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/version.h>
37 #include <asm/uaccess.h>
38 #include <asm/pgtable.h>
39 #include <asm/system.h>
41 #include <asm/processor.h>
43 #include <asm/mmu_context.h>
45 #include <asm/prctl.h>
46 #include <asm/kdebug.h>
48 #include <asm/proto.h>
51 asmlinkage extern void ret_from_fork(void);
53 unsigned long kernel_thread_flags = CLONE_VM | CLONE_UNTRACED;
55 atomic_t hlt_counter = ATOMIC_INIT(0);
57 unsigned long boot_option_idle_override = 0;
58 EXPORT_SYMBOL(boot_option_idle_override);
61 * Powermanagement idle function, if any..
63 void (*pm_idle)(void);
65 void disable_hlt(void)
67 atomic_inc(&hlt_counter);
70 EXPORT_SYMBOL(disable_hlt);
74 atomic_dec(&hlt_counter);
77 EXPORT_SYMBOL(enable_hlt);
80 * We use this if we don't have any better
83 void default_idle(void)
85 if (!atomic_read(&hlt_counter)) {
95 * On SMP it's slightly faster (but much more power-consuming!)
96 * to poll the ->need_resched flag instead of waiting for the
97 * cross-CPU IPI to arrive. Use this option with caution.
99 static void poll_idle (void)
106 * Deal with another CPU just having chosen a thread to
109 oldval = test_and_clear_thread_flag(TIF_NEED_RESCHED);
112 set_thread_flag(TIF_POLLING_NRFLAG);
119 "i" (_TIF_NEED_RESCHED),
120 "m" (current_thread_info()->flags));
127 * The idle thread. There's no useful work to be
128 * done, so just try to conserve power and have a
129 * low exit latency (ie sit in a loop waiting for
130 * somebody to say that they'd like to reschedule)
134 /* endless idle loop with no priority at all */
136 while (!need_resched()) {
139 * Mark this as an RCU critical section so that
140 * synchronize_kernel() in the unload path waits
141 * for our completion.
155 * This uses new MONITOR/MWAIT instructions on P4 processors with PNI,
156 * which can obviate IPI to trigger checking of need_resched.
157 * We execute MONITOR against need_resched and enter optimized wait state
158 * through MWAIT. Whenever someone changes need_resched, we would be woken
159 * up from MWAIT (without an IPI).
161 static void mwait_idle(void)
165 if (!need_resched()) {
166 set_thread_flag(TIF_POLLING_NRFLAG);
168 __monitor((void *)¤t_thread_info()->flags, 0, 0);
172 } while (!need_resched());
173 clear_thread_flag(TIF_POLLING_NRFLAG);
177 void __init select_idle_routine(const struct cpuinfo_x86 *c)
180 if (cpu_has(c, X86_FEATURE_MWAIT)) {
182 * Skip, if setup has overridden idle.
183 * One CPU supports mwait => All CPUs supports mwait
187 printk("using mwait in idle threads.\n");
190 pm_idle = mwait_idle;
195 static int __init idle_setup (char *str)
197 if (!strncmp(str, "poll", 4)) {
198 printk("using polling idle threads.\n");
202 boot_option_idle_override = 1;
206 __setup("idle=", idle_setup);
208 /* Prints also some state that isn't saved in the pt_regs */
209 void __show_regs(struct pt_regs * regs)
211 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L, fs, gs, shadowgs;
212 unsigned int fsindex,gsindex;
213 unsigned int ds,cs,es;
217 printk("Pid: %d, comm: %.20s %s %s\n",
218 current->pid, current->comm, print_tainted(), UTS_RELEASE);
219 printk("RIP: %04lx:[<%016lx>] ", regs->cs & 0xffff, regs->rip);
220 printk_address(regs->rip);
221 printk("\nRSP: %04lx:%016lx EFLAGS: %08lx\n", regs->ss, regs->rsp, regs->eflags);
222 printk("RAX: %016lx RBX: %016lx RCX: %016lx\n",
223 regs->rax, regs->rbx, regs->rcx);
224 printk("RDX: %016lx RSI: %016lx RDI: %016lx\n",
225 regs->rdx, regs->rsi, regs->rdi);
226 printk("RBP: %016lx R08: %016lx R09: %016lx\n",
227 regs->rbp, regs->r8, regs->r9);
