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>
41 #include <asm/uaccess.h>
42 #include <asm/pgtable.h>
43 #include <asm/system.h>
45 #include <asm/processor.h>
47 #include <asm/mmu_context.h>
49 #include <asm/prctl.h>
51 #include <asm/proto.h>
55 asmlinkage extern void ret_from_fork(void);
57 unsigned long kernel_thread_flags = CLONE_VM | CLONE_UNTRACED;
59 static ATOMIC_NOTIFIER_HEAD(idle_notifier);
61 void idle_notifier_register(struct notifier_block *n)
63 atomic_notifier_chain_register(&idle_notifier, n);
69 atomic_notifier_call_chain(&idle_notifier, IDLE_START, NULL);
72 static void __exit_idle(void)
74 if (test_and_clear_bit_pda(0, isidle) == 0)
76 atomic_notifier_call_chain(&idle_notifier, IDLE_END, NULL);
79 /* Called from interrupts to signify idle end */
82 /* idle loop has pid 0 */
88 #ifdef CONFIG_HOTPLUG_CPU
89 DECLARE_PER_CPU(int, cpu_state);
92 /* We halt the CPU with physical CPU hotplug */
93 static inline void play_dead(void)
99 __get_cpu_var(cpu_state) = CPU_DEAD;
106 static inline void play_dead(void)
110 #endif /* CONFIG_HOTPLUG_CPU */
113 * The idle thread. There's no useful work to be
114 * done, so just try to conserve power and have a
115 * low exit latency (ie sit in a loop waiting for
116 * somebody to say that they'd like to reschedule)
120 current_thread_info()->status |= TS_POLLING;
121 /* endless idle loop with no priority at all */
123 tick_nohz_stop_sched_tick(1);
124 while (!need_resched()) {
128 if (cpu_is_offline(smp_processor_id()))
131 * Idle routines should keep interrupts disabled
132 * from here on, until they go to idle.
133 * Otherwise, idle callbacks can misfire.
137 /* Don't trace irqs off for idle */
138 stop_critical_timings();
140 start_critical_timings();
141 /* In many cases the interrupt that ended idle
142 has already called exit_idle. But some idle
143 loops can be woken up without interrupt. */
147 tick_nohz_restart_sched_tick();
148 preempt_enable_no_resched();
154 /* Prints also some state that isn't saved in the pt_regs */
155 void __show_regs(struct pt_regs * regs)
157 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L, fs, gs, shadowgs;
158 unsigned long d0, d1, d2, d3, d6, d7;
159 unsigned int fsindex, gsindex;
160 unsigned int ds, cs, es;
164 printk("Pid: %d, comm: %.20s %s %s %.*s\n",
165 current->pid, current->comm, print_tainted(),
166 init_utsname()->release,
167 (int)strcspn(init_utsname()->version, " "),
168 init_utsname()->version);
169 printk("RIP: %04lx:[<%016lx>] ", regs->cs & 0xffff, regs->ip);
170 printk_address(regs->ip, 1);
171 printk("RSP: %04lx:%016lx EFLAGS: %08lx\n", regs->ss, regs->sp,
173 printk("RAX: %016lx RBX: %016lx RCX: %016lx\n",
174 regs->ax, regs->bx, regs->cx);
175 printk("RDX: %016lx RSI: %016lx RDI: %016lx\n",
176 regs->dx, regs->si, regs->di);
177 printk("RBP: %016lx R08: %016lx R09: %016lx\n",
178 regs->bp, regs->r8, regs->r9);
179 printk("R10: %016lx R11: %016lx R12: %016lx\n",
180 regs->r10, regs->r11, regs->r12);
181 printk("R13: %016lx R14: %016lx R15: %016lx\n",
182 regs->r13, regs->r14, regs->r15);
184 asm("movl %%ds,%0" : "=r" (ds));
185 asm("movl %%cs,%0" : "=r" (cs));
186 asm("movl %%es,%0" : "=r" (es));
187 asm("movl %%fs,%0" : "=r" (fsindex));
188 asm("movl %%gs,%0" : "=r" (gsindex));
190 rdmsrl(MSR_FS_BASE, fs);
191 rdmsrl(MSR_GS_BASE, gs);
192 rdmsrl(MSR_KERNEL_GS_BASE, shadowgs);
199 printk("FS: %016lx(%04x) GS:%016lx(%04x) knlGS:%016lx\n",
200 fs,fsindex,gs,gsindex,shadowgs);
201 printk("CS: %04x DS: %04x ES: %04x CR0: %016lx\n", cs, ds, es, cr0);
202 printk("CR2: %016lx CR3: %016lx CR4: %016lx\n", cr2, cr3, cr4);
207 printk("DR0: %016lx DR1: %016lx DR2: %016lx\n", d0, d1, d2);
211 printk("DR3: %016lx DR6: %016lx DR7: %016lx\n", d3, d6, d7);
214 void show_regs(struct pt_regs *regs)
216 printk("CPU %d:", smp_processor_id());
218 show_trace(NULL, regs, (void *)(regs + 1), regs->bp);
222 * Free current thread data structures etc..
