2 * linux/arch/x86-64/kernel/process.c
4 * Copyright (C) 1995 Linus Torvalds
6 * Pentium III FXSR, SSE support
7 * Gareth Hughes <gareth@valinux.com>, May 2000
12 * CPU hotplug support - ashok.raj@intel.com
13 * $Id: process.c,v 1.38 2002/01/15 10:08:03 ak Exp $
17 * This file handles the architecture-dependent parts of process handling..
22 #include <linux/cpu.h>
23 #include <linux/errno.h>
24 #include <linux/sched.h>
25 #include <linux/kernel.h>
27 #include <linux/elfcore.h>
28 #include <linux/smp.h>
29 #include <linux/slab.h>
30 #include <linux/user.h>
31 #include <linux/module.h>
32 #include <linux/a.out.h>
33 #include <linux/interrupt.h>
34 #include <linux/delay.h>
35 #include <linux/irq.h>
36 #include <linux/ptrace.h>
37 #include <linux/utsname.h>
38 #include <linux/random.h>
39 #include <linux/kprobes.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>
50 #include <asm/kdebug.h>
52 #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_t hlt_counter = ATOMIC_INIT(0);
61 unsigned long boot_option_idle_override = 0;
62 EXPORT_SYMBOL(boot_option_idle_override);
65 * Powermanagement idle function, if any..
67 void (*pm_idle)(void);
68 static DEFINE_PER_CPU(unsigned int, cpu_idle_state);
70 void disable_hlt(void)
72 atomic_inc(&hlt_counter);
75 EXPORT_SYMBOL(disable_hlt);
79 atomic_dec(&hlt_counter);
82 EXPORT_SYMBOL(enable_hlt);
85 * We use this if we don't have any better
88 void default_idle(void)
90 if (!atomic_read(&hlt_counter)) {
100 * On SMP it's slightly faster (but much more power-consuming!)
101 * to poll the ->need_resched flag instead of waiting for the
102 * cross-CPU IPI to arrive. Use this option with caution.
104 static void poll_idle (void)
111 * Deal with another CPU just having chosen a thread to
114 oldval = test_and_clear_thread_flag(TIF_NEED_RESCHED);
117 set_thread_flag(TIF_POLLING_NRFLAG);
124 "i" (_TIF_NEED_RESCHED),
125 "m" (current_thread_info()->flags));
131 void cpu_idle_wait(void)
133 unsigned int cpu, this_cpu = get_cpu();
136 set_cpus_allowed(current, cpumask_of_cpu(this_cpu));
140 for_each_online_cpu(cpu) {
141 per_cpu(cpu_idle_state, cpu) = 1;
145 __get_cpu_var(cpu_idle_state) = 0;
150 for_each_online_cpu(cpu) {
151 if (cpu_isset(cpu, map) && !per_cpu(cpu_idle_state, cpu))
154 cpus_and(map, map, cpu_online_map);
155 } while (!cpus_empty(map));
157 EXPORT_SYMBOL_GPL(cpu_idle_wait);
159 #ifdef CONFIG_HOTPLUG_CPU
160 DECLARE_PER_CPU(int, cpu_state);
163 /* We don't actually take CPU down, just spin without interrupts. */
164 static inline void play_dead(void)
170 __get_cpu_var(cpu_state) = CPU_DEAD;
176 static inline void play_dead(void)
180 #endif /* CONFIG_HOTPLUG_CPU */
183 * The idle thread. There's no useful work to be
184 * done, so just try to conserve power and have a
185 * low exit latency (ie sit in a loop waiting for
186 * somebody to say that they'd like to reschedule)
190 /* endless idle loop with no priority at all */
192 while (!need_resched()) {
195 if (__get_cpu_var(cpu_idle_state))
196 __get_cpu_var(cpu_idle_state) = 0;
202 if (cpu_is_offline(smp_processor_id()))
212 * This uses new MONITOR/MWAIT instructions on P4 processors with PNI,
213 * which can obviate IPI to trigger checking of need_resched.
214 * We execute MONITOR against need_resched and enter optimized wait state
215 * through MWAIT. Whenever someone changes need_resched, we would be woken
216 * up from MWAIT (without an IPI).