228 printk("R10: %016lx R11: %016lx R12: %016lx\n",
229 regs->r10, regs->r11, regs->r12);
230 printk("R13: %016lx R14: %016lx R15: %016lx\n",
231 regs->r13, regs->r14, regs->r15);
233 asm("movl %%ds,%0" : "=r" (ds));
234 asm("movl %%cs,%0" : "=r" (cs));
235 asm("movl %%es,%0" : "=r" (es));
236 asm("movl %%fs,%0" : "=r" (fsindex));
237 asm("movl %%gs,%0" : "=r" (gsindex));
239 rdmsrl(MSR_FS_BASE, fs);
240 rdmsrl(MSR_GS_BASE, gs);
241 rdmsrl(MSR_KERNEL_GS_BASE, shadowgs);
243 asm("movq %%cr0, %0": "=r" (cr0));
244 asm("movq %%cr2, %0": "=r" (cr2));
245 asm("movq %%cr3, %0": "=r" (cr3));
246 asm("movq %%cr4, %0": "=r" (cr4));
248 printk("FS: %016lx(%04x) GS:%016lx(%04x) knlGS:%016lx\n",
249 fs,fsindex,gs,gsindex,shadowgs);
250 printk("CS: %04x DS: %04x ES: %04x CR0: %016lx\n", cs, ds, es, cr0);
251 printk("CR2: %016lx CR3: %016lx CR4: %016lx\n", cr2, cr3, cr4);
254 void show_regs(struct pt_regs *regs)
257 show_trace(®s->rsp);
260 EXPORT_SYMBOL_GPL(show_regs);
263 * Free current thread data structures etc..
265 void exit_thread(void)
267 struct task_struct *me = current;
268 struct thread_struct *t = &me->thread;
269 if (me->thread.io_bitmap_ptr) {
270 struct tss_struct *tss = &per_cpu(init_tss, get_cpu());
272 kfree(t->io_bitmap_ptr);
273 t->io_bitmap_ptr = NULL;
275 * Careful, clear this in the TSS too:
277 memset(tss->io_bitmap, 0xff, t->io_bitmap_max);
278 t->io_bitmap_max = 0;
283 void flush_thread(void)
285 struct task_struct *tsk = current;
286 struct thread_info *t = current_thread_info();
288 if (t->flags & _TIF_ABI_PENDING)
289 t->flags ^= (_TIF_ABI_PENDING | _TIF_IA32);
291 tsk->thread.debugreg0 = 0;
292 tsk->thread.debugreg1 = 0;
293 tsk->thread.debugreg2 = 0;
294 tsk->thread.debugreg3 = 0;
295 tsk->thread.debugreg6 = 0;
296 tsk->thread.debugreg7 = 0;
297 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
299 * Forget coprocessor state..
305 void release_thread(struct task_struct *dead_task)
308 if (dead_task->mm->context.size) {
309 printk("WARNING: dead process %8s still has LDT? <%p/%d>\n",
311 dead_task->mm->context.ldt,
312 dead_task->mm->context.size);
318 static inline void set_32bit_tls(struct task_struct *t, int tls, u32 addr)
320 struct user_desc ud = {
327 struct n_desc_struct *desc = (void *)t->thread.tls_array;
329 desc->a = LDT_entry_a(&ud);
330 desc->b = LDT_entry_b(&ud);
333 static inline u32 read_32bit_tls(struct task_struct *t, int tls)
335 struct desc_struct *desc = (void *)t->thread.tls_array;
338 (((u32)desc->base1) << 16) |
339 (((u32)desc->base2) << 24);
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 rsp,
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 *) (THREAD_SIZE + (unsigned long) p->thread_info)) - 1;
364 childregs->rsp = rsp;
366 childregs->rsp = (unsigned long)childregs;
369 p->thread.rsp = (unsigned long) childregs;
370 p->thread.rsp0 = (unsigned long) (childregs+1);
371 p->thread.userrsp = me->thread.userrsp;
373 set_ti_thread_flag(p->thread_info, TIF_FORK);
375 p->thread.fs = me->thread.fs;
376 p->thread.gs = me->thread.gs;
378 asm("movl %%gs,%0" : "=m" (p->thread.gsindex));
379 asm("movl %%fs,%0" : "=m" (p->thread.fsindex));
380 asm("movl %%es,%0" : "=m" (p->thread.es));
381 asm("movl %%ds,%0" : "=m" (p->thread.ds));
383 if (unlikely(me->thread.io_bitmap_ptr != NULL)) {
384 p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL);
385 if (!p->thread.io_bitmap_ptr) {
386 p->thread.io_bitmap_max = 0;
389 memcpy(p->thread.io_bitmap_ptr, me->thread.io_bitmap_ptr, IO_BITMAP_BYTES);
393 * Set a new TLS for the child thread?