224 void exit_thread(void)
226 struct task_struct *me = current;
227 struct thread_struct *t = &me->thread;
229 if (me->thread.io_bitmap_ptr) {
230 struct tss_struct *tss = &per_cpu(init_tss, get_cpu());
232 kfree(t->io_bitmap_ptr);
233 t->io_bitmap_ptr = NULL;
234 clear_thread_flag(TIF_IO_BITMAP);
236 * Careful, clear this in the TSS too:
238 memset(tss->io_bitmap, 0xff, t->io_bitmap_max);
239 t->io_bitmap_max = 0;
244 void flush_thread(void)
246 struct task_struct *tsk = current;
248 if (test_tsk_thread_flag(tsk, TIF_ABI_PENDING)) {
249 clear_tsk_thread_flag(tsk, TIF_ABI_PENDING);
250 if (test_tsk_thread_flag(tsk, TIF_IA32)) {
251 clear_tsk_thread_flag(tsk, TIF_IA32);
253 set_tsk_thread_flag(tsk, TIF_IA32);
254 current_thread_info()->status |= TS_COMPAT;
257 clear_tsk_thread_flag(tsk, TIF_DEBUG);
259 tsk->thread.debugreg0 = 0;
260 tsk->thread.debugreg1 = 0;
261 tsk->thread.debugreg2 = 0;
262 tsk->thread.debugreg3 = 0;
263 tsk->thread.debugreg6 = 0;
264 tsk->thread.debugreg7 = 0;
265 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
267 * Forget coprocessor state..
269 tsk->fpu_counter = 0;
274 void release_thread(struct task_struct *dead_task)
277 if (dead_task->mm->context.size) {
278 printk("WARNING: dead process %8s still has LDT? <%p/%d>\n",
280 dead_task->mm->context.ldt,
281 dead_task->mm->context.size);
287 static inline void set_32bit_tls(struct task_struct *t, int tls, u32 addr)
289 struct user_desc ud = {
296 struct desc_struct *desc = t->thread.tls_array;
301 static inline u32 read_32bit_tls(struct task_struct *t, int tls)
303 return get_desc_base(&t->thread.tls_array[tls]);
307 * This gets called before we allocate a new thread and copy
308 * the current task into it.
310 void prepare_to_copy(struct task_struct *tsk)
315 int copy_thread(int nr, unsigned long clone_flags, unsigned long sp,
316 unsigned long unused,
317 struct task_struct * p, struct pt_regs * regs)
320 struct pt_regs * childregs;
321 struct task_struct *me = current;
323 childregs = ((struct pt_regs *)
324 (THREAD_SIZE + task_stack_page(p))) - 1;
330 childregs->sp = (unsigned long)childregs;
332 p->thread.sp = (unsigned long) childregs;
333 p->thread.sp0 = (unsigned long) (childregs+1);
334 p->thread.usersp = me->thread.usersp;
336 set_tsk_thread_flag(p, TIF_FORK);
338 p->thread.fs = me->thread.fs;
339 p->thread.gs = me->thread.gs;
341 savesegment(gs, p->thread.gsindex);
342 savesegment(fs, p->thread.fsindex);
343 savesegment(es, p->thread.es);
344 savesegment(ds, p->thread.ds);
346 if (unlikely(test_tsk_thread_flag(me, TIF_IO_BITMAP))) {
347 p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL);
348 if (!p->thread.io_bitmap_ptr) {
349 p->thread.io_bitmap_max = 0;
352 memcpy(p->thread.io_bitmap_ptr, me->thread.io_bitmap_ptr,
354 set_tsk_thread_flag(p, TIF_IO_BITMAP);
358 * Set a new TLS for the child thread?
