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 */
102 __get_cpu_var(cpu_state
) = CPU_DEAD
;
105 * With physical CPU hotplug, we should halt the cpu
112 static inline void play_dead(void)
116 #endif /* CONFIG_HOTPLUG_CPU */
119 * The idle thread. There's no useful work to be
120 * done, so just try to conserve power and have a
121 * low exit latency (ie sit in a loop waiting for
122 * somebody to say that they'd like to reschedule)
126 int cpu
= smp_processor_id();
128 current_thread_info()->status
|= TS_POLLING
;
130 /* endless idle loop with no priority at all */
132 tick_nohz_stop_sched_tick();
133 while (!need_resched()) {
138 if (rcu_pending(cpu
))
139 rcu_check_callbacks(cpu
, 0);
141 if (cpu_is_offline(cpu
))
145 __get_cpu_var(irq_stat
).idle_timestamp
= jiffies
;
146 /* Don't trace irqs off for idle */
147 stop_critical_timings();
149 start_critical_timings();
151 tick_nohz_restart_sched_tick();
152 preempt_enable_no_resched();
158 void __show_registers(struct pt_regs
*regs
, int all
)
160 unsigned long cr0
= 0L, cr2
= 0L, cr3
= 0L, cr4
= 0L;
161 unsigned long d0
, d1
, d2
, d3
, d6
, d7
;
163 unsigned short ss
, gs
;
166 if (user_mode_vm(regs
)) {
168 ss
= regs
->ss
& 0xffff;
171 sp
= (unsigned long) (®s
->sp
);
178 board
= dmi_get_system_info(DMI_PRODUCT_NAME
);
181 printk("Pid: %d, comm: %s %s (%s %.*s) %s\n",
182 task_pid_nr(current
), current
->comm
,
183 print_tainted(), init_utsname()->release
,
184 (int)strcspn(init_utsname()->version
, " "),
185 init_utsname()->version
, board
);
187 printk("EIP: %04x:[<%08lx>] EFLAGS: %08lx CPU: %d\n",
188 (u16
)regs
->cs
, regs
->ip
, regs
->flags
,
190 print_symbol("EIP is at %s\n", regs
->ip
);
192 printk("EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
193 regs
->ax
, regs
->bx
, regs
->cx
, regs
->dx
);
194 printk("ESI: %08lx EDI: %08lx EBP: %08lx ESP: %08lx\n",
195 regs
->si
, regs
->di
, regs
->bp
, sp
);
196 printk(" DS: %04x ES: %04x FS: %04x GS: %04x SS: %04x\n",
197 (u16
)regs
->ds
, (u16
)regs
->es
, (u16
)regs
->fs
, gs
, ss
);
205 cr4
= read_cr4_safe();
206 printk("CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n",
213 printk("DR0: %08lx DR1: %08lx DR2: %08lx DR3: %08lx\n",
218 printk("DR6: %08lx DR7: %08lx\n",
222 void show_regs(struct pt_regs
*regs
)
224 __show_registers(regs
, 1);
225 show_trace(NULL
, regs
, ®s
->sp
, regs
->bp
);
229 * This gets run with %bx containing the
230 * function to call, and %dx containing
233 extern void kernel_thread_helper(void);
236 * Create a kernel thread
238 int kernel_thread(int (*fn
)(void *), void * arg
, unsigned long flags
)
242 memset(®s
, 0, sizeof(regs
));
244 regs
.bx
= (unsigned long) fn
;
245 regs
.dx
= (unsigned long) arg
;
249 regs
.fs
= __KERNEL_PERCPU
;
251 regs
.ip
= (unsigned long) kernel_thread_helper
;
252 regs
.cs
= __KERNEL_CS
| get_kernel_rpl();
253 regs
.flags
= X86_EFLAGS_IF
| X86_EFLAGS_SF
| X86_EFLAGS_PF
| 0x2;
255 /* Ok, create the new process.. */
256 return do_fork(flags
| CLONE_VM
| CLONE_UNTRACED
, 0, ®s
, 0, NULL
, NULL
);
258 EXPORT_SYMBOL(kernel_thread
);
261 * Free current thread data structures etc..
