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);
58 * Powermanagement idle function, if any..
60 void (*pm_idle)(void);
62 void disable_hlt(void)
64 atomic_inc(&hlt_counter);
67 EXPORT_SYMBOL(disable_hlt);
71 atomic_dec(&hlt_counter);
74 EXPORT_SYMBOL(enable_hlt);
77 * We use this if we don't have any better
80 void default_idle(void)
82 if (!atomic_read(&hlt_counter)) {
92 * On SMP it's slightly faster (but much more power-consuming!)
93 * to poll the ->need_resched flag instead of waiting for the
94 * cross-CPU IPI to arrive. Use this option with caution.
96 static void poll_idle (void)
103 * Deal with another CPU just having chosen a thread to
106 oldval = test_and_clear_thread_flag(TIF_NEED_RESCHED);
109 set_thread_flag(TIF_POLLING_NRFLAG);
116 "i" (_TIF_NEED_RESCHED),
117 "m" (current_thread_info()->flags));
124 * The idle thread. There's no useful work to be
125 * done, so just try to conserve power and have a
126 * low exit latency (ie sit in a loop waiting for
127 * somebody to say that they'd like to reschedule)
131 /* endless idle loop with no priority at all */
133 while (!need_resched()) {
136 * Mark this as an RCU critical section so that
137 * synchronize_kernel() in the unload path waits
138 * for our completion.
152 * This uses new MONITOR/MWAIT instructions on P4 processors with PNI,
153 * which can obviate IPI to trigger checking of need_resched.
154 * We execute MONITOR against need_resched and enter optimized wait state
155 * through MWAIT. Whenever someone changes need_resched, we would be woken
156 * up from MWAIT (without an IPI).
158 static void mwait_idle(void)
162 if (!need_resched()) {
163 set_thread_flag(TIF_POLLING_NRFLAG);
165 __monitor((void *)¤t_thread_info()->flags, 0, 0);
169 } while (!need_resched());
170 clear_thread_flag(TIF_POLLING_NRFLAG);
174 void __init select_idle_routine(const struct cpuinfo_x86 *c)
177 if (cpu_has(c, X86_FEATURE_MWAIT)) {
179 * Skip, if setup has overridden idle.
180 * One CPU supports mwait => All CPUs supports mwait
184 printk("using mwait in idle threads.\n");
187 pm_idle = mwait_idle;
192 static int __init idle_setup (char *str)
194 if (!strncmp(str, "poll", 4)) {
195 printk("using polling idle threads.\n");
202 __setup("idle=", idle_setup);
204 /* Prints also some state that isn't saved in the pt_regs */
205 void __show_regs(struct pt_regs * regs)
207 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L, fs, gs, shadowgs;
208 unsigned int fsindex,gsindex;
209 unsigned int ds,cs,es;
213 printk("Pid: %d, comm: %.20s %s %s\n",
214 current->pid, current->comm, print_tainted(), UTS_RELEASE);
215 printk("RIP: %04lx:[<%016lx>] ", regs->cs & 0xffff, regs->rip);
216 printk_address(regs->rip);
217 printk("\nRSP: %04lx:%016lx EFLAGS: %08lx\n", regs->ss, regs->rsp, regs->eflags);
218 printk("RAX: %016lx RBX: %016lx RCX: %016lx\n",
219 regs->rax, regs->rbx, regs->rcx);
220 printk("RDX: %016lx RSI: %016lx RDI: %016lx\n",
221 regs->rdx, regs->rsi, regs->rdi);
222 printk("RBP: %016lx R08: %016lx R09: %016lx\n",
223 regs->rbp, regs->r8, regs->r9);
224 printk("R10: %016lx R11: %016lx R12: %016lx\n",
225 regs->r10, regs->r11, regs->r12);
226 printk("R13: %016lx R14: %016lx R15: %016lx\n",
227 regs->r13, regs->r14, regs->r15);
229 asm("movl %%ds,%0" : "=r" (ds));
230 asm("movl %%cs,%0" : "=r" (cs));
231 asm("movl %%es,%0" : "=r" (es));
232 asm("movl %%fs,%0" : "=r" (fsindex));
233 asm("movl %%gs,%0" : "=r" (gsindex));
235 rdmsrl(MSR_FS_BASE, fs);
236 rdmsrl(MSR_GS_BASE, gs);
237 rdmsrl(MSR_KERNEL_GS_BASE, shadowgs);
239 asm("movq %%cr0, %0": "=r" (cr0));
240 asm("movq %%cr2, %0": "=r" (cr2));
241 asm("movq %%cr3, %0": "=r" (cr3));
242 asm("movq %%cr4, %0": "=r" (cr4));
244 printk("FS: %016lx(%04x) GS:%016lx(%04x) knlGS:%016lx\n",
245 fs,fsindex,gs,gsindex,shadowgs);
246 printk("CS: %04x DS: %04x ES: %04x CR0: %016lx\n", cs, ds, es, cr0);
247 printk("CR2: %016lx CR3: %016lx CR4: %016lx\n", cr2, cr3, cr4);
250 void show_regs(struct pt_regs *regs)
253 show_trace(®s->rsp);
257 * Free current thread data structures etc..