360 if (clone_flags & CLONE_SETTLS) {
361 #ifdef CONFIG_IA32_EMULATION
362 if (test_thread_flag(TIF_IA32))
363 err = do_set_thread_area(p, -1,
364 (struct user_desc __user *)childregs->si, 0);
367 err = do_arch_prctl(p, ARCH_SET_FS, childregs->r8);
373 if (err && p->thread.io_bitmap_ptr) {
374 kfree(p->thread.io_bitmap_ptr);
375 p->thread.io_bitmap_max = 0;
381 start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp)
389 write_pda(oldrsp, new_sp);
390 regs->cs = __USER_CS;
391 regs->ss = __USER_DS;
395 * Free the old FP and other extended state
397 free_thread_xstate(current);
399 EXPORT_SYMBOL_GPL(start_thread);
401 static void hard_disable_TSC(void)
403 write_cr4(read_cr4() | X86_CR4_TSD);
406 void disable_TSC(void)
409 if (!test_and_set_thread_flag(TIF_NOTSC))
411 * Must flip the CPU state synchronously with
412 * TIF_NOTSC in the current running context.
418 static void hard_enable_TSC(void)
420 write_cr4(read_cr4() & ~X86_CR4_TSD);
423 static void enable_TSC(void)
426 if (test_and_clear_thread_flag(TIF_NOTSC))
428 * Must flip the CPU state synchronously with
429 * TIF_NOTSC in the current running context.
435 int get_tsc_mode(unsigned long adr)
439 if (test_thread_flag(TIF_NOTSC))
440 val = PR_TSC_SIGSEGV;
444 return put_user(val, (unsigned int __user *)adr);
447 int set_tsc_mode(unsigned int val)
449 if (val == PR_TSC_SIGSEGV)
451 else if (val == PR_TSC_ENABLE)
460 * This special macro can be used to load a debugging register
462 #define loaddebug(thread, r) set_debugreg(thread->debugreg ## r, r)
464 static inline void __switch_to_xtra(struct task_struct *prev_p,
465 struct task_struct *next_p,
466 struct tss_struct *tss)
468 struct thread_struct *prev, *next;
469 unsigned long debugctl;
471 prev = &prev_p->thread,
472 next = &next_p->thread;
474 debugctl = prev->debugctlmsr;
475 if (next->ds_area_msr != prev->ds_area_msr) {
476 /* we clear debugctl to make sure DS
477 * is not in use when we change it */
479 update_debugctlmsr(0);
480 wrmsrl(MSR_IA32_DS_AREA, next->ds_area_msr);
483 if (next->debugctlmsr != debugctl)
484 update_debugctlmsr(next->debugctlmsr);
486 if (test_tsk_thread_flag(next_p, TIF_DEBUG)) {
496 if (test_tsk_thread_flag(prev_p, TIF_NOTSC) ^
497 test_tsk_thread_flag(next_p, TIF_NOTSC)) {
498 /* prev and next are different */
499 if (test_tsk_thread_flag(next_p, TIF_NOTSC))
505 if (test_tsk_thread_flag(next_p, TIF_IO_BITMAP)) {
507 * Copy the relevant range of the IO bitmap.
508 * Normally this is 128 bytes or less:
510 memcpy(tss->io_bitmap, next->io_bitmap_ptr,
511 max(prev->io_bitmap_max, next->io_bitmap_max));
512 } else if (test_tsk_thread_flag(prev_p, TIF_IO_BITMAP)) {
514 * Clear any possible leftover bits:
516 memset(tss->io_bitmap, 0xff, prev->io_bitmap_max);
520 if (test_tsk_thread_flag(prev_p, TIF_BTS_TRACE_TS))
521 ptrace_bts_take_timestamp(prev_p, BTS_TASK_DEPARTS);
523 if (test_tsk_thread_flag(next_p, TIF_BTS_TRACE_TS))
524 ptrace_bts_take_timestamp(next_p, BTS_TASK_ARRIVES);
529 * switch_to(x,y) should switch tasks from x to y.
