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 * $Id: process.c,v 1.38 2002/01/15 10:08:03 ak Exp $
16 * This file handles the architecture-dependent parts of process handling..
21 #include <linux/errno.h>
22 #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/module.h>
30 #include <linux/a.out.h>
31 #include <linux/interrupt.h>
32 #include <linux/delay.h>
33 #include <linux/irq.h>
34 #include <linux/ptrace.h>
35 #include <linux/utsname.h>
36 #include <linux/random.h>
38 #include <asm/uaccess.h>
39 #include <asm/pgtable.h>
40 #include <asm/system.h>
42 #include <asm/processor.h>
44 #include <asm/mmu_context.h>
46 #include <asm/prctl.h>
47 #include <asm/kdebug.h>
49 #include <asm/proto.h>
52 asmlinkage extern void ret_from_fork(void);
54 unsigned long kernel_thread_flags = CLONE_VM | CLONE_UNTRACED;
56 static atomic_t hlt_counter = ATOMIC_INIT(0);
58 unsigned long boot_option_idle_override = 0;
59 EXPORT_SYMBOL(boot_option_idle_override);
62 * Powermanagement idle function, if any..
64 void (*pm_idle)(void);
65 static DEFINE_PER_CPU(unsigned int, cpu_idle_state);
67 void disable_hlt(void)
69 atomic_inc(&hlt_counter);
72 EXPORT_SYMBOL(disable_hlt);
76 atomic_dec(&hlt_counter);
79 EXPORT_SYMBOL(enable_hlt);
82 * We use this if we don't have any better
85 void default_idle(void)
87 if (!atomic_read(&hlt_counter)) {
97 * On SMP it's slightly faster (but much more power-consuming!)
98 * to poll the ->need_resched flag instead of waiting for the
99 * cross-CPU IPI to arrive. Use this option with caution.
101 static void poll_idle (void)
108 * Deal with another CPU just having chosen a thread to
111 oldval = test_and_clear_thread_flag(TIF_NEED_RESCHED);
114 set_thread_flag(TIF_POLLING_NRFLAG);
121 "i" (_TIF_NEED_RESCHED),
122 "m" (current_thread_info()->flags));
128 void cpu_idle_wait(void)
130 unsigned int cpu, this_cpu = get_cpu();
133 set_cpus_allowed(current, cpumask_of_cpu(this_cpu));
137 for_each_online_cpu(cpu) {
138 per_cpu(cpu_idle_state, cpu) = 1;
142 __get_cpu_var(cpu_idle_state) = 0;
147 for_each_online_cpu(cpu) {
148 if (cpu_isset(cpu, map) && !per_cpu(cpu_idle_state, cpu))
151 cpus_and(map, map, cpu_online_map);
152 } while (!cpus_empty(map));
154 EXPORT_SYMBOL_GPL(cpu_idle_wait);
157 * The idle thread. There's no useful work to be
158 * done, so just try to conserve power and have a
159 * low exit latency (ie sit in a loop waiting for
160 * somebody to say that they'd like to reschedule)
164 /* endless idle loop with no priority at all */
166 while (!need_resched()) {
169 if (__get_cpu_var(cpu_idle_state))
170 __get_cpu_var(cpu_idle_state) = 0;
184 * This uses new MONITOR/MWAIT instructions on P4 processors with PNI,
185 * which can obviate IPI to trigger checking of need_resched.
186 * We execute MONITOR against need_resched and enter optimized wait state
187 * through MWAIT. Whenever someone changes need_resched, we would be woken
188 * up from MWAIT (without an IPI).
190 static void mwait_idle(void)
194 if (!need_resched()) {
195 set_thread_flag(TIF_POLLING_NRFLAG);
197 __monitor((void *)¤t_thread_info()->flags, 0, 0);
201 } while (!need_resched());
202 clear_thread_flag(TIF_POLLING_NRFLAG);
206 void __init select_idle_routine(const struct cpuinfo_x86 *c)
209 if (cpu_has(c, X86_FEATURE_MWAIT)) {
211 * Skip, if setup has overridden idle.
