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 void (*idle)(void) = pm_idle;
136 while (!need_resched())
143 * This uses new MONITOR/MWAIT instructions on P4 processors with PNI,
144 * which can obviate IPI to trigger checking of need_resched.
145 * We execute MONITOR against need_resched and enter optimized wait state
146 * through MWAIT. Whenever someone changes need_resched, we would be woken
147 * up from MWAIT (without an IPI).
149 static void mwait_idle(void)
153 if (!need_resched()) {
154 set_thread_flag(TIF_POLLING_NRFLAG);
156 __monitor((void *)¤t_thread_info()->flags, 0, 0);
160 } while (!need_resched());
161 clear_thread_flag(TIF_POLLING_NRFLAG);
165 void __init select_idle_routine(const struct cpuinfo_x86 *c)
168 if (cpu_has(c, X86_FEATURE_MWAIT)) {
170 * Skip, if setup has overridden idle.
171 * Also, take care of system with asymmetric CPUs.
172 * Use, mwait_idle only if all cpus support it.
173 * If not, we fallback to default_idle()
177 printk("using mwait in idle threads.\n");
180 pm_idle = mwait_idle;
184 pm_idle = default_idle;
188 static int __init idle_setup (char *str)
190 if (!strncmp(str, "poll", 4)) {
191 printk("using polling idle threads.\n");
198 __setup("idle=", idle_setup);
200 /* Prints also some state that isn't saved in the pt_regs */
201 void __show_regs(struct pt_regs * regs)
203 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L, fs, gs, shadowgs;
204 unsigned int fsindex,gsindex;
205 unsigned int ds,cs,es;
209 printk("Pid: %d, comm: %.20s %s %s\n",
210 current->pid, current->comm, print_tainted(), UTS_RELEASE);
211 printk("RIP: %04lx:[<%016lx>] ", regs->cs & 0xffff, regs->rip);
212 printk_address(regs->rip);
213 printk("\nRSP: %04lx:%016lx EFLAGS: %08lx\n", regs->ss, regs->rsp, regs->eflags);
214 printk("RAX: %016lx RBX: %016lx RCX: %016lx\n",
215 regs->rax, regs->rbx, regs->rcx);
216 printk("RDX: %016lx RSI: %016lx RDI: %016lx\n",
217 regs->rdx, regs->rsi, regs->rdi);
218 printk("RBP: %016lx R08: %016lx R09: %016lx\n",
219 regs->rbp, regs->r8, regs->r9);
220 printk("R10: %016lx R11: %016lx R12: %016lx\n",
221 regs->r10, regs->r11, regs->r12);
222 printk("R13: %016lx R14: %016lx R15: %016lx\n",
223 regs->r13, regs->r14, regs->r15);
225 asm("movl %%ds,%0" : "=r" (ds));
226 asm("movl %%cs,%0" : "=r" (cs));
227 asm("movl %%es,%0" : "=r" (es));
228 asm("movl %%fs,%0" : "=r" (fsindex));
229 asm("movl %%gs,%0" : "=r" (gsindex));
231 rdmsrl(MSR_FS_BASE, fs);
232 rdmsrl(MSR_GS_BASE, gs);
233 rdmsrl(MSR_KERNEL_GS_BASE, shadowgs);
235 asm("movq %%cr0, %0": "=r" (cr0));
236 asm("movq %%cr2, %0": "=r" (cr2));
237 asm("movq %%cr3, %0": "=r" (cr3));
238 asm("movq %%cr4, %0": "=r" (cr4));
240 printk("FS: %016lx(%04x) GS:%016lx(%04x) knlGS:%016lx\n",
241 fs,fsindex,gs,gsindex,shadowgs);
242 printk("CS: %04x DS: %04x ES: %04x CR0: %016lx\n", cs, ds, es, cr0);
243 printk("CR2: %016lx CR3: %016lx CR4: %016lx\n", cr2, cr3, cr4);
246 void show_regs(struct pt_regs *regs)
249 show_trace(®s->rsp);
253 * Free current thread data structures etc..