531 * This could still be optimized:
532 * - fold all the options into a flag word and test it with a single test.
533 * - could test fs/gs bitsliced
535 * Kprobes not supported here. Set the probe on schedule instead.
538 __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
540 struct thread_struct *prev = &prev_p->thread;
541 struct thread_struct *next = &next_p->thread;
542 int cpu = smp_processor_id();
543 struct tss_struct *tss = &per_cpu(init_tss, cpu);
544 unsigned fsindex, gsindex;
546 /* we're going to use this soon, after a few expensive things */
547 if (next_p->fpu_counter>5)
548 prefetch(next->xstate);
551 * Reload esp0, LDT and the page table pointer:
557 * This won't pick up thread selector changes, but I guess that is ok.
559 savesegment(es, prev->es);
560 if (unlikely(next->es | prev->es))
561 loadsegment(es, next->es);
563 savesegment(ds, prev->ds);
564 if (unlikely(next->ds | prev->ds))
565 loadsegment(ds, next->ds);
568 /* We must save %fs and %gs before load_TLS() because
569 * %fs and %gs may be cleared by load_TLS().
571 * (e.g. xen_load_tls())
573 savesegment(fs, fsindex);
574 savesegment(gs, gsindex);
579 * Leave lazy mode, flushing any hypercalls made here.
580 * This must be done before restoring TLS segments so
581 * the GDT and LDT are properly updated, and must be
582 * done before math_state_restore, so the TS bit is up
585 arch_leave_lazy_cpu_mode();
590 * Segment register != 0 always requires a reload. Also
591 * reload when it has changed. When prev process used 64bit
592 * base always reload to avoid an information leak.
594 if (unlikely(fsindex | next->fsindex | prev->fs)) {
595 loadsegment(fs, next->fsindex);
597 * Check if the user used a selector != 0; if yes
598 * clear 64bit base, since overloaded base is always
599 * mapped to the Null selector
604 /* when next process has a 64bit base use it */
606 wrmsrl(MSR_FS_BASE, next->fs);
607 prev->fsindex = fsindex;
609 if (unlikely(gsindex | next->gsindex | prev->gs)) {
610 load_gs_index(next->gsindex);
615 wrmsrl(MSR_KERNEL_GS_BASE, next->gs);
616 prev->gsindex = gsindex;
618 /* Must be after DS reload */
622 * Switch the PDA and FPU contexts.
624 prev->usersp = read_pda(oldrsp);
625 write_pda(oldrsp, next->usersp);
626 write_pda(pcurrent, next_p);
628 write_pda(kernelstack,
629 (unsigned long)task_stack_page(next_p) +
630 THREAD_SIZE - PDA_STACKOFFSET);
631 #ifdef CONFIG_CC_STACKPROTECTOR
632 write_pda(stack_canary, next_p->stack_canary);
634 * Build time only check to make sure the stack_canary is at
635 * offset 40 in the pda; this is a gcc ABI requirement
637 BUILD_BUG_ON(offsetof(struct x8664_pda, stack_canary) != 40);
641 * Now maybe reload the debug registers and handle I/O bitmaps
643 if (unlikely(task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT ||
644 task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV))
645 __switch_to_xtra(prev_p, next_p, tss);
647 /* If the task has used fpu the last 5 timeslices, just do a full
648 * restore of the math state immediately to avoid the trap; the
649 * chances of needing FPU soon are obviously high now
651 * tsk_used_math() checks prevent calling math_state_restore(),
652 * which can sleep in the case of !tsk_used_math()
654 if (tsk_used_math(next_p) && next_p->fpu_counter > 5)
655 math_state_restore();
660 * sys_execve() executes a new program.
663 long sys_execve(char __user *name, char __user * __user *argv,
664 char __user * __user *envp, struct pt_regs *regs)
669 filename = getname(name);
670 error = PTR_ERR(filename);
671 if (IS_ERR(filename))
673 error = do_execve(filename, argv, envp, regs);
678 void set_personality_64bit(void)
680 /* inherit personality from parent */
682 /* Make sure to be in 64bit mode */
683 clear_thread_flag(TIF_IA32);
685 /* TBD: overwrites user setup. Should have two bits.
