2 * Copyright (C) 1995 Linus Torvalds
4 * Pentium III FXSR, SSE support
5 * Gareth Hughes <gareth@valinux.com>, May 2000
10 * CPU hotplug support - ashok.raj@intel.com
14 * This file handles the architecture-dependent parts of process handling..
19 #include <linux/cpu.h>
20 #include <linux/errno.h>
21 #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/interrupt.h>
30 #include <linux/utsname.h>
31 #include <linux/delay.h>
32 #include <linux/module.h>
33 #include <linux/ptrace.h>
34 #include <linux/random.h>
35 #include <linux/notifier.h>
36 #include <linux/kprobes.h>
37 #include <linux/kdebug.h>
38 #include <linux/tick.h>
39 #include <linux/prctl.h>
40 #include <linux/uaccess.h>
42 #include <linux/ftrace.h>
43 #include <linux/dmi.h>
45 #include <asm/pgtable.h>
46 #include <asm/system.h>
47 #include <asm/processor.h>
49 #include <asm/mmu_context.h>
51 #include <asm/prctl.h>
53 #include <asm/proto.h>
56 #include <asm/syscalls.h>
59 asmlinkage extern void ret_from_fork(void);
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);
80 atomic_notifier_call_chain(&idle_notifier, IDLE_START, NULL);
83 static void __exit_idle(void)
85 if (test_and_clear_bit_pda(0, isidle) == 0)
87 atomic_notifier_call_chain(&idle_notifier, IDLE_END, NULL);
90 /* Called from interrupts to signify idle end */
93 /* idle loop has pid 0 */
100 static inline void play_dead(void)
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)
114 current_thread_info()->status |= TS_POLLING;
115 /* endless idle loop with no priority at all */
117 tick_nohz_stop_sched_tick(1);
118 while (!need_resched()) {
122 if (cpu_is_offline(smp_processor_id()))
125 * Idle routines should keep interrupts disabled
126 * from here on, until they go to idle.
127 * Otherwise, idle callbacks can misfire.
131 /* Don't trace irqs off for idle */
132 stop_critical_timings();
134 start_critical_timings();
135 /* In many cases the interrupt that ended idle
136 has already called exit_idle. But some idle
137 loops can be woken up without interrupt. */
141 tick_nohz_restart_sched_tick();
142 preempt_enable_no_resched();
148 /* Prints also some state that isn't saved in the pt_regs */
149 void __show_regs(struct pt_regs *regs, int all)
151 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L, fs, gs, shadowgs;
152 unsigned long d0, d1, d2, d3, d6, d7;
153 unsigned int fsindex, gsindex;
154 unsigned int ds, cs, es;
159 board = dmi_get_system_info(DMI_PRODUCT_NAME);
162 printk(KERN_INFO "Pid: %d, comm: %.20s %s %s %.*s %s\n",
163 current->pid, current->comm, print_tainted(),
164 init_utsname()->release,
165 (int)strcspn(init_utsname()->version, " "),
166 init_utsname()->version, board);
167 printk(KERN_INFO "RIP: %04lx:[<%016lx>] ", regs->cs & 0xffff, regs->ip);
168 printk_address(regs->ip, 1);
169 printk(KERN_INFO "RSP: %04lx:%016lx EFLAGS: %08lx\n", regs->ss,
170 regs->sp, regs->flags);
171 printk(KERN_INFO "RAX: %016lx RBX: %016lx RCX: %016lx\n",
172 regs->ax, regs->bx, regs->cx);
173 printk(KERN_INFO "RDX: %016lx RSI: %016lx RDI: %016lx\n",