259 void exit_thread(void)
261 struct task_struct *me = current;
262 struct thread_struct *t = &me->thread;
263 if (me->thread.io_bitmap_ptr) {
264 struct tss_struct *tss = &per_cpu(init_tss, get_cpu());
266 kfree(t->io_bitmap_ptr);
267 t->io_bitmap_ptr = NULL;
269 * Careful, clear this in the TSS too:
271 memset(tss->io_bitmap, 0xff, t->io_bitmap_max);
272 t->io_bitmap_max = 0;
277 void flush_thread(void)
279 struct task_struct *tsk = current;
280 struct thread_info *t = current_thread_info();
282 if (t->flags & _TIF_ABI_PENDING)
283 t->flags ^= (_TIF_ABI_PENDING | _TIF_IA32);
285 tsk->thread.debugreg0 = 0;
286 tsk->thread.debugreg1 = 0;
287 tsk->thread.debugreg2 = 0;
288 tsk->thread.debugreg3 = 0;
289 tsk->thread.debugreg6 = 0;
290 tsk->thread.debugreg7 = 0;
291 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
293 * Forget coprocessor state..
299 void release_thread(struct task_struct *dead_task)
302 if (dead_task->mm->context.size) {
303 printk("WARNING: dead process %8s still has LDT? <%p/%d>\n",
305 dead_task->mm->context.ldt,
306 dead_task->mm->context.size);
312 static inline void set_32bit_tls(struct task_struct *t, int tls, u32 addr)
314 struct user_desc ud = {
321 struct n_desc_struct *desc = (void *)t->thread.tls_array;
323 desc->a = LDT_entry_a(&ud);
324 desc->b = LDT_entry_b(&ud);
327 static inline u32 read_32bit_tls(struct task_struct *t, int tls)
329 struct desc_struct *desc = (void *)t->thread.tls_array;
332 (((u32)desc->base1) << 16) |
333 (((u32)desc->base2) << 24);
337 * This gets called before we allocate a new thread and copy
338 * the current task into it.
340 void prepare_to_copy(struct task_struct *tsk)
345 int copy_thread(int nr, unsigned long clone_flags, unsigned long rsp,
346 unsigned long unused,
347 struct task_struct * p, struct pt_regs * regs)
350 struct pt_regs * childregs;
351 struct task_struct *me = current;
353 childregs = ((struct pt_regs *) (THREAD_SIZE + (unsigned long) p->thread_info)) - 1;
358 childregs->rsp = rsp;
360 childregs->rsp = (unsigned long)childregs;
362 p->set_child_tid = p->clear_child_tid = NULL;
364 p->thread.rsp = (unsigned long) childregs;
365 p->thread.rsp0 = (unsigned long) (childregs+1);
366 p->thread.userrsp = me->thread.userrsp;
368 set_ti_thread_flag(p->thread_info, TIF_FORK);
370 p->thread.fs = me->thread.fs;
371 p->thread.gs = me->thread.gs;
373 asm("movl %%gs,%0" : "=m" (p->thread.gsindex));
374 asm("movl %%fs,%0" : "=m" (p->thread.fsindex));
375 asm("movl %%es,%0" : "=m" (p->thread.es));
376 asm("movl %%ds,%0" : "=m" (p->thread.ds));
378 if (unlikely(me->thread.io_bitmap_ptr != NULL)) {
379 p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL);
380 if (!p->thread.io_bitmap_ptr) {
381 p->thread.io_bitmap_max = 0;
384 memcpy(p->thread.io_bitmap_ptr, me->thread.io_bitmap_ptr, IO_BITMAP_BYTES);
388 * Set a new TLS for the child thread?