212 * One CPU supports mwait => All CPUs supports mwait
216 printk("using mwait in idle threads.\n");
219 pm_idle = mwait_idle;
224 static int __init idle_setup (char *str)
226 if (!strncmp(str, "poll", 4)) {
227 printk("using polling idle threads.\n");
231 boot_option_idle_override = 1;
235 __setup("idle=", idle_setup);
237 /* Prints also some state that isn't saved in the pt_regs */
238 void __show_regs(struct pt_regs * regs)
240 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L, fs, gs, shadowgs;
241 unsigned int fsindex,gsindex;
242 unsigned int ds,cs,es;
246 printk("Pid: %d, comm: %.20s %s %s\n",
247 current->pid, current->comm, print_tainted(), system_utsname.release);
248 printk("RIP: %04lx:[<%016lx>] ", regs->cs & 0xffff, regs->rip);
249 printk_address(regs->rip);
250 printk("\nRSP: %04lx:%016lx EFLAGS: %08lx\n", regs->ss, regs->rsp, regs->eflags);
251 printk("RAX: %016lx RBX: %016lx RCX: %016lx\n",
252 regs->rax, regs->rbx, regs->rcx);
253 printk("RDX: %016lx RSI: %016lx RDI: %016lx\n",
254 regs->rdx, regs->rsi, regs->rdi);
255 printk("RBP: %016lx R08: %016lx R09: %016lx\n",
256 regs->rbp, regs->r8, regs->r9);
257 printk("R10: %016lx R11: %016lx R12: %016lx\n",
258 regs->r10, regs->r11, regs->r12);
259 printk("R13: %016lx R14: %016lx R15: %016lx\n",
260 regs->r13, regs->r14, regs->r15);
262 asm("movl %%ds,%0" : "=r" (ds));
263 asm("movl %%cs,%0" : "=r" (cs));
264 asm("movl %%es,%0" : "=r" (es));
265 asm("movl %%fs,%0" : "=r" (fsindex));
266 asm("movl %%gs,%0" : "=r" (gsindex));
268 rdmsrl(MSR_FS_BASE, fs);
269 rdmsrl(MSR_GS_BASE, gs);
270 rdmsrl(MSR_KERNEL_GS_BASE, shadowgs);
272 asm("movq %%cr0, %0": "=r" (cr0));
273 asm("movq %%cr2, %0": "=r" (cr2));
274 asm("movq %%cr3, %0": "=r" (cr3));
275 asm("movq %%cr4, %0": "=r" (cr4));
277 printk("FS: %016lx(%04x) GS:%016lx(%04x) knlGS:%016lx\n",
278 fs,fsindex,gs,gsindex,shadowgs);
279 printk("CS: %04x DS: %04x ES: %04x CR0: %016lx\n", cs, ds, es, cr0);
280 printk("CR2: %016lx CR3: %016lx CR4: %016lx\n", cr2, cr3, cr4);
283 void show_regs(struct pt_regs *regs)
286 show_trace(®s->rsp);
289 EXPORT_SYMBOL_GPL(show_regs);
292 * Free current thread data structures etc..
294 void exit_thread(void)
296 struct task_struct *me = current;
297 struct thread_struct *t = &me->thread;
298 if (me->thread.io_bitmap_ptr) {
299 struct tss_struct *tss = &per_cpu(init_tss, get_cpu());
301 kfree(t->io_bitmap_ptr);
302 t->io_bitmap_ptr = NULL;
304 * Careful, clear this in the TSS too:
306 memset(tss->io_bitmap, 0xff, t->io_bitmap_max);
307 t->io_bitmap_max = 0;
312 void flush_thread(void)
314 struct task_struct *tsk = current;
315 struct thread_info *t = current_thread_info();
317 if (t->flags & _TIF_ABI_PENDING)
318 t->flags ^= (_TIF_ABI_PENDING | _TIF_IA32);
320 tsk->thread.debugreg0 = 0;
321 tsk->thread.debugreg1 = 0;
322 tsk->thread.debugreg2 = 0;
323 tsk->thread.debugreg3 = 0;
324 tsk->thread.debugreg6 = 0;
325 tsk->thread.debugreg7 = 0;
326 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
328 * Forget coprocessor state..
334 void release_thread(struct task_struct *dead_task)
337 if (dead_task->mm->context.size) {
338 printk("WARNING: dead process %8s still has LDT? <%p/%d>\n",
340 dead_task->mm->context.ldt,
341 dead_task->mm->context.size);
347 static inline void set_32bit_tls(struct task_struct *t, int tls, u32 addr)
349 struct user_desc ud = {
356 struct n_desc_struct *desc = (void *)t->thread.tls_array;
358 desc->a = LDT_entry_a(&ud);
359 desc->b = LDT_entry_b(&ud);
362 static inline u32 read_32bit_tls(struct task_struct *t, int tls)
364 struct desc_struct *desc = (void *)t->thread.tls_array;
367 (((u32)desc->base1) << 16) |
368 (((u32)desc->base2) << 24);
372 * This gets called before we allocate a new thread and copy
373 * the current task into it.