174 regs->dx, regs->si, regs->di);
175 printk(KERN_INFO "RBP: %016lx R08: %016lx R09: %016lx\n",
176 regs->bp, regs->r8, regs->r9);
177 printk(KERN_INFO "R10: %016lx R11: %016lx R12: %016lx\n",
178 regs->r10, regs->r11, regs->r12);
179 printk(KERN_INFO "R13: %016lx R14: %016lx R15: %016lx\n",
180 regs->r13, regs->r14, regs->r15);
182 asm("movl %%ds,%0" : "=r" (ds));
183 asm("movl %%cs,%0" : "=r" (cs));
184 asm("movl %%es,%0" : "=r" (es));
185 asm("movl %%fs,%0" : "=r" (fsindex));
186 asm("movl %%gs,%0" : "=r" (gsindex));
188 rdmsrl(MSR_FS_BASE, fs);
189 rdmsrl(MSR_GS_BASE, gs);
190 rdmsrl(MSR_KERNEL_GS_BASE, shadowgs);
200 printk(KERN_INFO "FS: %016lx(%04x) GS:%016lx(%04x) knlGS:%016lx\n",
201 fs, fsindex, gs, gsindex, shadowgs);
202 printk(KERN_INFO "CS: %04x DS: %04x ES: %04x CR0: %016lx\n", cs, ds,
204 printk(KERN_INFO "CR2: %016lx CR3: %016lx CR4: %016lx\n", cr2, cr3,
210 printk(KERN_INFO "DR0: %016lx DR1: %016lx DR2: %016lx\n", d0, d1, d2);
214 printk(KERN_INFO "DR3: %016lx DR6: %016lx DR7: %016lx\n", d3, d6, d7);
217 void show_regs(struct pt_regs *regs)
219 printk(KERN_INFO "CPU %d:", smp_processor_id());
220 __show_regs(regs, 1);
221 show_trace(NULL, regs, (void *)(regs + 1), regs->bp);
225 * Free current thread data structures etc..
227 void exit_thread(void)
229 struct task_struct *me = current;
230 struct thread_struct *t = &me->thread;
232 if (me->thread.io_bitmap_ptr) {
233 struct tss_struct *tss = &per_cpu(init_tss, get_cpu());
235 kfree(t->io_bitmap_ptr);
236 t->io_bitmap_ptr = NULL;
237 clear_thread_flag(TIF_IO_BITMAP);
239 * Careful, clear this in the TSS too:
241 memset(tss->io_bitmap, 0xff, t->io_bitmap_max);
242 t->io_bitmap_max = 0;
246 ds_exit_thread(current);
249 void flush_thread(void)
251 struct task_struct *tsk = current;
253 if (test_tsk_thread_flag(tsk, TIF_ABI_PENDING)) {
254 clear_tsk_thread_flag(tsk, TIF_ABI_PENDING);
255 if (test_tsk_thread_flag(tsk, TIF_IA32)) {
256 clear_tsk_thread_flag(tsk, TIF_IA32);
258 set_tsk_thread_flag(tsk, TIF_IA32);
259 current_thread_info()->status |= TS_COMPAT;
262 clear_tsk_thread_flag(tsk, TIF_DEBUG);
264 tsk->thread.debugreg0 = 0;
265 tsk->thread.debugreg1 = 0;
266 tsk->thread.debugreg2 = 0;
267 tsk->thread.debugreg3 = 0;
268 tsk->thread.debugreg6 = 0;
269 tsk->thread.debugreg7 = 0;
270 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
272 * Forget coprocessor state..
274 tsk->fpu_counter = 0;
279 void release_thread(struct task_struct *dead_task)
282 if (dead_task->mm->context.size) {
283 printk("WARNING: dead process %8s still has LDT? <%p/%d>\n",
285 dead_task->mm->context.ldt,
286 dead_task->mm->context.size);
292 static inline void set_32bit_tls(struct task_struct *t, int tls, u32 addr)
294 struct user_desc ud = {
301 struct desc_struct *desc = t->thread.tls_array;
306 static inline u32 read_32bit_tls(struct task_struct *t, int tls)
308 return get_desc_base(&t->thread.tls_array[tls]);
312 * This gets called before we allocate a new thread and copy
313 * the current task into it.
