2 * Copyright (C) 1995 Linus Torvalds
4 * Pentium III FXSR, SSE support
5 * Gareth Hughes <gareth@valinux.com>, May 2000
9 * This file handles the architecture-dependent parts of process handling..
14 #include <linux/cpu.h>
15 #include <linux/errno.h>
16 #include <linux/sched.h>
18 #include <linux/kernel.h>
20 #include <linux/elfcore.h>
21 #include <linux/smp.h>
22 #include <linux/stddef.h>
23 #include <linux/slab.h>
24 #include <linux/vmalloc.h>
25 #include <linux/user.h>
26 #include <linux/interrupt.h>
27 #include <linux/utsname.h>
28 #include <linux/delay.h>
29 #include <linux/reboot.h>
30 #include <linux/init.h>
31 #include <linux/mc146818rtc.h>
32 #include <linux/module.h>
33 #include <linux/kallsyms.h>
34 #include <linux/ptrace.h>
35 #include <linux/random.h>
36 #include <linux/personality.h>
37 #include <linux/tick.h>
38 #include <linux/percpu.h>
39 #include <linux/prctl.h>
40 #include <linux/dmi.h>
42 #include <asm/uaccess.h>
43 #include <asm/pgtable.h>
44 #include <asm/system.h>
47 #include <asm/processor.h>
50 #ifdef CONFIG_MATH_EMULATION
51 #include <asm/math_emu.h>
54 #include <linux/err.h>
56 #include <asm/tlbflush.h>
58 #include <asm/kdebug.h>
60 asmlinkage
void ret_from_fork(void) __asm__("ret_from_fork");
62 DEFINE_PER_CPU(struct task_struct
*, current_task
) = &init_task
;
63 EXPORT_PER_CPU_SYMBOL(current_task
);
65 DEFINE_PER_CPU(int, cpu_number
);
66 EXPORT_PER_CPU_SYMBOL(cpu_number
);
69 * Return saved PC of a blocked thread.
71 unsigned long thread_saved_pc(struct task_struct
*tsk
)
73 return ((unsigned long *)tsk
->thread
.sp
)[3];
76 #ifdef CONFIG_HOTPLUG_CPU
79 static void cpu_exit_clear(void)
81 int cpu
= raw_smp_processor_id();
88 cpu_clear(cpu
, cpu_callout_map
);
89 cpu_clear(cpu
, cpu_callin_map
);
94 /* We don't actually take CPU down, just spin without interrupts. */
95 static inline void play_dead(void)
97 /* This must be done before dead CPU ack */
101 __get_cpu_var(cpu_state
) = CPU_DEAD
;
104 * With physical CPU hotplug, we should halt the cpu
107 /* mask all interrupts, flush any and all caches, and halt */
111 static inline void play_dead(void)
115 #endif /* CONFIG_HOTPLUG_CPU */
118 * The idle thread. There's no useful work to be
119 * done, so just try to conserve power and have a
120 * low exit latency (ie sit in a loop waiting for
121 * somebody to say that they'd like to reschedule)
125 int cpu
= smp_processor_id();
127 current_thread_info()->status
|= TS_POLLING
;
129 /* endless idle loop with no priority at all */
131 tick_nohz_stop_sched_tick(1);
132 while (!need_resched()) {
137 if (rcu_pending(cpu
))
138 rcu_check_callbacks(cpu
, 0);
140 if (cpu_is_offline(cpu
))
144 __get_cpu_var(irq_stat
).idle_timestamp
= jiffies
;
145 /* Don't trace irqs off for idle */
146 stop_critical_timings();
148 start_critical_timings();
150 tick_nohz_restart_sched_tick();
151 preempt_enable_no_resched();
157 void __show_registers(struct pt_regs
*regs
, int all
)
159 unsigned long cr0
= 0L, cr2
= 0L, cr3
= 0L, cr4
= 0L;
160 unsigned long d0
, d1
, d2
, d3
, d6
, d7
;
162 unsigned short ss
, gs
;
165 if (user_mode_vm(regs
)) {
167 ss
= regs
->ss
& 0xffff;
170 sp
= (unsigned long) (®s
->sp
);
