x86-64: make compat_start_thread() match start_thread()
[linux-2.6.git] / arch / x86 / kernel / process_64.c
blob7cf0a6b6d4bb5b723322022947baf5927cc42f35
1 /*
2 * Copyright (C) 1995 Linus Torvalds
4 * Pentium III FXSR, SSE support
5 * Gareth Hughes <gareth@valinux.com>, May 2000
7 * X86-64 port
8 * Andi Kleen.
10 * CPU hotplug support - ashok.raj@intel.com
14 * This file handles the architecture-dependent parts of process handling..
17 #include <linux/stackprotector.h>
18 #include <linux/cpu.h>
19 #include <linux/errno.h>
20 #include <linux/sched.h>
21 #include <linux/fs.h>
22 #include <linux/kernel.h>
23 #include <linux/mm.h>
24 #include <linux/elfcore.h>
25 #include <linux/smp.h>
26 #include <linux/slab.h>
27 #include <linux/user.h>
28 #include <linux/interrupt.h>
29 #include <linux/utsname.h>
30 #include <linux/delay.h>
31 #include <linux/module.h>
32 #include <linux/ptrace.h>
33 #include <linux/notifier.h>
34 #include <linux/kprobes.h>
35 #include <linux/kdebug.h>
36 #include <linux/tick.h>
37 #include <linux/prctl.h>
38 #include <linux/uaccess.h>
39 #include <linux/io.h>
40 #include <linux/ftrace.h>
41 #include <linux/dmi.h>
43 #include <asm/pgtable.h>
44 #include <asm/system.h>
45 #include <asm/processor.h>
46 #include <asm/i387.h>
47 #include <asm/mmu_context.h>
48 #include <asm/prctl.h>
49 #include <asm/desc.h>
50 #include <asm/proto.h>
51 #include <asm/ia32.h>
52 #include <asm/idle.h>
53 #include <asm/syscalls.h>
54 #include <asm/ds.h>
56 asmlinkage extern void ret_from_fork(void);
58 DEFINE_PER_CPU(unsigned long, old_rsp);
59 static DEFINE_PER_CPU(unsigned char, is_idle);
61 unsigned long kernel_thread_flags = CLONE_VM | CLONE_UNTRACED;
63 static ATOMIC_NOTIFIER_HEAD(idle_notifier);
65 void idle_notifier_register(struct notifier_block *n)
67 atomic_notifier_chain_register(&idle_notifier, n);
69 EXPORT_SYMBOL_GPL(idle_notifier_register);
71 void idle_notifier_unregister(struct notifier_block *n)
73 atomic_notifier_chain_unregister(&idle_notifier, n);
75 EXPORT_SYMBOL_GPL(idle_notifier_unregister);
77 void enter_idle(void)
79 percpu_write(is_idle, 1);
80 atomic_notifier_call_chain(&idle_notifier, IDLE_START, NULL);
83 static void __exit_idle(void)
85 if (x86_test_and_clear_bit_percpu(0, is_idle) == 0)
86 return;
87 atomic_notifier_call_chain(&idle_notifier, IDLE_END, NULL);
90 /* Called from interrupts to signify idle end */
91 void exit_idle(void)
93 /* idle loop has pid 0 */
94 if (current->pid)
95 return;
96 __exit_idle();
99 #ifndef CONFIG_SMP
100 static inline void play_dead(void)
102 BUG();
104 #endif
107 * The idle thread. There's no useful work to be
108 * done, so just try to conserve power and have a
109 * low exit latency (ie sit in a loop waiting for
110 * somebody to say that they'd like to reschedule)
112 void cpu_idle(void)
114 current_thread_info()->status |= TS_POLLING;
117 * If we're the non-boot CPU, nothing set the stack canary up
118 * for us. CPU0 already has it initialized but no harm in
119 * doing it again. This is a good place for updating it, as
120 * we wont ever return from this function (so the invalid
121 * canaries already on the stack wont ever trigger).
123 boot_init_stack_canary();
125 /* endless idle loop with no priority at all */
126 while (1) {
127 tick_nohz_stop_sched_tick(1);
128 while (!need_resched()) {
130 rmb();
132 if (cpu_is_offline(smp_processor_id()))
133 play_dead();
135 * Idle routines should keep interrupts disabled
136 * from here on, until they go to idle.
137 * Otherwise, idle callbacks can misfire.