395 if (clone_flags & CLONE_SETTLS) {
396 #ifdef CONFIG_IA32_EMULATION
397 if (test_thread_flag(TIF_IA32))
398 err = ia32_child_tls(p, childregs);
401 err = do_arch_prctl(p, ARCH_SET_FS, childregs->r8);
407 if (err && p->thread.io_bitmap_ptr) {
408 kfree(p->thread.io_bitmap_ptr);
409 p->thread.io_bitmap_max = 0;
415 * This special macro can be used to load a debugging register
417 #define loaddebug(thread,r) set_debug(thread->debugreg ## r, r)
420 * switch_to(x,y) should switch tasks from x to y.
422 * This could still be optimized:
423 * - fold all the options into a flag word and test it with a single test.
424 * - could test fs/gs bitsliced
426 struct task_struct *__switch_to(struct task_struct *prev_p, struct task_struct *next_p)
428 struct thread_struct *prev = &prev_p->thread,
429 *next = &next_p->thread;
430 int cpu = smp_processor_id();
431 struct tss_struct *tss = &per_cpu(init_tss, cpu);
436 * Reload esp0, LDT and the page table pointer:
438 tss->rsp0 = next->rsp0;
442 * This won't pick up thread selector changes, but I guess that is ok.
444 asm volatile("movl %%es,%0" : "=m" (prev->es));
445 if (unlikely(next->es | prev->es))
446 loadsegment(es, next->es);
448 asm volatile ("movl %%ds,%0" : "=m" (prev->ds));
449 if (unlikely(next->ds | prev->ds))
450 loadsegment(ds, next->ds);
459 asm volatile("movl %%fs,%0" : "=g" (fsindex));
460 /* segment register != 0 always requires a reload.
461 also reload when it has changed.
462 when prev process used 64bit base always reload
463 to avoid an information leak. */
464 if (unlikely(fsindex | next->fsindex | prev->fs)) {
465 loadsegment(fs, next->fsindex);
466 /* check if the user used a selector != 0
467 * if yes clear 64bit base, since overloaded base
468 * is always mapped to the Null selector
473 /* when next process has a 64bit base use it */
475 wrmsrl(MSR_FS_BASE, next->fs);
476 prev->fsindex = fsindex;
480 asm volatile("movl %%gs,%0" : "=g" (gsindex));
481 if (unlikely(gsindex | next->gsindex | prev->gs)) {
482 load_gs_index(next->gsindex);
487 wrmsrl(MSR_KERNEL_GS_BASE, next->gs);
488 prev->gsindex = gsindex;
492 * Switch the PDA context.
494 prev->userrsp = read_pda(oldrsp);
495 write_pda(oldrsp, next->userrsp);
496 write_pda(pcurrent, next_p);
497 write_pda(kernelstack, (unsigned long)next_p->thread_info + THREAD_SIZE - PDA_STACKOFFSET);
500 * Now maybe reload the debug registers
502 if (unlikely(next->debugreg7)) {
514 * Handle the IO bitmap
516 if (unlikely(prev->io_bitmap_ptr || next->io_bitmap_ptr)) {
517 if (next->io_bitmap_ptr)
519 * Copy the relevant range of the IO bitmap.
520 * Normally this is 128 bytes or less:
522 memcpy(tss->io_bitmap, next->io_bitmap_ptr,
523 max(prev->io_bitmap_max, next->io_bitmap_max));
526 * Clear any possible leftover bits:
528 memset(tss->io_bitmap, 0xff, prev->io_bitmap_max);
536 * sys_execve() executes a new program.