315 void prepare_to_copy(struct task_struct *tsk)
320 int copy_thread(int nr, unsigned long clone_flags, unsigned long sp,
321 unsigned long unused,
322 struct task_struct *p, struct pt_regs *regs)
325 struct pt_regs *childregs;
326 struct task_struct *me = current;
328 childregs = ((struct pt_regs *)
329 (THREAD_SIZE + task_stack_page(p))) - 1;
335 childregs->sp = (unsigned long)childregs;
337 p->thread.sp = (unsigned long) childregs;
338 p->thread.sp0 = (unsigned long) (childregs+1);
339 p->thread.usersp = me->thread.usersp;
341 set_tsk_thread_flag(p, TIF_FORK);
343 p->thread.fs = me->thread.fs;
344 p->thread.gs = me->thread.gs;
346 savesegment(gs, p->thread.gsindex);
347 savesegment(fs, p->thread.fsindex);
348 savesegment(es, p->thread.es);
349 savesegment(ds, p->thread.ds);
351 if (unlikely(test_tsk_thread_flag(me, TIF_IO_BITMAP))) {
352 p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL);
353 if (!p->thread.io_bitmap_ptr) {
354 p->thread.io_bitmap_max = 0;
357 memcpy(p->thread.io_bitmap_ptr, me->thread.io_bitmap_ptr,
359 set_tsk_thread_flag(p, TIF_IO_BITMAP);
363 * Set a new TLS for the child thread?
365 if (clone_flags & CLONE_SETTLS) {
366 #ifdef CONFIG_IA32_EMULATION
367 if (test_thread_flag(TIF_IA32))
368 err = do_set_thread_area(p, -1,
369 (struct user_desc __user *)childregs->si, 0);
372 err = do_arch_prctl(p, ARCH_SET_FS, childregs->r8);
377 ds_copy_thread(p, me);
379 clear_tsk_thread_flag(p, TIF_DEBUGCTLMSR);
380 p->thread.debugctlmsr = 0;
384 if (err && p->thread.io_bitmap_ptr) {
385 kfree(p->thread.io_bitmap_ptr);
386 p->thread.io_bitmap_max = 0;
392 start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp)
400 write_pda(oldrsp, new_sp);
401 regs->cs = __USER_CS;
402 regs->ss = __USER_DS;
406 * Free the old FP and other extended state
408 free_thread_xstate(current);
410 EXPORT_SYMBOL_GPL(start_thread);
412 static void hard_disable_TSC(void)
414 write_cr4(read_cr4() | X86_CR4_TSD);
417 void disable_TSC(void)
420 if (!test_and_set_thread_flag(TIF_NOTSC))
422 * Must flip the CPU state synchronously with
423 * TIF_NOTSC in the current running context.
429 static void hard_enable_TSC(void)
431 write_cr4(read_cr4() & ~X86_CR4_TSD);
434 static void enable_TSC(void)
437 if (test_and_clear_thread_flag(TIF_NOTSC))
439 * Must flip the CPU state synchronously with
440 * TIF_NOTSC in the current running context.
446 int get_tsc_mode(unsigned long adr)
450 if (test_thread_flag(TIF_NOTSC))
451 val = PR_TSC_SIGSEGV;
455 return put_user(val, (unsigned int __user *)adr);
458 int set_tsc_mode(unsigned int val)
460 if (val == PR_TSC_SIGSEGV)
462 else if (val == PR_TSC_ENABLE)
471 * This special macro can be used to load a debugging register
473 #define loaddebug(thread, r) set_debugreg(thread->debugreg ## r, r)
475 static inline void __switch_to_xtra(struct task_struct *prev_p,
476 struct task_struct *next_p,
477 struct tss_struct *tss)
479 struct thread_struct *prev, *next;
481 prev = &prev_p->thread,
482 next = &next_p->thread;
484 if (test_tsk_thread_flag(next_p, TIF_DS_AREA_MSR) ||
485 test_tsk_thread_flag(prev_p, TIF_DS_AREA_MSR))
486 ds_switch_to(prev_p, next_p);
487 else if (next->debugctlmsr != prev->debugctlmsr)
488 update_debugctlmsr(next->debugctlmsr);
490 if (test_tsk_thread_flag(next_p, TIF_DEBUG)) {
500 if (test_tsk_thread_flag(prev_p, TIF_NOTSC) ^
501 test_tsk_thread_flag(next_p, TIF_NOTSC)) {
502 /* prev and next are different */
503 if (test_tsk_thread_flag(next_p, TIF_NOTSC))
509 if (test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
511 * Copy the relevant range of the IO bitmap.