263 void exit_thread(void)
265 /* The process may have allocated an io port bitmap... nuke it. */
266 if (unlikely(test_thread_flag(TIF_IO_BITMAP
))) {
267 struct task_struct
*tsk
= current
;
268 struct thread_struct
*t
= &tsk
->thread
;
270 struct tss_struct
*tss
= &per_cpu(init_tss
, cpu
);
272 kfree(t
->io_bitmap_ptr
);
273 t
->io_bitmap_ptr
= NULL
;
274 clear_thread_flag(TIF_IO_BITMAP
);
276 * Careful, clear this in the TSS too:
278 memset(tss
->io_bitmap
, 0xff, tss
->io_bitmap_max
);
279 t
->io_bitmap_max
= 0;
280 tss
->io_bitmap_owner
= NULL
;
281 tss
->io_bitmap_max
= 0;
282 tss
->x86_tss
.io_bitmap_base
= INVALID_IO_BITMAP_OFFSET
;
287 void flush_thread(void)
289 struct task_struct
*tsk
= current
;
291 tsk
->thread
.debugreg0
= 0;
292 tsk
->thread
.debugreg1
= 0;
293 tsk
->thread
.debugreg2
= 0;
294 tsk
->thread
.debugreg3
= 0;
295 tsk
->thread
.debugreg6
= 0;
296 tsk
->thread
.debugreg7
= 0;
297 memset(tsk
->thread
.tls_array
, 0, sizeof(tsk
->thread
.tls_array
));
298 clear_tsk_thread_flag(tsk
, TIF_DEBUG
);
300 * Forget coprocessor state..
302 tsk
->fpu_counter
= 0;
307 void release_thread(struct task_struct
*dead_task
)
309 BUG_ON(dead_task
->mm
);
310 release_vm86_irqs(dead_task
);
314 * This gets called before we allocate a new thread and copy
315 * the current task into it.
317 void prepare_to_copy(struct task_struct
*tsk
)
322 int copy_thread(int nr
, unsigned long clone_flags
, unsigned long sp
,
323 unsigned long unused
,
324 struct task_struct
* p
, struct pt_regs
* regs
)
326 struct pt_regs
* childregs
;
327 struct task_struct
*tsk
;
330 childregs
= task_pt_regs(p
);
335 p
->thread
.sp
= (unsigned long) childregs
;
336 p
->thread
.sp0
= (unsigned long) (childregs
+1);
338 p
->thread
.ip
= (unsigned long) ret_from_fork
;
340 savesegment(gs
, p
->thread
.gs
);
343 if (unlikely(test_tsk_thread_flag(tsk
, TIF_IO_BITMAP
))) {
344 p
->thread
.io_bitmap_ptr
= kmemdup(tsk
->thread
.io_bitmap_ptr
,
345 IO_BITMAP_BYTES
, GFP_KERNEL
);
346 if (!p
->thread
.io_bitmap_ptr
) {
347 p
->thread
.io_bitmap_max
= 0;
350 set_tsk_thread_flag(p
, TIF_IO_BITMAP
);
356 * Set a new TLS for the child thread?
358 if (clone_flags
& CLONE_SETTLS
)
359 err
= do_set_thread_area(p
, -1,
360 (struct user_desc __user
*)childregs
->si
, 0);
362 if (err
&& p
->thread
.io_bitmap_ptr
) {
363 kfree(p
->thread
.io_bitmap_ptr
);
364 p
->thread
.io_bitmap_max
= 0;
370 start_thread(struct pt_regs
*regs
, unsigned long new_ip
, unsigned long new_sp
)
372 __asm__("movl %0, %%gs" :: "r"(0));
375 regs
->ds
= __USER_DS
;
376 regs
->es
= __USER_DS
;
377 regs
->ss
= __USER_DS
;
378 regs
->cs
= __USER_CS
;
382 * Free the old FP and other extended state
384 free_thread_xstate(current
);
386 EXPORT_SYMBOL_GPL(start_thread
);
388 static void hard_disable_TSC(void)
390 write_cr4(read_cr4() | X86_CR4_TSD
);
393 void disable_TSC(void)
396 if (!test_and_set_thread_flag(TIF_NOTSC
))
398 * Must flip the CPU state synchronously with
399 * TIF_NOTSC in the current running context.
405 static void hard_enable_TSC(void)
407 write_cr4(read_cr4() & ~X86_CR4_TSD
);
410 static void enable_TSC(void)
413 if (test_and_clear_thread_flag(TIF_NOTSC
))
415 * Must flip the CPU state synchronously with
416 * TIF_NOTSC in the current running context.