139 local_irq_disable();
140 enter_idle();
141 /* Don't trace irqs off for idle */
142 stop_critical_timings();
143 pm_idle();
144 start_critical_timings();
145 /* In many cases the interrupt that ended idle
146 has already called exit_idle. But some idle
147 loops can be woken up without interrupt. */
148 __exit_idle();
151 tick_nohz_restart_sched_tick();
152 preempt_enable_no_resched();
153 schedule();
154 preempt_disable();
158 /* Prints also some state that isn't saved in the pt_regs */
159 void __show_regs(struct pt_regs *regs, int all)
161 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L, fs, gs, shadowgs;
162 unsigned long d0, d1, d2, d3, d6, d7;
163 unsigned int fsindex, gsindex;
164 unsigned int ds, cs, es;
165 const char *board;
167 printk("\n");
168 print_modules();
169 board = dmi_get_system_info(DMI_PRODUCT_NAME);
170 if (!board)
171 board = "";
172 printk(KERN_INFO "Pid: %d, comm: %.20s %s %s %.*s %s\n",
173 current->pid, current->comm, print_tainted(),
174 init_utsname()->release,
175 (int)strcspn(init_utsname()->version, " "),
176 init_utsname()->version, board);
177 printk(KERN_INFO "RIP: %04lx:[<%016lx>] ", regs->cs & 0xffff, regs->ip);
178 printk_address(regs->ip, 1);
179 printk(KERN_INFO "RSP: %04lx:%016lx EFLAGS: %08lx\n", regs->ss,
180 regs->sp, regs->flags);
181 printk(KERN_INFO "RAX: %016lx RBX: %016lx RCX: %016lx\n",
182 regs->ax, regs->bx, regs->cx);
183 printk(KERN_INFO "RDX: %016lx RSI: %016lx RDI: %016lx\n",
184 regs->dx, regs->si, regs->di);
185 printk(KERN_INFO "RBP: %016lx R08: %016lx R09: %016lx\n",
186 regs->bp, regs->r8, regs->r9);
187 printk(KERN_INFO "R10: %016lx R11: %016lx R12: %016lx\n",
188 regs->r10, regs->r11, regs->r12);
189 printk(KERN_INFO "R13: %016lx R14: %016lx R15: %016lx\n",
190 regs->r13, regs->r14, regs->r15);
192 asm("movl %%ds,%0" : "=r" (ds));
193 asm("movl %%cs,%0" : "=r" (cs));
194 asm("movl %%es,%0" : "=r" (es));
195 asm("movl %%fs,%0" : "=r" (fsindex));
196 asm("movl %%gs,%0" : "=r" (gsindex));
198 rdmsrl(MSR_FS_BASE, fs);
199 rdmsrl(MSR_GS_BASE, gs);
200 rdmsrl(MSR_KERNEL_GS_BASE, shadowgs);
202 if (!all)
203 return;
205 cr0 = read_cr0();
206 cr2 = read_cr2();
207 cr3 = read_cr3();
208 cr4 = read_cr4();
210 printk(KERN_INFO "FS: %016lx(%04x) GS:%016lx(%04x) knlGS:%016lx\n",
211 fs, fsindex, gs, gsindex, shadowgs);
212 printk(KERN_INFO "CS: %04x DS: %04x ES: %04x CR0: %016lx\n", cs, ds,
213 es, cr0);
214 printk(KERN_INFO "CR2: %016lx CR3: %016lx CR4: %016lx\n", cr2, cr3,
215 cr4);
217 get_debugreg(d0, 0);
218 get_debugreg(d1, 1);
219 get_debugreg(d2, 2);
220 printk(KERN_INFO "DR0: %016lx DR1: %016lx DR2: %016lx\n", d0, d1, d2);
221 get_debugreg(d3, 3);
222 get_debugreg(d6, 6);
223 get_debugreg(d7, 7);
224 printk(KERN_INFO "DR3: %016lx DR6: %016lx DR7: %016lx\n", d3, d6, d7);
227 void show_regs(struct pt_regs *regs)
229 printk(KERN_INFO "CPU %d:", smp_processor_id());
230 __show_regs(regs, 1);
231 show_trace(NULL, regs, (void *)(regs + 1), regs->bp);
234 void release_thread(struct task_struct *dead_task)
236 if (dead_task->mm) {
237 if (dead_task->mm->context.size) {
238 printk("WARNING: dead process %8s still has LDT? <%p/%d>\n",
239 dead_task->comm,
240 dead_task->mm->context.ldt,
241 dead_task->mm->context.size);
242 BUG();
247 static inline void set_32bit_tls(struct task_struct *t, int tls, u32 addr)
249 struct user_desc ud = {
250 .base_addr = addr,
251 .limit = 0xfffff,
252 .seg_32bit = 1,
253 .limit_in_pages = 1,
254 .useable = 1,
256 struct desc_struct *desc = t->thread.tls_array;
257 desc += tls;
258 fill_ldt(desc, &ud);
261 static inline u32 read_32bit_tls(struct task_struct *t, int tls)
263 return get_desc_base(&t->thread.tls_array[tls]);
267 * This gets called before we allocate a new thread and copy
268 * the current task into it.