375 void prepare_to_copy(struct task_struct *tsk)
380 int copy_thread(int nr, unsigned long clone_flags, unsigned long rsp,
381 unsigned long unused,
382 struct task_struct * p, struct pt_regs * regs)
385 struct pt_regs * childregs;
386 struct task_struct *me = current;
388 childregs = ((struct pt_regs *) (THREAD_SIZE + (unsigned long) p->thread_info)) - 1;
393 childregs->rsp = rsp;
395 childregs->rsp = (unsigned long)childregs;
398 p->thread.rsp = (unsigned long) childregs;
399 p->thread.rsp0 = (unsigned long) (childregs+1);
400 p->thread.userrsp = me->thread.userrsp;
402 set_ti_thread_flag(p->thread_info, TIF_FORK);
404 p->thread.fs = me->thread.fs;
405 p->thread.gs = me->thread.gs;
407 asm("mov %%gs,%0" : "=m" (p->thread.gsindex));
408 asm("mov %%fs,%0" : "=m" (p->thread.fsindex));
409 asm("mov %%es,%0" : "=m" (p->thread.es));
410 asm("mov %%ds,%0" : "=m" (p->thread.ds));
412 if (unlikely(me->thread.io_bitmap_ptr != NULL)) {
413 p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL);
414 if (!p->thread.io_bitmap_ptr) {
415 p->thread.io_bitmap_max = 0;
418 memcpy(p->thread.io_bitmap_ptr, me->thread.io_bitmap_ptr, IO_BITMAP_BYTES);
422 * Set a new TLS for the child thread?
424 if (clone_flags & CLONE_SETTLS) {
425 #ifdef CONFIG_IA32_EMULATION
426 if (test_thread_flag(TIF_IA32))
427 err = ia32_child_tls(p, childregs);
430 err = do_arch_prctl(p, ARCH_SET_FS, childregs->r8);
436 if (err && p->thread.io_bitmap_ptr) {
437 kfree(p->thread.io_bitmap_ptr);
438 p->thread.io_bitmap_max = 0;
444 * This special macro can be used to load a debugging register
446 #define loaddebug(thread,r) set_debug(thread->debugreg ## r, r)
449 * switch_to(x,y) should switch tasks from x to y.
451 * This could still be optimized:
452 * - fold all the options into a flag word and test it with a single test.
453 * - could test fs/gs bitsliced
455 struct task_struct *__switch_to(struct task_struct *prev_p, struct task_struct *next_p)
457 struct thread_struct *prev = &prev_p->thread,
458 *next = &next_p->thread;
459 int cpu = smp_processor_id();
460 struct tss_struct *tss = &per_cpu(init_tss, cpu);
465 * Reload esp0, LDT and the page table pointer:
467 tss->rsp0 = next->rsp0;
471 * This won't pick up thread selector changes, but I guess that is ok.
473 asm volatile("mov %%es,%0" : "=m" (prev->es));
474 if (unlikely(next->es | prev->es))
475 loadsegment(es, next->es);
477 asm volatile ("mov %%ds,%0" : "=m" (prev->ds));
478 if (unlikely(next->ds | prev->ds))
479 loadsegment(ds, next->ds);
488 asm volatile("movl %%fs,%0" : "=r" (fsindex));
489 /* segment register != 0 always requires a reload.
490 also reload when it has changed.
491 when prev process used 64bit base always reload
492 to avoid an information leak. */
493 if (unlikely(fsindex | next->fsindex | prev->fs)) {
494 loadsegment(fs, next->fsindex);
495 /* check if the user used a selector != 0
496 * if yes clear 64bit base, since overloaded base
497 * is always mapped to the Null selector
502 /* when next process has a 64bit base use it */
504 wrmsrl(MSR_FS_BASE, next->fs);
505 prev->fsindex = fsindex;
509 asm volatile("movl %%gs,%0" : "=r" (gsindex));
510 if (unlikely(gsindex | next->gsindex | prev->gs)) {
511 load_gs_index(next->gsindex);
516 wrmsrl(MSR_KERNEL_GS_BASE, next->gs);
517 prev->gsindex = gsindex;
521 * Switch the PDA context.
523 prev->userrsp = read_pda(oldrsp);
524 write_pda(oldrsp, next->userrsp);
525 write_pda(pcurrent, next_p);
526 write_pda(kernelstack, (unsigned long)next_p->thread_info + THREAD_SIZE - PDA_STACKOFFSET);
529 * Now maybe reload the debug registers
531 if (unlikely(next->debugreg7)) {
543 * Handle the IO bitmap
545 if (unlikely(prev->io_bitmap_ptr || next->io_bitmap_ptr)) {
546 if (next->io_bitmap_ptr)
548 * Copy the relevant range of the IO bitmap.