218 static void mwait_idle(void)
222 if (!need_resched()) {
223 set_thread_flag(TIF_POLLING_NRFLAG);
225 __monitor((void *)¤t_thread_info()->flags, 0, 0);
229 } while (!need_resched());
230 clear_thread_flag(TIF_POLLING_NRFLAG);
234 void __cpuinit select_idle_routine(const struct cpuinfo_x86 *c)
237 if (cpu_has(c, X86_FEATURE_MWAIT)) {
239 * Skip, if setup has overridden idle.
240 * One CPU supports mwait => All CPUs supports mwait
244 printk("using mwait in idle threads.\n");
247 pm_idle = mwait_idle;
252 static int __init idle_setup (char *str)
254 if (!strncmp(str, "poll", 4)) {
255 printk("using polling idle threads.\n");
259 boot_option_idle_override = 1;
263 __setup("idle=", idle_setup);
265 /* Prints also some state that isn't saved in the pt_regs */
266 void __show_regs(struct pt_regs * regs)
268 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L, fs, gs, shadowgs;
269 unsigned int fsindex,gsindex;
270 unsigned int ds,cs,es;
274 printk("Pid: %d, comm: %.20s %s %s\n",
275 current->pid, current->comm, print_tainted(), system_utsname.release);
276 printk("RIP: %04lx:[<%016lx>] ", regs->cs & 0xffff, regs->rip);
277 printk_address(regs->rip);
278 printk("\nRSP: %04lx:%016lx EFLAGS: %08lx\n", regs->ss, regs->rsp, regs->eflags);
279 printk("RAX: %016lx RBX: %016lx RCX: %016lx\n",
280 regs->rax, regs->rbx, regs->rcx);
281 printk("RDX: %016lx RSI: %016lx RDI: %016lx\n",
282 regs->rdx, regs->rsi, regs->rdi);
283 printk("RBP: %016lx R08: %016lx R09: %016lx\n",
284 regs->rbp, regs->r8, regs->r9);
285 printk("R10: %016lx R11: %016lx R12: %016lx\n",
286 regs->r10, regs->r11, regs->r12);
287 printk("R13: %016lx R14: %016lx R15: %016lx\n",
288 regs->r13, regs->r14, regs->r15);
290 asm("movl %%ds,%0" : "=r" (ds));
291 asm("movl %%cs,%0" : "=r" (cs));
292 asm("movl %%es,%0" : "=r" (es));
293 asm("movl %%fs,%0" : "=r" (fsindex));
294 asm("movl %%gs,%0" : "=r" (gsindex));
296 rdmsrl(MSR_FS_BASE, fs);
297 rdmsrl(MSR_GS_BASE, gs);
298 rdmsrl(MSR_KERNEL_GS_BASE, shadowgs);
300 asm("movq %%cr0, %0": "=r" (cr0));
301 asm("movq %%cr2, %0": "=r" (cr2));
302 asm("movq %%cr3, %0": "=r" (cr3));
303 asm("movq %%cr4, %0": "=r" (cr4));
305 printk("FS: %016lx(%04x) GS:%016lx(%04x) knlGS:%016lx\n",
306 fs,fsindex,gs,gsindex,shadowgs);
307 printk("CS: %04x DS: %04x ES: %04x CR0: %016lx\n", cs, ds, es, cr0);
308 printk("CR2: %016lx CR3: %016lx CR4: %016lx\n", cr2, cr3, cr4);
311 void show_regs(struct pt_regs *regs)
313 printk("CPU %d:", smp_processor_id());
315 show_trace(®s->rsp);
319 * Free current thread data structures etc..
321 void exit_thread(void)
323 struct task_struct *me = current;
324 struct thread_struct *t = &me->thread;
327 * Remove function-return probe instances associated with this task
328 * and put them back on the free list. Do not insert an exit probe for
329 * this function, it will be disabled by kprobe_flush_task if you do.