177 board
= dmi_get_system_info(DMI_PRODUCT_NAME
);
180 printk("Pid: %d, comm: %s %s (%s %.*s) %s\n",
181 task_pid_nr(current
), current
->comm
,
182 print_tainted(), init_utsname()->release
,
183 (int)strcspn(init_utsname()->version
, " "),
184 init_utsname()->version
, board
);
186 printk("EIP: %04x:[<%08lx>] EFLAGS: %08lx CPU: %d\n",
187 (u16
)regs
->cs
, regs
->ip
, regs
->flags
,
189 print_symbol("EIP is at %s\n", regs
->ip
);
191 printk("EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
192 regs
->ax
, regs
->bx
, regs
->cx
, regs
->dx
);
193 printk("ESI: %08lx EDI: %08lx EBP: %08lx ESP: %08lx\n",
194 regs
->si
, regs
->di
, regs
->bp
, sp
);
195 printk(" DS: %04x ES: %04x FS: %04x GS: %04x SS: %04x\n",
196 (u16
)regs
->ds
, (u16
)regs
->es
, (u16
)regs
->fs
, gs
, ss
);
204 cr4
= read_cr4_safe();
205 printk("CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n",
212 printk("DR0: %08lx DR1: %08lx DR2: %08lx DR3: %08lx\n",
217 printk("DR6: %08lx DR7: %08lx\n",
221 void show_regs(struct pt_regs
*regs
)
223 __show_registers(regs
, 1);
224 show_trace(NULL
, regs
, ®s
->sp
, regs
->bp
);
228 * This gets run with %bx containing the
229 * function to call, and %dx containing
232 extern void kernel_thread_helper(void);
235 * Create a kernel thread
237 int kernel_thread(int (*fn
)(void *), void * arg
, unsigned long flags
)
241 memset(®s
, 0, sizeof(regs
));
243 regs
.bx
= (unsigned long) fn
;
244 regs
.dx
= (unsigned long) arg
;
248 regs
.fs
= __KERNEL_PERCPU
;
250 regs
.ip
= (unsigned long) kernel_thread_helper
;
251 regs
.cs
= __KERNEL_CS
| get_kernel_rpl();
252 regs
.flags
= X86_EFLAGS_IF
| X86_EFLAGS_SF
| X86_EFLAGS_PF
| 0x2;
254 /* Ok, create the new process.. */
255 return do_fork(flags
| CLONE_VM
| CLONE_UNTRACED
, 0, ®s
, 0, NULL
, NULL
);
257 EXPORT_SYMBOL(kernel_thread
);
260 * Free current thread data structures etc..
262 void exit_thread(void)
264 /* The process may have allocated an io port bitmap... nuke it. */
265 if (unlikely(test_thread_flag(TIF_IO_BITMAP
))) {
266 struct task_struct
*tsk
= current
;
267 struct thread_struct
*t
= &tsk
->thread
;
269 struct tss_struct
*tss
= &per_cpu(init_tss
, cpu
);
271 kfree(t
->io_bitmap_ptr
);
272 t
->io_bitmap_ptr
= NULL
;
273 clear_thread_flag(TIF_IO_BITMAP
);
275 * Careful, clear this in the TSS too:
277 memset(tss
->io_bitmap
, 0xff, tss
->io_bitmap_max
);
278 t
->io_bitmap_max
= 0;
279 tss
->io_bitmap_owner
= NULL
;
280 tss
->io_bitmap_max
= 0;
281 tss
->x86_tss
.io_bitmap_base
= INVALID_IO_BITMAP_OFFSET
;
286 void flush_thread(void)
288 struct task_struct
*tsk
= current
;
290 tsk
->thread
.debugreg0
= 0;
291 tsk
->thread
.debugreg1
= 0;
292 tsk
->thread
.debugreg2
= 0;
293 tsk
->thread
.debugreg3
= 0;
294 tsk
->thread
.debugreg6
= 0;
295 tsk
->thread
.debugreg7
= 0;
296 memset(tsk
->thread
.tls_array
, 0, sizeof(tsk
->thread
.tls_array
));
297 clear_tsk_thread_flag(tsk
, TIF_DEBUG
);
299 * Forget coprocessor state..
301 tsk
->fpu_counter
= 0;
306 void release_thread(struct task_struct
*dead_task
)
308 BUG_ON(dead_task
->mm
);
309 release_vm86_irqs(dead_task
);
313 * This gets called before we allocate a new thread and copy
314 * the current task into it.