422 int get_tsc_mode(unsigned long adr
)
426 if (test_thread_flag(TIF_NOTSC
))
427 val
= PR_TSC_SIGSEGV
;
431 return put_user(val
, (unsigned int __user
*)adr
);
434 int set_tsc_mode(unsigned int val
)
436 if (val
== PR_TSC_SIGSEGV
)
438 else if (val
== PR_TSC_ENABLE
)
447 __switch_to_xtra(struct task_struct
*prev_p
, struct task_struct
*next_p
,
448 struct tss_struct
*tss
)
450 struct thread_struct
*prev
, *next
;
451 unsigned long debugctl
;
453 prev
= &prev_p
->thread
;
454 next
= &next_p
->thread
;
456 debugctl
= prev
->debugctlmsr
;
457 if (next
->ds_area_msr
!= prev
->ds_area_msr
) {
458 /* we clear debugctl to make sure DS
459 * is not in use when we change it */
461 update_debugctlmsr(0);
462 wrmsr(MSR_IA32_DS_AREA
, next
->ds_area_msr
, 0);
465 if (next
->debugctlmsr
!= debugctl
)
466 update_debugctlmsr(next
->debugctlmsr
);
468 if (test_tsk_thread_flag(next_p
, TIF_DEBUG
)) {
469 set_debugreg(next
->debugreg0
, 0);
470 set_debugreg(next
->debugreg1
, 1);
471 set_debugreg(next
->debugreg2
, 2);
472 set_debugreg(next
->debugreg3
, 3);
474 set_debugreg(next
->debugreg6
, 6);
475 set_debugreg(next
->debugreg7
, 7);
478 if (test_tsk_thread_flag(prev_p
, TIF_NOTSC
) ^
479 test_tsk_thread_flag(next_p
, TIF_NOTSC
)) {
480 /* prev and next are different */
481 if (test_tsk_thread_flag(next_p
, TIF_NOTSC
))
488 if (test_tsk_thread_flag(prev_p
, TIF_BTS_TRACE_TS
))
489 ptrace_bts_take_timestamp(prev_p
, BTS_TASK_DEPARTS
);
491 if (test_tsk_thread_flag(next_p
, TIF_BTS_TRACE_TS
))
492 ptrace_bts_take_timestamp(next_p
, BTS_TASK_ARRIVES
);
496 if (!test_tsk_thread_flag(next_p
, TIF_IO_BITMAP
)) {
498 * Disable the bitmap via an invalid offset. We still cache
499 * the previous bitmap owner and the IO bitmap contents:
501 tss
->x86_tss
.io_bitmap_base
= INVALID_IO_BITMAP_OFFSET
;
505 if (likely(next
== tss
->io_bitmap_owner
)) {
507 * Previous owner of the bitmap (hence the bitmap content)
508 * matches the next task, we dont have to do anything but
509 * to set a valid offset in the TSS:
511 tss
->x86_tss
.io_bitmap_base
= IO_BITMAP_OFFSET
;
515 * Lazy TSS's I/O bitmap copy. We set an invalid offset here
516 * and we let the task to get a GPF in case an I/O instruction
517 * is performed. The handler of the GPF will verify that the
518 * faulting task has a valid I/O bitmap and, it true, does the
519 * real copy and restart the instruction. This will save us
520 * redundant copies when the currently switched task does not
521 * perform any I/O during its timeslice.
523 tss
->x86_tss
.io_bitmap_base
= INVALID_IO_BITMAP_OFFSET_LAZY
;
527 * switch_to(x,yn) should switch tasks from x to y.
529 * We fsave/fwait so that an exception goes off at the right time
530 * (as a call from the fsave or fwait in effect) rather than to
531 * the wrong process. Lazy FP saving no longer makes any sense
532 * with modern CPU's, and this simplifies a lot of things (SMP
533 * and UP become the same).
535 * NOTE! We used to use the x86 hardware context switching. The
536 * reason for not using it any more becomes apparent when you
537 * try to recover gracefully from saved state that is no longer
538 * valid (stale segment register values in particular). With the
539 * hardware task-switch, there is no way to fix up bad state in
540 * a reasonable manner.
542 * The fact that Intel documents the hardware task-switching to
543 * be slow is a fairly red herring - this code is not noticeably
544 * faster. However, there _is_ some room for improvement here,
545 * so the performance issues may eventually be a valid point.
546 * More important, however, is the fact that this allows us much
549 * The return value (in %ax) will be the "prev" task after
550 * the task-switch, and shows up in ret_from_fork in entry.S,
553 struct task_struct
* __switch_to(struct task_struct
*prev_p
, struct task_struct
*next_p
)
555 struct thread_struct
*prev
= &prev_p
->thread
,
556 *next
= &next_p
->thread
;
557 int cpu
= smp_processor_id();
558 struct tss_struct
*tss
= &per_cpu(init_tss
, cpu
);
560 /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
562 __unlazy_fpu(prev_p
);
565 /* we're going to use this soon, after a few expensive things */
566 if (next_p
->fpu_counter
> 5)
567 prefetch(next
->xstate
);
575 * Save away %gs. No need to save %fs, as it was saved on the
576 * stack on entry. No need to save %es and %ds, as those are
577 * always kernel segments while inside the kernel. Doing this
578 * before setting the new TLS descriptors avoids the situation
579 * where we temporarily have non-reloadable segments in %fs
580 * and %gs. This could be an issue if the NMI handler ever
581 * used %fs or %gs (it does not today), or if the kernel is
582 * running inside of a hypervisor layer.
584 savesegment(gs
, prev
->gs
);
587 * Load the per-thread Thread-Local Storage descriptor.