331 kprobe_flush_task(me);
333 if (me->thread.io_bitmap_ptr) {
334 struct tss_struct *tss = &per_cpu(init_tss, get_cpu());
336 kfree(t->io_bitmap_ptr);
337 t->io_bitmap_ptr = NULL;
339 * Careful, clear this in the TSS too:
341 memset(tss->io_bitmap, 0xff, t->io_bitmap_max);
342 t->io_bitmap_max = 0;
347 void flush_thread(void)
349 struct task_struct *tsk = current;
350 struct thread_info *t = current_thread_info();
353 * Remove function-return probe instances associated with this task
354 * and put them back on the free list. Do not insert an exit probe for
355 * this function, it will be disabled by kprobe_flush_task if you do.
357 kprobe_flush_task(tsk);
359 if (t->flags & _TIF_ABI_PENDING)
360 t->flags ^= (_TIF_ABI_PENDING | _TIF_IA32);
362 tsk->thread.debugreg0 = 0;
363 tsk->thread.debugreg1 = 0;
364 tsk->thread.debugreg2 = 0;
365 tsk->thread.debugreg3 = 0;
366 tsk->thread.debugreg6 = 0;
367 tsk->thread.debugreg7 = 0;
368 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
370 * Forget coprocessor state..
376 void release_thread(struct task_struct *dead_task)
379 if (dead_task->mm->context.size) {
380 printk("WARNING: dead process %8s still has LDT? <%p/%d>\n",
382 dead_task->mm->context.ldt,
383 dead_task->mm->context.size);
389 static inline void set_32bit_tls(struct task_struct *t, int tls, u32 addr)
391 struct user_desc ud = {
398 struct n_desc_struct *desc = (void *)t->thread.tls_array;
400 desc->a = LDT_entry_a(&ud);
401 desc->b = LDT_entry_b(&ud);
404 static inline u32 read_32bit_tls(struct task_struct *t, int tls)
406 struct desc_struct *desc = (void *)t->thread.tls_array;
409 (((u32)desc->base1) << 16) |
410 (((u32)desc->base2) << 24);
414 * This gets called before we allocate a new thread and copy
415 * the current task into it.
417 void prepare_to_copy(struct task_struct *tsk)
422 int copy_thread(int nr, unsigned long clone_flags, unsigned long rsp,
423 unsigned long unused,
424 struct task_struct * p, struct pt_regs * regs)
427 struct pt_regs * childregs;
428 struct task_struct *me = current;
430 childregs = ((struct pt_regs *) (THREAD_SIZE + (unsigned long) p->thread_info)) - 1;
435 childregs->rsp = rsp;
437 childregs->rsp = (unsigned long)childregs;
440 p->thread.rsp = (unsigned long) childregs;
441 p->thread.rsp0 = (unsigned long) (childregs+1);
442 p->thread.userrsp = me->thread.userrsp;
444 set_ti_thread_flag(p->thread_info, TIF_FORK);
446 p->thread.fs = me->thread.fs;
447 p->thread.gs = me->thread.gs;
449 asm("mov %%gs,%0" : "=m" (p->thread.gsindex));
450 asm("mov %%fs,%0" : "=m" (p->thread.fsindex));
451 asm("mov %%es,%0" : "=m" (p->thread.es));
452 asm("mov %%ds,%0" : "=m" (p->thread.ds));
454 if (unlikely(me->thread.io_bitmap_ptr != NULL)) {
455 p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL);
456 if (!p->thread.io_bitmap_ptr) {
457 p->thread.io_bitmap_max = 0;
460 memcpy(p->thread.io_bitmap_ptr, me->thread.io_bitmap_ptr, IO_BITMAP_BYTES);
464 * Set a new TLS for the child thread?
466 if (clone_flags & CLONE_SETTLS) {
467 #ifdef CONFIG_IA32_EMULATION
468 if (test_thread_flag(TIF_IA32))
469 err = ia32_child_tls(p, childregs);
472 err = do_arch_prctl(p, ARCH_SET_FS, childregs->r8);
478 if (err && p->thread.io_bitmap_ptr) {
479 kfree(p->thread.io_bitmap_ptr);
480 p->thread.io_bitmap_max = 0;
486 * This function selects if the context switch from prev to next
487 * has to tweak the TSC disable bit in the cr4.