270 void prepare_to_copy(struct task_struct *tsk)
272 unlazy_fpu(tsk);
275 int copy_thread(unsigned long clone_flags, unsigned long sp,
276 unsigned long unused,
277 struct task_struct *p, struct pt_regs *regs)
279 int err;
280 struct pt_regs *childregs;
281 struct task_struct *me = current;
283 childregs = ((struct pt_regs *)
284 (THREAD_SIZE + task_stack_page(p))) - 1;
285 *childregs = *regs;
287 childregs->ax = 0;
288 childregs->sp = sp;
289 if (sp == ~0UL)
290 childregs->sp = (unsigned long)childregs;
292 p->thread.sp = (unsigned long) childregs;
293 p->thread.sp0 = (unsigned long) (childregs+1);
294 p->thread.usersp = me->thread.usersp;
296 set_tsk_thread_flag(p, TIF_FORK);
298 p->thread.fs = me->thread.fs;
299 p->thread.gs = me->thread.gs;
301 savesegment(gs, p->thread.gsindex);
302 savesegment(fs, p->thread.fsindex);
303 savesegment(es, p->thread.es);
304 savesegment(ds, p->thread.ds);
306 if (unlikely(test_tsk_thread_flag(me, TIF_IO_BITMAP))) {
307 p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL);
308 if (!p->thread.io_bitmap_ptr) {
309 p->thread.io_bitmap_max = 0;
310 return -ENOMEM;
312 memcpy(p->thread.io_bitmap_ptr, me->thread.io_bitmap_ptr,
313 IO_BITMAP_BYTES);
314 set_tsk_thread_flag(p, TIF_IO_BITMAP);
318 * Set a new TLS for the child thread?
320 if (clone_flags & CLONE_SETTLS) {
321 #ifdef CONFIG_IA32_EMULATION
322 if (test_thread_flag(TIF_IA32))
323 err = do_set_thread_area(p, -1,
324 (struct user_desc __user *)childregs->si, 0);
325 else
326 #endif
327 err = do_arch_prctl(p, ARCH_SET_FS, childregs->r8);
328 if (err)
329 goto out;
332 clear_tsk_thread_flag(p, TIF_DS_AREA_MSR);
333 p->thread.ds_ctx = NULL;
335 clear_tsk_thread_flag(p, TIF_DEBUGCTLMSR);
336 p->thread.debugctlmsr = 0;
338 err = 0;
339 out:
340 if (err && p->thread.io_bitmap_ptr) {
341 kfree(p->thread.io_bitmap_ptr);
342 p->thread.io_bitmap_max = 0;
344 return err;
347 void
348 start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp)
350 loadsegment(fs, 0);
351 loadsegment(es, 0);
352 loadsegment(ds, 0);
353 load_gs_index(0);
354 regs->ip = new_ip;
355 regs->sp = new_sp;
356 percpu_write(old_rsp, new_sp);
357 regs->cs = __USER_CS;
358 regs->ss = __USER_DS;
359 regs->flags = X86_EFLAGS_IF;
360 set_fs(USER_DS);
362 * Free the old FP and other extended state
364 free_thread_xstate(current);
366 EXPORT_SYMBOL_GPL(start_thread);
368 #ifdef CONFIG_IA32_EMULATION
369 void start_thread_ia32(struct pt_regs *regs, u32 new_ip, u32 new_sp)
371 loadsegment(fs, 0);
372 loadsegment(ds, __USER32_DS);
373 loadsegment(es, __USER32_DS);
374 load_gs_index(0);
375 regs->ip = new_ip;
376 regs->sp = new_sp;
377 percpu_write(old_rsp, new_sp);
378 regs->cs = __USER32_CS;
379 regs->ss = __USER32_DS;
380 regs->flags = X86_EFLAGS_IF;
381 set_fs(USER_DS);
383 * Free the old FP and other extended state
385 free_thread_xstate(current);
387 #endif
390 * switch_to(x,y) should switch tasks from x to y.
