss);
627 /* If the task has used fpu the last 5 timeslices, just do a full
628 * restore of the math state immediately to avoid the trap; the
629 * chances of needing FPU soon are obviously high now
631 if (next_p->fpu_counter>5)
632 math_state_restore();
637 * sys_execve() executes a new program.
640 long sys_execve(char __user *name, char __user * __user *argv,
641 char __user * __user *envp, struct pt_regs *regs)
646 filename = getname(name);
647 error = PTR_ERR(filename);
648 if (IS_ERR(filename))
650 error = do_execve(filename, argv, envp, regs);
655 void set_personality_64bit(void)
657 /* inherit personality from parent */
659 /* Make sure to be in 64bit mode */
660 clear_thread_flag(TIF_IA32);
662 /* TBD: overwrites user setup. Should have two bits.
663 But 64bit processes have always behaved this way,
664 so it's not too bad. The main problem is just that
665 32bit childs are affected again. */
666 current->personality &= ~READ_IMPLIES_EXEC;
669 asmlinkage long sys_fork(struct pt_regs *regs)
671 return do_fork(SIGCHLD, regs->sp, regs, 0, NULL, NULL);
675 sys_clone(unsigned long clone_flags, unsigned long newsp,
676 void __user *parent_tid, void __user *child_tid, struct pt_regs *regs)
680 return do_fork(clone_flags, newsp, regs, 0, parent_tid, child_tid);
684 * This is trivial, and on the face of it looks like it
685 * could equally well be done in user mode.
687 * Not so, for quite unobvious reasons - register pressure.
688 * In user mode vfork() cannot have a stack frame, and if
689 * done by calling the "clone()" system call directly, you
690 * do not have enough call-clobbered registers to hold all
691 * the information you need.
693 asmlinkage long sys_vfork(struct pt_regs *regs)
695 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->sp, regs, 0,
699 unsigned long get_wchan(struct task_struct *p)
705 if (!p || p == current || p->state==TASK_RUNNING)
707 stack = (unsigned long)task_stack_page(p);
708 if (p->thread.sp < stack || p->thread.sp > stack+THREAD_SIZE)
710 fp = *(u64 *)(p->thread.sp);
712 if (fp < (unsigned long)stack ||
713 fp > (unsigned long)stack+THREAD_SIZE)
716 if (!in_sched_functions(ip))
719 } while (count++ < 16);
723 long do_arch_prctl(struct task_struct *task, int code, unsigned long addr)
726 int doit = task == current;
731 if (addr >= TASK_SIZE_OF(task))
734 /* handle small bases via the GDT because that's faster to
736 if (addr <= 0xffffffff) {
737 set_32bit_tls(task, GS_TLS, addr);
739 load_TLS(&task->thread, cpu);
740 load_gs_index(GS_TLS_SEL);
742 task->thread.gsindex = GS_TLS_SEL;
745 task->thread.gsindex = 0;
746 task->thread.gs = addr;
749 ret = checking_wrmsrl(MSR_KERNEL_GS_BASE, addr);
755 /* Not strictly needed for fs, but do it for symmetry
757 if (addr >= TASK_SIZE_OF(task))
760 /* handle small bases via the GDT because that's faster to
762 if (addr <= 0xffffffff) {
763 set_32bit_tls(task, FS_TLS, addr);
765 load_TLS(&task->thread, cpu);
766 asm volatile("movl %0,%%fs" :: "r"(FS_TLS_SEL));
768 task->thread.fsindex = FS_TLS_SEL;
771 task->thread.fsindex = 0;
772 task->thread.fs = addr;
774 /* set the selector to 0 to not confuse
776 asm volatile("movl %0,%%fs" :: "r" (0));
777 ret = checking_wrmsrl(MSR_FS_BASE, addr);
784 if (task->thread.fsindex == FS_TLS_SEL)
785 base = read_32bit_tls(task, FS_TLS);
787 rdmsrl(MSR_FS_BASE, base);
789 base = task->thread.fs;
790 ret = put_user(base, (unsigned long __user *)addr);
796 if (task->thread.gsindex == GS_TLS_SEL)
797 base = read_32bit_tls(task, GS_TLS);
799 asm("movl %%gs,%0" : "=r" (gsindex));
801 rdmsrl(MSR_KERNEL_GS_BASE, base);
803 base = task->thread.gs;
806 base = task->thread.gs;
807 ret = put_user(base, (unsigned long __user *)addr);
819 long sys_arch_prctl(int code, unsigned long addr)
821 return do_arch_prctl(current, code, addr);
824 unsigned long arch_align_stack(unsigned long sp)
826 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
827 sp -= get_random_int() % 8192;
831 unsigned long arch_randomize_brk(struct mm_struct *mm)
833 unsigned long range_end = mm->brk + 0x02000000;
834 return randomize_range(mm->brk, range_end, 0) ? : mm->brk;