316 void prepare_to_copy(struct task_struct
*tsk
)
321 int copy_thread(int nr
, unsigned long clone_flags
, unsigned long sp
,
322 unsigned long unused
,
323 struct task_struct
* p
, struct pt_regs
* regs
)
325 struct pt_regs
* childregs
;
326 struct task_struct
*tsk
;
329 childregs
= task_pt_regs(p
);
334 p
->thread
.sp
= (unsigned long) childregs
;
335 p
->thread
.sp0
= (unsigned long) (childregs
+1);
337 p
->thread
.ip
= (unsigned long) ret_from_fork
;
339 savesegment(gs
, p
->thread
.gs
);
342 if (unlikely(test_tsk_thread_flag(tsk
, TIF_IO_BITMAP
))) {
343 p
->thread
.io_bitmap_ptr
= kmemdup(tsk
->thread
.io_bitmap_ptr
,
344 IO_BITMAP_BYTES
, GFP_KERNEL
);
345 if (!p
->thread
.io_bitmap_ptr
) {
346 p
->thread
.io_bitmap_max
= 0;
349 set_tsk_thread_flag(p
, TIF_IO_BITMAP
);
355 * Set a new TLS for the child thread?
357 if (clone_flags
& CLONE_SETTLS
)
358 err
= do_set_thread_area(p
, -1,
359 (struct user_desc __user
*)childregs
->si
, 0);
361 if (err
&& p
->thread
.io_bitmap_ptr
) {
362 kfree(p
->thread
.io_bitmap_ptr
);
363 p
->thread
.io_bitmap_max
= 0;
369 start_thread(struct pt_regs
*regs
, unsigned long new_ip
, unsigned long new_sp
)
371 __asm__("movl %0, %%gs" :: "r"(0));
374 regs
->ds
= __USER_DS
;
375 regs
->es
= __USER_DS
;
376 regs
->ss
= __USER_DS
;
377 regs
->cs
= __USER_CS
;
381 * Free the old FP and other extended state
383 free_thread_xstate(current
);
385 EXPORT_SYMBOL_GPL(start_thread
);
387 static void hard_disable_TSC(void)
389 write_cr4(read_cr4() | X86_CR4_TSD
);
392 void disable_TSC(void)
395 if (!test_and_set_thread_flag(TIF_NOTSC
))
397 * Must flip the CPU state synchronously with
398 * TIF_NOTSC in the current running context.
404 static void hard_enable_TSC(void)
406 write_cr4(read_cr4() & ~X86_CR4_TSD
);
409 static void enable_TSC(void)
412 if (test_and_clear_thread_flag(TIF_NOTSC
))
414 * Must flip the CPU state synchronously with
415 * TIF_NOTSC in the current running context.
421 int get_tsc_mode(unsigned long adr
)
425 if (test_thread_flag(TIF_NOTSC
))
426 val
= PR_TSC_SIGSEGV
;
430 return put_user(val
, (unsigned int __user
*)adr
);
433 int set_tsc_mode(unsigned int val
)
435 if (val
== PR_TSC_SIGSEGV
)
437 else if (val
== PR_TSC_ENABLE
)
446 __switch_to_xtra(struct task_struct
*prev_p
, struct task_struct
*next_p
,
447 struct tss_struct
*tss
)
449 struct thread_struct
*prev
, *next
;
450 unsigned long debugctl
;
452 prev
= &prev_p
->thread
;
453 next
= &next_p
->thread
;
455 debugctl
= prev
->debugctlmsr
;
456 if (next
->ds_area_msr
!= prev
->ds_area_msr
) {
457 /* we clear debugctl to make sure DS
458 * is not in use when we change it */
460 update_debugctlmsr(0);
461 wrmsr(MSR_IA32_DS_AREA
, next
->ds_area_msr
, 0);
464 if (next
->debugctlmsr
!= debugctl
)
465 update_debugctlmsr(next
->debugctlmsr
);
467 if (test_tsk_thread_flag(next_p
, TIF_DEBUG
)) {
468 set_debugreg(next
->debugreg0
, 0);
469 set_debugreg(next
->debugreg1
, 1);
470 set_debugreg(next
->debugreg2
, 2);
471 set_debugreg(next
->debugreg3
, 3);
473 set_debugreg(next
->debugreg6
, 6);
474 set_debugreg(next
->debugreg7
, 7);
477 if (test_tsk_thread_flag(prev_p
, TIF_NOTSC
) ^
478 test_tsk_thread_flag(next_p
, TIF_NOTSC
)) {
479 /* prev and next are different */
480 if (test_tsk_thread_flag(next_p
, TIF_NOTSC
))
487 if (test_tsk_thread_flag(prev_p
, TIF_BTS_TRACE_TS
))
488 ptrace_bts_take_timestamp(prev_p
, BTS_TASK_DEPARTS
);
490 if (test_tsk_thread_flag(next_p
, TIF_BTS_TRACE_TS
))
491 ptrace_bts_take_timestamp(next_p
, BTS_TASK_ARRIVES
);