686 But 64bit processes have always behaved this way,
687 so it's not too bad. The main problem is just that
688 32bit childs are affected again. */
689 current->personality &= ~READ_IMPLIES_EXEC;
692 asmlinkage long sys_fork(struct pt_regs *regs)
694 return do_fork(SIGCHLD, regs->sp, regs, 0, NULL, NULL);
698 sys_clone(unsigned long clone_flags, unsigned long newsp,
699 void __user *parent_tid, void __user *child_tid, struct pt_regs *regs)
703 return do_fork(clone_flags, newsp, regs, 0, parent_tid, child_tid);
707 * This is trivial, and on the face of it looks like it
708 * could equally well be done in user mode.
710 * Not so, for quite unobvious reasons - register pressure.
711 * In user mode vfork() cannot have a stack frame, and if
712 * done by calling the "clone()" system call directly, you
713 * do not have enough call-clobbered registers to hold all
714 * the information you need.
716 asmlinkage long sys_vfork(struct pt_regs *regs)
718 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs->sp, regs, 0,
722 unsigned long get_wchan(struct task_struct *p)
728 if (!p || p == current || p->state==TASK_RUNNING)
730 stack = (unsigned long)task_stack_page(p);
731 if (p->thread.sp < stack || p->thread.sp > stack+THREAD_SIZE)
733 fp = *(u64 *)(p->thread.sp);
735 if (fp < (unsigned long)stack ||
736 fp > (unsigned long)stack+THREAD_SIZE)
739 if (!in_sched_functions(ip))
742 } while (count++ < 16);
746 long do_arch_prctl(struct task_struct *task, int code, unsigned long addr)
749 int doit = task == current;
754 if (addr >= TASK_SIZE_OF(task))
757 /* handle small bases via the GDT because that's faster to
759 if (addr <= 0xffffffff) {
760 set_32bit_tls(task, GS_TLS, addr);
762 load_TLS(&task->thread, cpu);
763 load_gs_index(GS_TLS_SEL);
765 task->thread.gsindex = GS_TLS_SEL;
768 task->thread.gsindex = 0;
769 task->thread.gs = addr;
772 ret = checking_wrmsrl(MSR_KERNEL_GS_BASE, addr);
778 /* Not strictly needed for fs, but do it for symmetry
780 if (addr >= TASK_SIZE_OF(task))
783 /* handle small bases via the GDT because that's faster to
785 if (addr <= 0xffffffff) {
786 set_32bit_tls(task, FS_TLS, addr);
788 load_TLS(&task->thread, cpu);
789 loadsegment(fs, FS_TLS_SEL);
791 task->thread.fsindex = FS_TLS_SEL;
794 task->thread.fsindex = 0;
795 task->thread.fs = addr;
797 /* set the selector to 0 to not confuse
800 ret = checking_wrmsrl(MSR_FS_BASE, addr);
807 if (task->thread.fsindex == FS_TLS_SEL)
808 base = read_32bit_tls(task, FS_TLS);
810 rdmsrl(MSR_FS_BASE, base);
812 base = task->thread.fs;
813 ret = put_user(base, (unsigned long __user *)addr);
819 if (task->thread.gsindex == GS_TLS_SEL)
820 base = read_32bit_tls(task, GS_TLS);
822 savesegment(gs, gsindex);
824 rdmsrl(MSR_KERNEL_GS_BASE, base);
826 base = task->thread.gs;
829 base = task->thread.gs;
830 ret = put_user(base, (unsigned long __user *)addr);
842 long sys_arch_prctl(int code, unsigned long addr)
844 return do_arch_prctl(current, code, addr);
847 unsigned long arch_align_stack(unsigned long sp)
849 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
850 sp -= get_random_int() % 8192;
854 unsigned long arch_randomize_brk(struct mm_struct *mm)
856 unsigned long range_end = mm->brk + 0x02000000;
857 return randomize_range(mm->brk, range_end, 0) ? : mm->brk;