549 * Normally this is 128 bytes or less:
551 memcpy(tss->io_bitmap, next->io_bitmap_ptr,
552 max(prev->io_bitmap_max, next->io_bitmap_max));
555 * Clear any possible leftover bits:
557 memset(tss->io_bitmap, 0xff, prev->io_bitmap_max);
565 * sys_execve() executes a new program.
568 long sys_execve(char __user *name, char __user * __user *argv,
569 char __user * __user *envp, struct pt_regs regs)
574 filename = getname(name);
575 error = PTR_ERR(filename);
576 if (IS_ERR(filename))
578 error = do_execve(filename, argv, envp, ®s);
581 current->ptrace &= ~PT_DTRACE;
582 task_unlock(current);
588 void set_personality_64bit(void)
590 /* inherit personality from parent */
592 /* Make sure to be in 64bit mode */
593 clear_thread_flag(TIF_IA32);
596 asmlinkage long sys_fork(struct pt_regs *regs)
598 return do_fork(SIGCHLD, regs->rsp, regs, 0, NULL, NULL);
601 asmlinkage long sys_clone(unsigned long clone_flags, unsigned long newsp, void __user *parent_tid, void __user *child_tid, struct pt_regs *regs)
605 return do_fork(clone_flags, newsp, regs, 0, parent_tid, child_tid);
609 * This is trivial, and on the face of it looks like it
610 * could equally well be done in user mode.
612 * Not so, for quite unobvious reasons - register pressure.
613 * In user mode vfork() cannot have a stack frame, and if
614 * done by calling the "clone()" system call directly, you
615 * do not have enough call-clobbered registers to hold all
616 * the information you need.
618 asmlinkage long sys_vfork(struct pt_regs *regs)
620 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->rsp, regs, 0,
624 unsigned long get_wchan(struct task_struct *p)
630 if (!p || p == current || p->state==TASK_RUNNING)
632 stack = (unsigned long)p->thread_info;
633 if (p->thread.rsp < stack || p->thread.rsp > stack+THREAD_SIZE)
635 fp = *(u64 *)(p->thread.rsp);
637 if (fp < (unsigned long)stack || fp > (unsigned long)stack+THREAD_SIZE)
639 rip = *(u64 *)(fp+8);
640 if (!in_sched_functions(rip))
643 } while (count++ < 16);
647 long do_arch_prctl(struct task_struct *task, int code, unsigned long addr)
650 int doit = task == current;
655 if (addr >= TASK_SIZE)
658 /* handle small bases via the GDT because that's faster to
660 if (addr <= 0xffffffff) {
661 set_32bit_tls(task, GS_TLS, addr);
663 load_TLS(&task->thread, cpu);
664 load_gs_index(GS_TLS_SEL);
666 task->thread.gsindex = GS_TLS_SEL;
669 task->thread.gsindex = 0;
670 task->thread.gs = addr;
673 ret = checking_wrmsrl(MSR_KERNEL_GS_BASE, addr);
679 /* Not strictly needed for fs, but do it for symmetry
681 if (addr >= TASK_SIZE)
684 /* handle small bases via the GDT because that's faster to
686 if (addr <= 0xffffffff) {
687 set_32bit_tls(task, FS_TLS, addr);
689 load_TLS(&task->thread, cpu);
690 asm volatile("movl %0,%%fs" :: "r" (FS_TLS_SEL));
692 task->thread.fsindex = FS_TLS_SEL;
695 task->thread.fsindex = 0;
696 task->thread.fs = addr;
698 /* set the selector to 0 to not confuse
700 asm volatile("movl %0,%%fs" :: "r" (0));
701 ret = checking_wrmsrl(MSR_FS_BASE, addr);
708 if (task->thread.fsindex == FS_TLS_SEL)
709 base = read_32bit_tls(task, FS_TLS);
711 rdmsrl(MSR_FS_BASE, base);
713 base = task->thread.fs;
714 ret = put_user(base, (unsigned long __user *)addr);
719 if (task->thread.gsindex == GS_TLS_SEL)
720 base = read_32bit_tls(task, GS_TLS);
722 rdmsrl(MSR_KERNEL_GS_BASE, base);
724 base = task->thread.gs;
725 ret = put_user(base, (unsigned long __user *)addr);
737 long sys_arch_prctl(int code, unsigned long addr)
739 return do_arch_prctl(current, code, addr);
743 * Capture the user space registers if the task is not running (in user space)
745 int dump_task_regs(struct task_struct *tsk, elf_gregset_t *regs)
747 struct pt_regs *pp, ptregs;
749 pp = (struct pt_regs *)(tsk->thread.rsp0);
756 elf_core_copy_regs(regs, &ptregs);