392 * This could still be optimized:
393 * - fold all the options into a flag word and test it with a single test.
394 * - could test fs/gs bitsliced
396 * Kprobes not supported here. Set the probe on schedule instead.
397 * Function graph tracer not supported too.
399 __notrace_funcgraph struct task_struct *
400 __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
402 struct thread_struct *prev = &prev_p->thread;
403 struct thread_struct *next = &next_p->thread;
404 int cpu = smp_processor_id();
405 struct tss_struct *tss = &per_cpu(init_tss, cpu);
406 unsigned fsindex, gsindex;
407 bool preload_fpu;
410 * If the task has used fpu the last 5 timeslices, just do a full
411 * restore of the math state immediately to avoid the trap; the
412 * chances of needing FPU soon are obviously high now
414 preload_fpu = tsk_used_math(next_p) && next_p->fpu_counter > 5;
416 /* we're going to use this soon, after a few expensive things */
417 if (preload_fpu)
418 prefetch(next->xstate);
421 * Reload esp0, LDT and the page table pointer:
423 load_sp0(tss, next);
426 * Switch DS and ES.
427 * This won't pick up thread selector changes, but I guess that is ok.
429 savesegment(es, prev->es);
430 if (unlikely(next->es | prev->es))
431 loadsegment(es, next->es);
433 savesegment(ds, prev->ds);
434 if (unlikely(next->ds | prev->ds))
435 loadsegment(ds, next->ds);
438 /* We must save %fs and %gs before load_TLS() because
439 * %fs and %gs may be cleared by load_TLS().
441 * (e.g. xen_load_tls())
443 savesegment(fs, fsindex);
444 savesegment(gs, gsindex);
446 load_TLS(next, cpu);
448 /* Must be after DS reload */
449 unlazy_fpu(prev_p);
451 /* Make sure cpu is ready for new context */
452 if (preload_fpu)
453 clts();
456 * Leave lazy mode, flushing any hypercalls made here.
457 * This must be done before restoring TLS segments so
458 * the GDT and LDT are properly updated, and must be
459 * done before math_state_restore, so the TS bit is up
460 * to date.
462 arch_end_context_switch(next_p);
465 * Switch FS and GS.
467 * Segment register != 0 always requires a reload. Also
468 * reload when it has changed. When prev process used 64bit
469 * base always reload to avoid an information leak.
471 if (unlikely(fsindex | next->fsindex | prev->fs)) {
472 loadsegment(fs, next->fsindex);
474 * Check if the user used a selector != 0; if yes
475 * clear 64bit base, since overloaded base is always
476 * mapped to the Null selector
478 if (fsindex)
479 prev->fs = 0;
481 /* when next process has a 64bit base use it */
482 if (next->fs)
483 wrmsrl(MSR_FS_BASE, next->fs);
484 prev->fsindex = fsindex;
486 if (unlikely(gsindex | next->gsindex | prev->gs)) {
487 load_gs_index(next->gsindex);
488 if (gsindex)
489 prev->gs = 0;
491 if (next->gs)
492 wrmsrl(MSR_KERNEL_GS_BASE, next->gs);
493 prev->gsindex = gsindex;
496 * Switch the PDA and FPU contexts.
498 prev->usersp = percpu_read(old_rsp);
499 percpu_write(old_rsp, next->usersp);
500 percpu_write(current_task, next_p);
502 percpu_write(kernel_stack,
503 (unsigned long)task_stack_page(next_p) +
504 THREAD_SIZE - KERNEL_STACK_OFFSET);
507 * Now maybe reload the debug registers and handle I/O bitmaps
509 if (unlikely(task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT ||
510 task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV))
511 __switch_to_xtra(prev_p, next_p, tss);
514 * Preload the FPU context, now that we've determined that the
515 * task is likely to be using it.
517 if (preload_fpu)
518 __math_state_restore();
519 return prev_p;
523 * sys_execve() executes a new program.
525 asmlinkage
526 long sys_execve(char __user *name, char __user * __user *argv,
527 char __user * __user *envp, struct pt_regs *regs)
529 long error;
530 char *filename;
532 filename = getname(name);
533 error = PTR_ERR(filename);
534 if (IS_ERR(filename))
535 return error;
536 error = do_execve(filename, argv, envp, regs);
537 putname(filename);
538 return error;
541 void set_personality_64bit(void)
543 /* inherit personality from parent */
545 /* Make sure to be in 64bit mode */
546 clear_thread_flag(TIF_IA32);
548 /* TBD: overwrites user setup. Should have two bits.