255 void exit_thread(void)
257 struct task_struct *me = current;
258 if (me->thread.io_bitmap_ptr) {
259 struct tss_struct *tss = init_tss + get_cpu();
260 kfree(me->thread.io_bitmap_ptr);
261 me->thread.io_bitmap_ptr = NULL;
262 tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
267 void flush_thread(void)
269 struct task_struct *tsk = current;
270 struct thread_info *t = current_thread_info();
272 if (t->flags & _TIF_ABI_PENDING)
273 t->flags ^= (_TIF_ABI_PENDING | _TIF_IA32);
275 tsk->thread.debugreg0 = 0;
276 tsk->thread.debugreg1 = 0;
277 tsk->thread.debugreg2 = 0;
278 tsk->thread.debugreg3 = 0;
279 tsk->thread.debugreg6 = 0;
280 tsk->thread.debugreg7 = 0;
281 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
283 * Forget coprocessor state..
289 void release_thread(struct task_struct *dead_task)
292 if (dead_task->mm->context.size) {
293 printk("WARNING: dead process %8s still has LDT? <%p/%d>\n",
295 dead_task->mm->context.ldt,
296 dead_task->mm->context.size);
302 static inline void set_32bit_tls(struct task_struct *t, int tls, u32 addr)
304 struct user_desc ud = {
311 struct n_desc_struct *desc = (void *)t->thread.tls_array;
313 desc->a = LDT_entry_a(&ud);
314 desc->b = LDT_entry_b(&ud);
317 static inline u32 read_32bit_tls(struct task_struct *t, int tls)
319 struct desc_struct *desc = (void *)t->thread.tls_array;
322 (((u32)desc->base1) << 16) |
323 (((u32)desc->base2) << 24);
327 * This gets called before we allocate a new thread and copy
328 * the current task into it.
330 void prepare_to_copy(struct task_struct *tsk)
335 int copy_thread(int nr, unsigned long clone_flags, unsigned long rsp,
336 unsigned long unused,
337 struct task_struct * p, struct pt_regs * regs)
340 struct pt_regs * childregs;
341 struct task_struct *me = current;
343 childregs = ((struct pt_regs *) (THREAD_SIZE + (unsigned long) p->thread_info)) - 1;
348 childregs->rsp = rsp;
350 childregs->rsp = (unsigned long)childregs;
352 p->set_child_tid = p->clear_child_tid = NULL;
354 p->thread.rsp = (unsigned long) childregs;
355 p->thread.rsp0 = (unsigned long) (childregs+1);
356 p->thread.userrsp = me->thread.userrsp;
358 set_ti_thread_flag(p->thread_info, TIF_FORK);
360 p->thread.fs = me->thread.fs;
361 p->thread.gs = me->thread.gs;
363 asm("movl %%gs,%0" : "=m" (p->thread.gsindex));
364 asm("movl %%fs,%0" : "=m" (p->thread.fsindex));
365 asm("movl %%es,%0" : "=m" (p->thread.es));
366 asm("movl %%ds,%0" : "=m" (p->thread.ds));
368 if (unlikely(me->thread.io_bitmap_ptr != NULL)) {
369 p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL);
370 if (!p->thread.io_bitmap_ptr)
372 memcpy(p->thread.io_bitmap_ptr, me->thread.io_bitmap_ptr, IO_BITMAP_BYTES);
376 * Set a new TLS for the child thread?
378 if (clone_flags & CLONE_SETTLS) {
379 #ifdef CONFIG_IA32_EMULATION
380 if (test_thread_flag(TIF_IA32))
381 err = ia32_child_tls(p, childregs);
384 err = do_arch_prctl(p, ARCH_SET_FS, childregs->r8);
390 if (err && p->thread.io_bitmap_ptr)
391 kfree(p->thread.io_bitmap_ptr);
396 * This special macro can be used to load a debugging register
398 #define loaddebug(thread,r) set_debug(thread->debugreg ## r, r)
401 * switch_to(x,y) should switch tasks from x to y.