489 static inline void disable_tsc(struct task_struct *prev_p,
490 struct task_struct *next_p)
492 struct thread_info *prev, *next;
495 * gcc should eliminate the ->thread_info dereference if
496 * has_secure_computing returns 0 at compile time (SECCOMP=n).
498 prev = prev_p->thread_info;
499 next = next_p->thread_info;
501 if (has_secure_computing(prev) || has_secure_computing(next)) {
503 if (has_secure_computing(prev) &&
504 !has_secure_computing(next)) {
505 write_cr4(read_cr4() & ~X86_CR4_TSD);
506 } else if (!has_secure_computing(prev) &&
507 has_secure_computing(next))
508 write_cr4(read_cr4() | X86_CR4_TSD);
513 * This special macro can be used to load a debugging register
515 #define loaddebug(thread,r) set_debug(thread->debugreg ## r, r)
518 * switch_to(x,y) should switch tasks from x to y.
520 * This could still be optimized:
521 * - fold all the options into a flag word and test it with a single test.
522 * - could test fs/gs bitsliced
524 struct task_struct *__switch_to(struct task_struct *prev_p, struct task_struct *next_p)
526 struct thread_struct *prev = &prev_p->thread,
527 *next = &next_p->thread;
528 int cpu = smp_processor_id();
529 struct tss_struct *tss = &per_cpu(init_tss, cpu);
534 * Reload esp0, LDT and the page table pointer:
536 tss->rsp0 = next->rsp0;
540 * This won't pick up thread selector changes, but I guess that is ok.
542 asm volatile("mov %%es,%0" : "=m" (prev->es));
543 if (unlikely(next->es | prev->es))
544 loadsegment(es, next->es);
546 asm volatile ("mov %%ds,%0" : "=m" (prev->ds));
547 if (unlikely(next->ds | prev->ds))
548 loadsegment(ds, next->ds);
557 asm volatile("movl %%fs,%0" : "=r" (fsindex));
558 /* segment register != 0 always requires a reload.
559 also reload when it has changed.
560 when prev process used 64bit base always reload
561 to avoid an information leak. */
562 if (unlikely(fsindex | next->fsindex | prev->fs)) {
563 loadsegment(fs, next->fsindex);
564 /* check if the user used a selector != 0
565 * if yes clear 64bit base, since overloaded base
566 * is always mapped to the Null selector
571 /* when next process has a 64bit base use it */
573 wrmsrl(MSR_FS_BASE, next->fs);
574 prev->fsindex = fsindex;
578 asm volatile("movl %%gs,%0" : "=r" (gsindex));
579 if (unlikely(gsindex | next->gsindex | prev->gs)) {
580 load_gs_index(next->gsindex);
585 wrmsrl(MSR_KERNEL_GS_BASE, next->gs);
586 prev->gsindex = gsindex;
590 * Switch the PDA context.
592 prev->userrsp = read_pda(oldrsp);
593 write_pda(oldrsp, next->userrsp);
594 write_pda(pcurrent, next_p);
595 write_pda(kernelstack, (unsigned long)next_p->thread_info + THREAD_SIZE - PDA_STACKOFFSET);
598 * Now maybe reload the debug registers
600 if (unlikely(next->debugreg7)) {
612 * Handle the IO bitmap
614 if (unlikely(prev->io_bitmap_ptr || next->io_bitmap_ptr)) {
615 if (next->io_bitmap_ptr)
617 * Copy the relevant range of the IO bitmap.
618 * Normally this is 128 bytes or less:
620 memcpy(tss->io_bitmap, next->io_bitmap_ptr,
621 max(prev->io_bitmap_max, next->io_bitmap_max));
624 * Clear any possible leftover bits:
626 memset(tss->io_bitmap, 0xff, prev->io_bitmap_max);
630 disable_tsc(prev_p, next_p);
636 * sys_execve() executes a new program.
639 long sys_execve(char __user *name, char __user * __user *argv,
640 char __user * __user *envp, struct pt_regs regs)
645 filename = getname(name);
646 error = PTR_ERR(filename);
647 if (IS_ERR(filename))
649 error = do_execve(filename, argv, envp, ®s);
652 current->ptrace &= ~PT_DTRACE;
653 task_unlock(current);
659 void set_personality_64bit(void)
661 /* inherit personality from parent */
663 /* Make sure to be in 64bit mode */
664 clear_thread_flag(TIF_IA32);
666 /* TBD: overwrites user setup. Should have two bits.