495 if (!test_tsk_thread_flag(next_p
, TIF_IO_BITMAP
)) {
497 * Disable the bitmap via an invalid offset. We still cache
498 * the previous bitmap owner and the IO bitmap contents:
500 tss
->x86_tss
.io_bitmap_base
= INVALID_IO_BITMAP_OFFSET
;
504 if (likely(next
== tss
->io_bitmap_owner
)) {
506 * Previous owner of the bitmap (hence the bitmap content)
507 * matches the next task, we dont have to do anything but
508 * to set a valid offset in the TSS:
510 tss
->x86_tss
.io_bitmap_base
= IO_BITMAP_OFFSET
;
514 * Lazy TSS's I/O bitmap copy. We set an invalid offset here
515 * and we let the task to get a GPF in case an I/O instruction
516 * is performed. The handler of the GPF will verify that the
517 * faulting task has a valid I/O bitmap and, it true, does the
518 * real copy and restart the instruction. This will save us
519 * redundant copies when the currently switched task does not
520 * perform any I/O during its timeslice.
522 tss
->x86_tss
.io_bitmap_base
= INVALID_IO_BITMAP_OFFSET_LAZY
;
526 * switch_to(x,yn) should switch tasks from x to y.
528 * We fsave/fwait so that an exception goes off at the right time
529 * (as a call from the fsave or fwait in effect) rather than to
530 * the wrong process. Lazy FP saving no longer makes any sense
531 * with modern CPU's, and this simplifies a lot of things (SMP
532 * and UP become the same).
534 * NOTE! We used to use the x86 hardware context switching. The
535 * reason for not using it any more becomes apparent when you
536 * try to recover gracefully from saved state that is no longer
537 * valid (stale segment register values in particular). With the
538 * hardware task-switch, there is no way to fix up bad state in
539 * a reasonable manner.
541 * The fact that Intel documents the hardware task-switching to
542 * be slow is a fairly red herring - this code is not noticeably
543 * faster. However, there _is_ some room for improvement here,
544 * so the performance issues may eventually be a valid point.
545 * More important, however, is the fact that this allows us much
548 * The return value (in %ax) will be the "prev" task after
549 * the task-switch, and shows up in ret_from_fork in entry.S,
552 struct task_struct
* __switch_to(struct task_struct
*prev_p
, struct task_struct
*next_p
)
554 struct thread_struct
*prev
= &prev_p
->thread
,
555 *next
= &next_p
->thread
;
556 int cpu
= smp_processor_id();
557 struct tss_struct
*tss
= &per_cpu(init_tss
, cpu
);
559 /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
561 __unlazy_fpu(prev_p
);
564 /* we're going to use this soon, after a few expensive things */
565 if (next_p
->fpu_counter
> 5)
566 prefetch(next
->xstate
);
574 * Save away %gs. No need to save %fs, as it was saved on the
575 * stack on entry. No need to save %es and %ds, as those are
576 * always kernel segments while inside the kernel. Doing this
577 * before setting the new TLS descriptors avoids the situation
578 * where we temporarily have non-reloadable segments in %fs
579 * and %gs. This could be an issue if the NMI handler ever
580 * used %fs or %gs (it does not today), or if the kernel is
581 * running inside of a hypervisor layer.
583 savesegment(gs
, prev
->gs
);
586 * Load the per-thread Thread-Local Storage descriptor.
591 * Restore IOPL if needed. In normal use, the flags restore
592 * in the switch assembly will handle this. But if the kernel
593 * is running virtualized at a non-zero CPL, the popf will
594 * not restore flags, so it must be done in a separate step.