390 if (clone_flags & CLONE_SETTLS) {
391 #ifdef CONFIG_IA32_EMULATION
392 if (test_thread_flag(TIF_IA32))
393 err = ia32_child_tls(p, childregs);
396 err = do_arch_prctl(p, ARCH_SET_FS, childregs->r8);
402 if (err && p->thread.io_bitmap_ptr) {
403 kfree(p->thread.io_bitmap_ptr);
404 p->thread.io_bitmap_max = 0;
410 * This special macro can be used to load a debugging register
412 #define loaddebug(thread,r) set_debug(thread->debugreg ## r, r)
415 * switch_to(x,y) should switch tasks from x to y.
417 * This could still be optimized:
418 * - fold all the options into a flag word and test it with a single test.
419 * - could test fs/gs bitsliced
421 struct task_struct *__switch_to(struct task_struct *prev_p, struct task_struct *next_p)
423 struct thread_struct *prev = &prev_p->thread,
424 *next = &next_p->thread;
425 int cpu = smp_processor_id();
426 struct tss_struct *tss = &per_cpu(init_tss, cpu);
431 * Reload esp0, LDT and the page table pointer:
433 tss->rsp0 = next->rsp0;
437 * This won't pick up thread selector changes, but I guess that is ok.
439 asm volatile("movl %%es,%0" : "=m" (prev->es));
440 if (unlikely(next->es | prev->es))
441 loadsegment(es, next->es);
443 asm volatile ("movl %%ds,%0" : "=m" (prev->ds));
444 if (unlikely(next->ds | prev->ds))
445 loadsegment(ds, next->ds);
454 asm volatile("movl %%fs,%0" : "=g" (fsindex));
455 /* segment register != 0 always requires a reload.
456 also reload when it has changed.
457 when prev process used 64bit base always reload
458 to avoid an information leak. */
459 if (unlikely(fsindex | next->fsindex | prev->fs)) {
460 loadsegment(fs, next->fsindex);
461 /* check if the user used a selector != 0
462 * if yes clear 64bit base, since overloaded base
463 * is always mapped to the Null selector
468 /* when next process has a 64bit base use it */
470 wrmsrl(MSR_FS_BASE, next->fs);
471 prev->fsindex = fsindex;
475 asm volatile("movl %%gs,%0" : "=g" (gsindex));
476 if (unlikely(gsindex | next->gsindex | prev->gs)) {
477 load_gs_index(next->gsindex);
482 wrmsrl(MSR_KERNEL_GS_BASE, next->gs);
483 prev->gsindex = gsindex;
487 * Switch the PDA context.
489 prev->userrsp = read_pda(oldrsp);
490 write_pda(oldrsp, next->userrsp);
491 write_pda(pcurrent, next_p);
492 write_pda(kernelstack, (unsigned long)next_p->thread_info + THREAD_SIZE - PDA_STACKOFFSET);
495 * Now maybe reload the debug registers
497 if (unlikely(next->debugreg7)) {
509 * Handle the IO bitmap
511 if (unlikely(prev->io_bitmap_ptr || next->io_bitmap_ptr)) {
512 if (next->io_bitmap_ptr)
514 * Copy the relevant range of the IO bitmap.
515 * Normally this is 128 bytes or less:
517 memcpy(tss->io_bitmap, next->io_bitmap_ptr,
518 max(prev->io_bitmap_max, next->io_bitmap_max));
521 * Clear any possible leftover bits:
523 memset(tss->io_bitmap, 0xff, prev->io_bitmap_max);
531 * sys_execve() executes a new program.