592 * Restore IOPL if needed. In normal use, the flags restore
593 * in the switch assembly will handle this. But if the kernel
594 * is running virtualized at a non-zero CPL, the popf will
595 * not restore flags, so it must be done in a separate step.
597 if (get_kernel_rpl() && unlikely(prev
->iopl
!= next
->iopl
))
598 set_iopl_mask(next
->iopl
);
601 * Now maybe handle debug registers and/or IO bitmaps
603 if (unlikely(task_thread_info(prev_p
)->flags
& _TIF_WORK_CTXSW_PREV
||
604 task_thread_info(next_p
)->flags
& _TIF_WORK_CTXSW_NEXT
))
605 __switch_to_xtra(prev_p
, next_p
, tss
);
608 * Leave lazy mode, flushing any hypercalls made here.
609 * This must be done before restoring TLS segments so
610 * the GDT and LDT are properly updated, and must be
611 * done before math_state_restore, so the TS bit is up
614 arch_leave_lazy_cpu_mode();
616 /* If the task has used fpu the last 5 timeslices, just do a full
617 * restore of the math state immediately to avoid the trap; the
618 * chances of needing FPU soon are obviously high now
620 * tsk_used_math() checks prevent calling math_state_restore(),
621 * which can sleep in the case of !tsk_used_math()
623 if (tsk_used_math(next_p
) && next_p
->fpu_counter
> 5)
624 math_state_restore();
627 * Restore %gs if needed (which is common)
629 if (prev
->gs
| next
->gs
)
630 loadsegment(gs
, next
->gs
);
632 x86_write_percpu(current_task
, next_p
);
637 asmlinkage
int sys_fork(struct pt_regs regs
)
639 return do_fork(SIGCHLD
, regs
.sp
, ®s
, 0, NULL
, NULL
);
642 asmlinkage
int sys_clone(struct pt_regs regs
)
644 unsigned long clone_flags
;
646 int __user
*parent_tidptr
, *child_tidptr
;
648 clone_flags
= regs
.bx
;
650 parent_tidptr
= (int __user
*)regs
.dx
;
651 child_tidptr
= (int __user
*)regs
.di
;
654 return do_fork(clone_flags
, newsp
, ®s
, 0, parent_tidptr
, child_tidptr
);
658 * This is trivial, and on the face of it looks like it
659 * could equally well be done in user mode.
661 * Not so, for quite unobvious reasons - register pressure.
662 * In user mode vfork() cannot have a stack frame, and if
663 * done by calling the "clone()" system call directly, you
664 * do not have enough call-clobbered registers to hold all
665 * the information you need.
667 asmlinkage
int sys_vfork(struct pt_regs regs
)
669 return do_fork(CLONE_VFORK
| CLONE_VM
| SIGCHLD
, regs
.sp
, ®s
, 0, NULL
, NULL
);
673 * sys_execve() executes a new program.
675 asmlinkage
int sys_execve(struct pt_regs regs
)
680 filename
= getname((char __user
*) regs
.bx
);
681 error
= PTR_ERR(filename
);
682 if (IS_ERR(filename
))
684 error
= do_execve(filename
,
685 (char __user
* __user
*) regs
.cx
,
686 (char __user
* __user
*) regs
.dx
,
689 /* Make sure we don't return using sysenter.. */
690 set_thread_flag(TIF_IRET
);
697 #define top_esp (THREAD_SIZE - sizeof(unsigned long))
698 #define top_ebp (THREAD_SIZE - 2*sizeof(unsigned long))
700 unsigned long get_wchan(struct task_struct
*p
)
702 unsigned long bp
, sp
, ip
;
703 unsigned long stack_page
;
705 if (!p
|| p
== current
|| p
->state
== TASK_RUNNING
)
707 stack_page
= (unsigned long)task_stack_page(p
);
709 if (!stack_page
|| sp
< stack_page
|| sp
> top_esp
+stack_page
)
711 /* include/asm-i386/system.h:switch_to() pushes bp last. */
712 bp
= *(unsigned long *) sp
;
714 if (bp
< stack_page
|| bp
> top_ebp
+stack_page
)
716 ip
= *(unsigned long *) (bp
+4);
717 if (!in_sched_functions(ip
))
719 bp
= *(unsigned long *) bp
;
720 } while (count
++ < 16);
724 unsigned long arch_align_stack(unsigned long sp
)
726 if (!(current
->personality
& ADDR_NO_RANDOMIZE
) && randomize_va_space
)
727 sp
-= get_random_int() % 8192;
731 unsigned long arch_randomize_brk(struct mm_struct
*mm
)
733 unsigned long range_end
= mm
->brk
+ 0x02000000;
734 return randomize_range(mm
->brk
, range_end
, 0) ? : mm
->brk
;