667 But 64bit processes have always behaved this way,
668 so it's not too bad. The main problem is just that
669 32bit childs are affected again. */
670 current->personality &= ~READ_IMPLIES_EXEC;
673 asmlinkage long sys_fork(struct pt_regs *regs)
675 return do_fork(SIGCHLD, regs->rsp, regs, 0, NULL, NULL);
678 asmlinkage long sys_clone(unsigned long clone_flags, unsigned long newsp, void __user *parent_tid, void __user *child_tid, struct pt_regs *regs)
682 return do_fork(clone_flags, newsp, regs, 0, parent_tid, child_tid);
686 * This is trivial, and on the face of it looks like it
687 * could equally well be done in user mode.
689 * Not so, for quite unobvious reasons - register pressure.
690 * In user mode vfork() cannot have a stack frame, and if
691 * done by calling the "clone()" system call directly, you
692 * do not have enough call-clobbered registers to hold all
693 * the information you need.
695 asmlinkage long sys_vfork(struct pt_regs *regs)
697 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->rsp, regs, 0,
701 unsigned long get_wchan(struct task_struct *p)
707 if (!p || p == current || p->state==TASK_RUNNING)
709 stack = (unsigned long)p->thread_info;
710 if (p->thread.rsp < stack || p->thread.rsp > stack+THREAD_SIZE)
712 fp = *(u64 *)(p->thread.rsp);
714 if (fp < (unsigned long)stack || fp > (unsigned long)stack+THREAD_SIZE)
716 rip = *(u64 *)(fp+8);
717 if (!in_sched_functions(rip))
720 } while (count++ < 16);
724 long do_arch_prctl(struct task_struct *task, int code, unsigned long addr)
727 int doit = task == current;
732 if (addr >= TASK_SIZE_OF(task))
735 /* handle small bases via the GDT because that's faster to
737 if (addr <= 0xffffffff) {
738 set_32bit_tls(task, GS_TLS, addr);
740 load_TLS(&task->thread, cpu);
741 load_gs_index(GS_TLS_SEL);
743 task->thread.gsindex = GS_TLS_SEL;
746 task->thread.gsindex = 0;
747 task->thread.gs = addr;
750 ret = checking_wrmsrl(MSR_KERNEL_GS_BASE, addr);
756 /* Not strictly needed for fs, but do it for symmetry
758 if (addr >= TASK_SIZE_OF(task))
761 /* handle small bases via the GDT because that's faster to
763 if (addr <= 0xffffffff) {
764 set_32bit_tls(task, FS_TLS, addr);
766 load_TLS(&task->thread, cpu);
767 asm volatile("movl %0,%%fs" :: "r" (FS_TLS_SEL));
769 task->thread.fsindex = FS_TLS_SEL;
772 task->thread.fsindex = 0;
773 task->thread.fs = addr;
775 /* set the selector to 0 to not confuse
777 asm volatile("movl %0,%%fs" :: "r" (0));
778 ret = checking_wrmsrl(MSR_FS_BASE, addr);
785 if (task->thread.fsindex == FS_TLS_SEL)
786 base = read_32bit_tls(task, FS_TLS);
788 rdmsrl(MSR_FS_BASE, base);
790 base = task->thread.fs;
791 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 rdmsrl(MSR_KERNEL_GS_BASE, base);
801 base = task->thread.gs;
802 ret = put_user(base, (unsigned long __user *)addr);
814 long sys_arch_prctl(int code, unsigned long addr)
816 return do_arch_prctl(current, code, addr);
820 * Capture the user space registers if the task is not running (in user space)
822 int dump_task_regs(struct task_struct *tsk, elf_gregset_t *regs)
824 struct pt_regs *pp, ptregs;
826 pp = (struct pt_regs *)(tsk->thread.rsp0);
833 elf_core_copy_regs(regs, &ptregs);
838 unsigned long arch_align_stack(unsigned long sp)
840 if (randomize_va_space)
841 sp -= get_random_int() % 8192;