534 long sys_execve(char __user *name, char __user * __user *argv,
535 char __user * __user *envp, struct pt_regs regs)
540 filename = getname(name);
541 error = PTR_ERR(filename);
542 if (IS_ERR(filename))
544 error = do_execve(filename, argv, envp, ®s);
546 current->ptrace &= ~PT_DTRACE;
551 void set_personality_64bit(void)
553 /* inherit personality from parent */
555 /* Make sure to be in 64bit mode */
556 clear_thread_flag(TIF_IA32);
559 asmlinkage long sys_fork(struct pt_regs *regs)
561 return do_fork(SIGCHLD, regs->rsp, regs, 0, NULL, NULL);
564 asmlinkage long sys_clone(unsigned long clone_flags, unsigned long newsp, void __user *parent_tid, void __user *child_tid, struct pt_regs *regs)
568 return do_fork(clone_flags, newsp, regs, 0, parent_tid, child_tid);
572 * This is trivial, and on the face of it looks like it
573 * could equally well be done in user mode.
575 * Not so, for quite unobvious reasons - register pressure.
576 * In user mode vfork() cannot have a stack frame, and if
577 * done by calling the "clone()" system call directly, you
578 * do not have enough call-clobbered registers to hold all
579 * the information you need.
581 asmlinkage long sys_vfork(struct pt_regs *regs)
583 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->rsp, regs, 0,
587 unsigned long get_wchan(struct task_struct *p)
593 if (!p || p == current || p->state==TASK_RUNNING)
595 stack = (unsigned long)p->thread_info;
596 if (p->thread.rsp < stack || p->thread.rsp > stack+THREAD_SIZE)
598 fp = *(u64 *)(p->thread.rsp);
600 if (fp < (unsigned long)stack || fp > (unsigned long)stack+THREAD_SIZE)
602 rip = *(u64 *)(fp+8);
603 if (!in_sched_functions(rip))
606 } while (count++ < 16);
610 long do_arch_prctl(struct task_struct *task, int code, unsigned long addr)
613 int doit = task == current;
618 if (addr >= TASK_SIZE)
621 /* handle small bases via the GDT because that's faster to
623 if (addr <= 0xffffffff) {
624 set_32bit_tls(task, GS_TLS, addr);
626 load_TLS(&task->thread, cpu);
627 load_gs_index(GS_TLS_SEL);
629 task->thread.gsindex = GS_TLS_SEL;
632 task->thread.gsindex = 0;
633 task->thread.gs = addr;
636 ret = checking_wrmsrl(MSR_KERNEL_GS_BASE, addr);
642 /* Not strictly needed for fs, but do it for symmetry
644 if (addr >= TASK_SIZE)
647 /* handle small bases via the GDT because that's faster to
649 if (addr <= 0xffffffff) {
650 set_32bit_tls(task, FS_TLS, addr);
652 load_TLS(&task->thread, cpu);
653 asm volatile("movl %0,%%fs" :: "r" (FS_TLS_SEL));
655 task->thread.fsindex = FS_TLS_SEL;
658 task->thread.fsindex = 0;
659 task->thread.fs = addr;
661 /* set the selector to 0 to not confuse
663 asm volatile("movl %0,%%fs" :: "r" (0));
664 ret = checking_wrmsrl(MSR_FS_BASE, addr);
671 if (task->thread.fsindex == FS_TLS_SEL)
672 base = read_32bit_tls(task, FS_TLS);
674 rdmsrl(MSR_FS_BASE, base);
676 base = task->thread.fs;
677 ret = put_user(base, (unsigned long __user *)addr);
682 if (task->thread.gsindex == GS_TLS_SEL)
683 base = read_32bit_tls(task, GS_TLS);
685 rdmsrl(MSR_KERNEL_GS_BASE, base);
687 base = task->thread.gs;
688 ret = put_user(base, (unsigned long __user *)addr);
700 long sys_arch_prctl(int code, unsigned long addr)
702 return do_arch_prctl(current, code, addr);
706 * Capture the user space registers if the task is not running (in user space)
708 int dump_task_regs(struct task_struct *tsk, elf_gregset_t *regs)
710 struct pt_regs *pp, ptregs;
712 pp = (struct pt_regs *)(tsk->thread.rsp0);
719 elf_core_copy_regs(regs, &ptregs);