403 * This could still be optimized:
404 * - fold all the options into a flag word and test it with a single test.
405 * - could test fs/gs bitsliced
407 struct task_struct *__switch_to(struct task_struct *prev_p, struct task_struct *next_p)
409 struct thread_struct *prev = &prev_p->thread,
410 *next = &next_p->thread;
411 int cpu = smp_processor_id();
412 struct tss_struct *tss = init_tss + cpu;
417 * Reload esp0, LDT and the page table pointer:
419 tss->rsp0 = next->rsp0;
423 * This won't pick up thread selector changes, but I guess that is ok.
425 asm volatile("movl %%es,%0" : "=m" (prev->es));
426 if (unlikely(next->es | prev->es))
427 loadsegment(es, next->es);
429 asm volatile ("movl %%ds,%0" : "=m" (prev->ds));
430 if (unlikely(next->ds | prev->ds))
431 loadsegment(ds, next->ds);
440 asm volatile("movl %%fs,%0" : "=g" (fsindex));
441 /* segment register != 0 always requires a reload.
442 also reload when it has changed.
443 when prev process used 64bit base always reload
444 to avoid an information leak. */
445 if (unlikely(fsindex | next->fsindex | prev->fs)) {
446 loadsegment(fs, next->fsindex);
447 /* check if the user used a selector != 0
448 * if yes clear 64bit base, since overloaded base
449 * is always mapped to the Null selector
454 /* when next process has a 64bit base use it */
456 wrmsrl(MSR_FS_BASE, next->fs);
457 prev->fsindex = fsindex;
461 asm volatile("movl %%gs,%0" : "=g" (gsindex));
462 if (unlikely(gsindex | next->gsindex | prev->gs)) {
463 load_gs_index(next->gsindex);
468 wrmsrl(MSR_KERNEL_GS_BASE, next->gs);
469 prev->gsindex = gsindex;
473 * Switch the PDA context.
475 prev->userrsp = read_pda(oldrsp);
476 write_pda(oldrsp, next->userrsp);
477 write_pda(pcurrent, next_p);
478 write_pda(kernelstack, (unsigned long)next_p->thread_info + THREAD_SIZE - PDA_STACKOFFSET);
481 * Now maybe reload the debug registers
483 if (unlikely(next->debugreg7)) {
495 * Handle the IO bitmap
497 if (unlikely(prev->io_bitmap_ptr || next->io_bitmap_ptr)) {
498 if (next->io_bitmap_ptr) {
500 * 2 cachelines copy ... not good, but not that
501 * bad either. Anyone got something better?
502 * This only affects processes which use ioperm().
504 memcpy(tss->io_bitmap, next->io_bitmap_ptr, IO_BITMAP_BYTES);
505 tss->io_bitmap_base = IO_BITMAP_OFFSET;
508 * a bitmap offset pointing outside of the TSS limit
509 * causes a nicely controllable SIGSEGV if a process
510 * tries to use a port IO instruction. The first
511 * sys_ioperm() call sets up the bitmap properly.
513 tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
521 * sys_execve() executes a new program.
524 long sys_execve(char *name, char **argv,char **envp, struct pt_regs regs)
529 filename = getname(name);
530 error = PTR_ERR(filename);
531 if (IS_ERR(filename))
533 error = do_execve(filename, argv, envp, ®s);
535 current->ptrace &= ~PT_DTRACE;
540 void set_personality_64bit(void)
542 /* inherit personality from parent */
544 /* Make sure to be in 64bit mode */
545 clear_thread_flag(TIF_IA32);
548 asmlinkage long sys_fork(struct pt_regs regs)
550 return do_fork(SIGCHLD, regs.rsp, ®s, 0, NULL, NULL);
553 asmlinkage long sys_clone(unsigned long clone_flags, unsigned long newsp, void *parent_tid, void *child_tid, struct pt_regs regs)
557 return do_fork(clone_flags & ~CLONE_IDLETASK, newsp, ®s, 0,
558 parent_tid, child_tid);
562 * This is trivial, and on the face of it looks like it
563 * could equally well be done in user mode.