512 * Normally this is 128 bytes or less:
514 memcpy(tss->io_bitmap, next->io_bitmap_ptr,
515 max(prev->io_bitmap_max, next->io_bitmap_max));
516 } else if (test_tsk_thread_flag(prev_p, TIF_IO_BITMAP)) {
518 * Clear any possible leftover bits:
520 memset(tss->io_bitmap, 0xff, prev->io_bitmap_max);
525 * switch_to(x,y) should switch tasks from x to y.
527 * This could still be optimized:
528 * - fold all the options into a flag word and test it with a single test.
529 * - could test fs/gs bitsliced
531 * Kprobes not supported here. Set the probe on schedule instead.
532 * Function graph tracer not supported too.
534 __notrace_funcgraph struct task_struct *
535 __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
537 struct thread_struct *prev = &prev_p->thread;
538 struct thread_struct *next = &next_p->thread;
539 int cpu = smp_processor_id();
540 struct tss_struct *tss = &per_cpu(init_tss, cpu);
541 unsigned fsindex, gsindex;
543 /* we're going to use this soon, after a few expensive things */
544 if (next_p->fpu_counter > 5)
545 prefetch(next->xstate);
548 * Reload esp0, LDT and the page table pointer:
554 * This won't pick up thread selector changes, but I guess that is ok.
556 savesegment(es, prev->es);
557 if (unlikely(next->es | prev->es))
558 loadsegment(es, next->es);
560 savesegment(ds, prev->ds);
561 if (unlikely(next->ds | prev->ds))
562 loadsegment(ds, next->ds);
565 /* We must save %fs and %gs before load_TLS() because
566 * %fs and %gs may be cleared by load_TLS().
568 * (e.g. xen_load_tls())
570 savesegment(fs, fsindex);
571 savesegment(gs, gsindex);
576 * Leave lazy mode, flushing any hypercalls made here.
577 * This must be done before restoring TLS segments so
578 * the GDT and LDT are properly updated, and must be
579 * done before math_state_restore, so the TS bit is up
582 arch_leave_lazy_cpu_mode();
587 * Segment register != 0 always requires a reload. Also
588 * reload when it has changed. When prev process used 64bit
589 * base always reload to avoid an information leak.
591 if (unlikely(fsindex | next->fsindex | prev->fs)) {
592 loadsegment(fs, next->fsindex);
594 * Check if the user used a selector != 0; if yes
595 * clear 64bit base, since overloaded base is always
596 * mapped to the Null selector
601 /* when next process has a 64bit base use it */
603 wrmsrl(MSR_FS_BASE, next->fs);
604 prev->fsindex = fsindex;
606 if (unlikely(gsindex | next->gsindex | prev->gs)) {
607 load_gs_index(next->gsindex);
612 wrmsrl(MSR_KERNEL_GS_BASE, next->gs);
613 prev->gsindex = gsindex;
615 /* Must be after DS reload */
619 * Switch the PDA and FPU contexts.
621 prev->usersp = read_pda(oldrsp);
622 write_pda(oldrsp, next->usersp);
623 write_pda(pcurrent, next_p);
625 write_pda(kernelstack,
626 (unsigned long)task_stack_page(next_p) +
627 THREAD_SIZE - PDA_STACKOFFSET);
628 #ifdef CONFIG_CC_STACKPROTECTOR
629 write_pda(stack_canary, next_p->stack_canary);
631 * Build time only check to make sure the stack_canary is at
632 * offset 40 in the pda; this is a gcc ABI requirement
634 BUILD_BUG_ON(offsetof(struct x8664_pda, stack_canary) != 40);
638 * Now maybe reload the debug registers and handle I/O bitmaps
640 if (unlikely(task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT ||
641 task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV))
642 __switch_to_xtra(prev_p, next_p, tss);
644 /* If the task has used fpu the last 5 timeslices, just do a full
645 * restore of the math state immediately to avoid the trap; the
646 * chances of needing FPU soon are obviously high now
648 * tsk_used_math() checks prevent calling math_state_restore(),
649 * which can sleep in the case of !tsk_used_math()
651 if (tsk_used_math(next_p) && next_p->fpu_counter > 5)
652 math_state_restore();
657 * sys_execve() executes a new program.