539 long sys_execve(char __user *name, char __user * __user *argv,
540 char __user * __user *envp, struct pt_regs regs)
545 filename = getname(name);
546 error = PTR_ERR(filename);
547 if (IS_ERR(filename))
549 error = do_execve(filename, argv, envp, ®s);
552 current->ptrace &= ~PT_DTRACE;
553 task_unlock(current);
559 void set_personality_64bit(void)
561 /* inherit personality from parent */
563 /* Make sure to be in 64bit mode */
564 clear_thread_flag(TIF_IA32);
567 asmlinkage long sys_fork(struct pt_regs *regs)
569 return do_fork(SIGCHLD, regs->rsp, regs, 0, NULL, NULL);
572 asmlinkage long sys_clone(unsigned long clone_flags, unsigned long newsp, void __user *parent_tid, void __user *child_tid, struct pt_regs *regs)
576 return do_fork(clone_flags, newsp, regs, 0, parent_tid, child_tid);
580 * This is trivial, and on the face of it looks like it
581 * could equally well be done in user mode.
583 * Not so, for quite unobvious reasons - register pressure.
584 * In user mode vfork() cannot have a stack frame, and if
585 * done by calling the "clone()" system call directly, you
586 * do not have enough call-clobbered registers to hold all
587 * the information you need.
589 asmlinkage long sys_vfork(struct pt_regs *regs)
591 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->rsp, regs, 0,
595 unsigned long get_wchan(struct task_struct *p)
601 if (!p || p == current || p->state==TASK_RUNNING)
603 stack = (unsigned long)p->thread_info;
604 if (p->thread.rsp < stack || p->thread.rsp > stack+THREAD_SIZE)
606 fp = *(u64 *)(p->thread.rsp);
608 if (fp < (unsigned long)stack || fp > (unsigned long)stack+THREAD_SIZE)
610 rip = *(u64 *)(fp+8);
611 if (!in_sched_functions(rip))
614 } while (count++ < 16);
618 long do_arch_prctl(struct task_struct *task, int code, unsigned long addr)
621 int doit = task == current;
626 if (addr >= TASK_SIZE)
629 /* handle small bases via the GDT because that's faster to
631 if (addr <= 0xffffffff) {
632 set_32bit_tls(task, GS_TLS, addr);
634 load_TLS(&task->thread, cpu);
635 load_gs_index(GS_TLS_SEL);
637 task->thread.gsindex = GS_TLS_SEL;
640 task->thread.gsindex = 0;
641 task->thread.gs = addr;
644 ret = checking_wrmsrl(MSR_KERNEL_GS_BASE, addr);
650 /* Not strictly needed for fs, but do it for symmetry
652 if (addr >= TASK_SIZE)
655 /* handle small bases via the GDT because that's faster to
657 if (addr <= 0xffffffff) {
658 set_32bit_tls(task, FS_TLS, addr);
660 load_TLS(&task->thread, cpu);
661 asm volatile("movl %0,%%fs" :: "r" (FS_TLS_SEL));
663 task->thread.fsindex = FS_TLS_SEL;
666 task->thread.fsindex = 0;
667 task->thread.fs = addr;
669 /* set the selector to 0 to not confuse
671 asm volatile("movl %0,%%fs" :: "r" (0));
672 ret = checking_wrmsrl(MSR_FS_BASE, addr);
679 if (task->thread.fsindex == FS_TLS_SEL)
680 base = read_32bit_tls(task, FS_TLS);
682 rdmsrl(MSR_FS_BASE, base);
684 base = task->thread.fs;
685 ret = put_user(base, (unsigned long __user *)addr);
690 if (task->thread.gsindex == GS_TLS_SEL)
691 base = read_32bit_tls(task, GS_TLS);
693 rdmsrl(MSR_KERNEL_GS_BASE, base);
695 base = task->thread.gs;
696 ret = put_user(base, (unsigned long __user *)addr);
708 long sys_arch_prctl(int code, unsigned long addr)
710 return do_arch_prctl(current, code, addr);
714 * Capture the user space registers if the task is not running (in user space)
716 int dump_task_regs(struct task_struct *tsk, elf_gregset_t *regs)
718 struct pt_regs *pp, ptregs;
720 pp = (struct pt_regs *)(tsk->thread.rsp0);
727 elf_core_copy_regs(regs, &ptregs);