565 * Not so, for quite unobvious reasons - register pressure.
566 * In user mode vfork() cannot have a stack frame, and if
567 * done by calling the "clone()" system call directly, you
568 * do not have enough call-clobbered registers to hold all
569 * the information you need.
571 asmlinkage long sys_vfork(struct pt_regs regs)
573 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs.rsp, ®s, 0,
578 * These bracket the sleeping functions..
580 #define first_sched ((unsigned long) scheduling_functions_start_here)
581 #define last_sched ((unsigned long) scheduling_functions_end_here)
583 unsigned long get_wchan(struct task_struct *p)
589 if (!p || p == current || p->state==TASK_RUNNING)
591 stack = (unsigned long)p->thread_info;
592 if (p->thread.rsp < stack || p->thread.rsp > stack+THREAD_SIZE)
594 fp = *(u64 *)(p->thread.rsp);
596 if (fp < (unsigned long)stack || fp > (unsigned long)stack+THREAD_SIZE)
598 rip = *(u64 *)(fp+8);
599 if (rip < first_sched || rip >= last_sched)
602 } while (count++ < 16);
608 long do_arch_prctl(struct task_struct *task, int code, unsigned long addr)
611 int doit = task == current;
616 if (addr >= TASK_SIZE)
619 /* handle small bases via the GDT because that's faster to
621 if (addr <= 0xffffffff) {
622 set_32bit_tls(task, GS_TLS, addr);
624 load_TLS(&task->thread, cpu);
625 load_gs_index(GS_TLS_SEL);
627 task->thread.gsindex = GS_TLS_SEL;
630 task->thread.gsindex = 0;
631 task->thread.gs = addr;
634 ret = checking_wrmsrl(MSR_KERNEL_GS_BASE, addr);
640 /* Not strictly needed for fs, but do it for symmetry
642 if (addr >= TASK_SIZE)
645 /* handle small bases via the GDT because that's faster to
647 if (addr <= 0xffffffff) {
648 set_32bit_tls(task, FS_TLS, addr);
650 load_TLS(&task->thread, cpu);
651 asm volatile("movl %0,%%fs" :: "r" (FS_TLS_SEL));
653 task->thread.fsindex = FS_TLS_SEL;
656 task->thread.fsindex = 0;
657 task->thread.fs = addr;
659 /* set the selector to 0 to not confuse
661 asm volatile("movl %0,%%fs" :: "r" (0));
662 ret = checking_wrmsrl(MSR_FS_BASE, addr);
669 if (task->thread.fsindex == FS_TLS_SEL)
670 base = read_32bit_tls(task, FS_TLS);
672 rdmsrl(MSR_FS_BASE, base);
674 base = task->thread.fs;
675 ret = put_user(base, (unsigned long *)addr);
680 if (task->thread.gsindex == GS_TLS_SEL)
681 base = read_32bit_tls(task, GS_TLS);
683 rdmsrl(MSR_KERNEL_GS_BASE, base);
685 base = task->thread.gs;
686 ret = put_user(base, (unsigned long *)addr);
698 long sys_arch_prctl(int code, unsigned long addr)
700 return do_arch_prctl(current, code, addr);
704 * Capture the user space registers if the task is not running (in user space)
706 int dump_task_regs(struct task_struct *tsk, elf_gregset_t *regs)
708 struct pt_regs *pp, ptregs;
710 pp = (struct pt_regs *)(tsk->thread.rsp0);
717 elf_core_copy_regs(regs, &ptregs);