660 long sys_execve(char __user *name, char __user * __user *argv,
661 char __user * __user *envp, struct pt_regs *regs)
666 filename = getname(name);
667 error = PTR_ERR(filename);
668 if (IS_ERR(filename))
670 error = do_execve(filename, argv, envp, regs);
675 void set_personality_64bit(void)
677 /* inherit personality from parent */
679 /* Make sure to be in 64bit mode */
680 clear_thread_flag(TIF_IA32);
682 /* TBD: overwrites user setup. Should have two bits.
683 But 64bit processes have always behaved this way,
684 so it's not too bad. The main problem is just that
685 32bit childs are affected again. */
686 current->personality &= ~READ_IMPLIES_EXEC;
689 asmlinkage long sys_fork(struct pt_regs *regs)
691 return do_fork(SIGCHLD, regs->sp, regs, 0, NULL, NULL);
695 sys_clone(unsigned long clone_flags, unsigned long newsp,
696 void __user *parent_tid, void __user *child_tid, struct pt_regs *regs)
700 return do_fork(clone_flags, newsp, regs, 0, parent_tid, child_tid);
704 * This is trivial, and on the face of it looks like it
705 * could equally well be done in user mode.
707 * Not so, for quite unobvious reasons - register pressure.
708 * In user mode vfork() cannot have a stack frame, and if
709 * done by calling the "clone()" system call directly, you
710 * do not have enough call-clobbered registers to hold all
711 * the information you need.
713 asmlinkage long sys_vfork(struct pt_regs *regs)
715 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->sp, regs, 0,
719 unsigned long get_wchan(struct task_struct *p)
725 if (!p || p == current || p->state == TASK_RUNNING)
727 stack = (unsigned long)task_stack_page(p);
728 if (p->thread.sp < stack || p->thread.sp >= stack+THREAD_SIZE)
730 fp = *(u64 *)(p->thread.sp);
732 if (fp < (unsigned long)stack ||
733 fp >= (unsigned long)stack+THREAD_SIZE)
736 if (!in_sched_functions(ip))
739 } while (count++ < 16);
743 long do_arch_prctl(struct task_struct *task, int code, unsigned long addr)
746 int doit = task == current;
751 if (addr >= TASK_SIZE_OF(task))
754 /* handle small bases via the GDT because that's faster to
756 if (addr <= 0xffffffff) {
757 set_32bit_tls(task, GS_TLS, addr);
759 load_TLS(&task->thread, cpu);
760 load_gs_index(GS_TLS_SEL);
762 task->thread.gsindex = GS_TLS_SEL;
765 task->thread.gsindex = 0;
766 task->thread.gs = addr;
769 ret = checking_wrmsrl(MSR_KERNEL_GS_BASE, addr);
775 /* Not strictly needed for fs, but do it for symmetry
777 if (addr >= TASK_SIZE_OF(task))
780 /* handle small bases via the GDT because that's faster to
782 if (addr <= 0xffffffff) {
783 set_32bit_tls(task, FS_TLS, addr);
785 load_TLS(&task->thread, cpu);
786 loadsegment(fs, FS_TLS_SEL);
788 task->thread.fsindex = FS_TLS_SEL;
791 task->thread.fsindex = 0;
792 task->thread.fs = addr;
794 /* set the selector to 0 to not confuse
797 ret = checking_wrmsrl(MSR_FS_BASE, addr);
804 if (task->thread.fsindex == FS_TLS_SEL)
805 base = read_32bit_tls(task, FS_TLS);
807 rdmsrl(MSR_FS_BASE, base);
809 base = task->thread.fs;
810 ret = put_user(base, (unsigned long __user *)addr);
816 if (task->thread.gsindex == GS_TLS_SEL)
817 base = read_32bit_tls(task, GS_TLS);
819 savesegment(gs, gsindex);
821 rdmsrl(MSR_KERNEL_GS_BASE, base);
823 base = task->thread.gs;
825 base = task->thread.gs;
826 ret = put_user(base, (unsigned long __user *)addr);
838 long sys_arch_prctl(int code, unsigned long addr)
840 return do_arch_prctl(current, code, addr);
843 unsigned long arch_align_stack(unsigned long sp)
845 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
846 sp -= get_random_int() % 8192;
850 unsigned long arch_randomize_brk(struct mm_struct *mm)
852 unsigned long range_end = mm->brk + 0x02000000;
853 return randomize_range(mm->brk, range_end, 0) ? : mm->brk;