596 if (get_kernel_rpl() && unlikely(prev
->iopl
!= next
->iopl
))
597 set_iopl_mask(next
->iopl
);
600 * Now maybe handle debug registers and/or IO bitmaps
602 if (unlikely(task_thread_info(prev_p
)->flags
& _TIF_WORK_CTXSW_PREV
||
603 task_thread_info(next_p
)->flags
& _TIF_WORK_CTXSW_NEXT
))
604 __switch_to_xtra(prev_p
, next_p
, tss
);
607 * Leave lazy mode, flushing any hypercalls made here.
608 * This must be done before restoring TLS segments so
609 * the GDT and LDT are properly updated, and must be
610 * done before math_state_restore, so the TS bit is up
613 arch_leave_lazy_cpu_mode();
615 /* If the task has used fpu the last 5 timeslices, just do a full
616 * restore of the math state immediately to avoid the trap; the
617 * chances of needing FPU soon are obviously high now
619 * tsk_used_math() checks prevent calling math_state_restore(),
620 * which can sleep in the case of !tsk_used_math()
622 if (tsk_used_math(next_p
) && next_p
->fpu_counter
> 5)
623 math_state_restore();
626 * Restore %gs if needed (which is common)
628 if (prev
->gs
| next
->gs
)
629 loadsegment(gs
, next
->gs
);
631 x86_write_percpu(current_task
, next_p
);
636 asmlinkage
int sys_fork(struct pt_regs regs
)
638 return do_fork(SIGCHLD
, regs
.sp
, ®s
, 0, NULL
, NULL
);
641 asmlinkage
int sys_clone(struct pt_regs regs
)
643 unsigned long clone_flags
;
645 int __user
*parent_tidptr
, *child_tidptr
;
647 clone_flags
= regs
.bx
;
649 parent_tidptr
= (int __user
*)regs
.dx
;
650 child_tidptr
= (int __user
*)regs
.di
;
653 return do_fork(clone_flags
, newsp
, ®s
, 0, parent_tidptr
, child_tidptr
);
657 * This is trivial, and on the face of it looks like it
658 * could equally well be done in user mode.
660 * Not so, for quite unobvious reasons - register pressure.
661 * In user mode vfork() cannot have a stack frame, and if
662 * done by calling the "clone()" system call directly, you
663 * do not have enough call-clobbered registers to hold all
664 * the information you need.
666 asmlinkage
int sys_vfork(struct pt_regs regs
)
668 return do_fork(CLONE_VFORK
| CLONE_VM
| SIGCHLD
, regs
.sp
, ®s
, 0, NULL
, NULL
);
672 * sys_execve() executes a new program.
674 asmlinkage
int sys_execve(struct pt_regs regs
)
679 filename
= getname((char __user
*) regs
.bx
);
680 error
= PTR_ERR(filename
);
681 if (IS_ERR(filename
))
683 error
= do_execve(filename
,
684 (char __user
* __user
*) regs
.cx
,
685 (char __user
* __user
*) regs
.dx
,
688 /* Make sure we don't return using sysenter.. */
689 set_thread_flag(TIF_IRET
);
696 #define top_esp (THREAD_SIZE - sizeof(unsigned long))
697 #define top_ebp (THREAD_SIZE - 2*sizeof(unsigned long))
699 unsigned long get_wchan(struct task_struct
*p
)
701 unsigned long bp
, sp
, ip
;
702 unsigned long stack_page
;
704 if (!p
|| p
== current
|| p
->state
== TASK_RUNNING
)
706 stack_page
= (unsigned long)task_stack_page(p
);
708 if (!stack_page
|| sp
< stack_page
|| sp
> top_esp
+stack_page
)
710 /* include/asm-i386/system.h:switch_to() pushes bp last. */
711 bp
= *(unsigned long *) sp
;
713 if (bp
< stack_page
|| bp
> top_ebp
+stack_page
)
715 ip
= *(unsigned long *) (bp
+4);
716 if (!in_sched_functions(ip
))
718 bp
= *(unsigned long *) bp
;
719 } while (count
++ < 16);
723 unsigned long arch_align_stack(unsigned long sp
)
725 if (!(current
->personality
& ADDR_NO_RANDOMIZE
) && randomize_va_space
)
726 sp
-= get_random_int() % 8192;
730 unsigned long arch_randomize_brk(struct mm_struct
*mm
)
732 unsigned long range_end
= mm
->brk
+ 0x02000000;
733 return randomize_range(mm
->brk
, range_end
, 0) ? : mm
->brk
;