549 But 64bit processes have always behaved this way,
550 so it's not too bad. The main problem is just that
551 32bit childs are affected again. */
552 current->personality &= ~READ_IMPLIES_EXEC;
555 asmlinkage long
556 sys_clone(unsigned long clone_flags, unsigned long newsp,
557 void __user *parent_tid, void __user *child_tid, struct pt_regs *regs)
559 if (!newsp)
560 newsp = regs->sp;
561 return do_fork(clone_flags, newsp, regs, 0, parent_tid, child_tid);
564 unsigned long get_wchan(struct task_struct *p)
566 unsigned long stack;
567 u64 fp, ip;
568 int count = 0;
570 if (!p || p == current || p->state == TASK_RUNNING)
571 return 0;
572 stack = (unsigned long)task_stack_page(p);
573 if (p->thread.sp < stack || p->thread.sp >= stack+THREAD_SIZE)
574 return 0;
575 fp = *(u64 *)(p->thread.sp);
576 do {
577 if (fp < (unsigned long)stack ||
578 fp >= (unsigned long)stack+THREAD_SIZE)
579 return 0;
580 ip = *(u64 *)(fp+8);
581 if (!in_sched_functions(ip))
582 return ip;
583 fp = *(u64 *)fp;
584 } while (count++ < 16);
585 return 0;
588 long do_arch_prctl(struct task_struct *task, int code, unsigned long addr)
590 int ret = 0;
591 int doit = task == current;
592 int cpu;
594 switch (code) {
595 case ARCH_SET_GS:
596 if (addr >= TASK_SIZE_OF(task))
597 return -EPERM;
598 cpu = get_cpu();
599 /* handle small bases via the GDT because that's faster to
600 switch. */
601 if (addr <= 0xffffffff) {
602 set_32bit_tls(task, GS_TLS, addr);
603 if (doit) {
604 load_TLS(&task->thread, cpu);
605 load_gs_index(GS_TLS_SEL);
607 task->thread.gsindex = GS_TLS_SEL;
608 task->thread.gs = 0;
609 } else {
610 task->thread.gsindex = 0;
611 task->thread.gs = addr;
612 if (doit) {
613 load_gs_index(0);
614 ret = checking_wrmsrl(MSR_KERNEL_GS_BASE, addr);
617 put_cpu();
618 break;
619 case ARCH_SET_FS:
620 /* Not strictly needed for fs, but do it for symmetry
621 with gs */
622 if (addr >= TASK_SIZE_OF(task))
623 return -EPERM;
624 cpu = get_cpu();
625 /* handle small bases via the GDT because that's faster to
626 switch. */
627 if (addr <= 0xffffffff) {
628 set_32bit_tls(task, FS_TLS, addr);
629 if (doit) {
630 load_TLS(&task->thread, cpu);
631 loadsegment(fs, FS_TLS_SEL);
633 task->thread.fsindex = FS_TLS_SEL;
634 task->thread.fs = 0;
635 } else {
636 task->thread.fsindex = 0;
637 task->thread.fs = addr;
638 if (doit) {
639 /* set the selector to 0 to not confuse
640 __switch_to */
641 loadsegment(fs, 0);
642 ret = checking_wrmsrl(MSR_FS_BASE, addr);
645 put_cpu();
646 break;
647 case ARCH_GET_FS: {
648 unsigned long base;
649 if (task->thread.fsindex == FS_TLS_SEL)
650 base = read_32bit_tls(task, FS_TLS);
651 else if (doit)
652 rdmsrl(MSR_FS_BASE, base);
653 else
654 base = task->thread.fs;
655 ret = put_user(base, (unsigned long __user *)addr);
656 break;
658 case ARCH_GET_GS: {
659 unsigned long base;
660 unsigned gsindex;
661 if (task->thread.gsindex == GS_TLS_SEL)
662 base = read_32bit_tls(task, GS_TLS);
663 else if (doit) {
664 savesegment(gs, gsindex);
665 if (gsindex)
666 rdmsrl(MSR_KERNEL_GS_BASE, base);
667 else
668 base = task->thread.gs;
669 } else
670 base = task->thread.gs;
671 ret = put_user(base, (unsigned long __user *)addr);
672 break;
675 default:
676 ret = -EINVAL;
677 break;
680 return ret;
683 long sys_arch_prctl(int code, unsigned long addr)
685 return do_arch_prctl(current, code, addr);