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/stackprotector.h>
15 #include <linux/cpu.h>
16 #include <linux/errno.h>
17 #include <linux/sched.h>
19 #include <linux/kernel.h>
21 #include <linux/elfcore.h>
22 #include <linux/smp.h>
23 #include <linux/stddef.h>
24 #include <linux/slab.h>
25 #include <linux/vmalloc.h>
26 #include <linux/user.h>
27 #include <linux/interrupt.h>
28 #include <linux/utsname.h>
29 #include <linux/delay.h>
30 #include <linux/reboot.h>
31 #include <linux/init.h>
32 #include <linux/mc146818rtc.h>
33 #include <linux/module.h>
34 #include <linux/kallsyms.h>
35 #include <linux/ptrace.h>
36 #include <linux/random.h>
37 #include <linux/personality.h>
38 #include <linux/tick.h>
39 #include <linux/percpu.h>
40 #include <linux/prctl.h>
41 #include <linux/dmi.h>
42 #include <linux/ftrace.h>
43 #include <linux/uaccess.h>
45 #include <linux/kdebug.h>
47 #include <asm/pgtable.h>
48 #include <asm/system.h>
50 #include <asm/processor.h>
53 #ifdef CONFIG_MATH_EMULATION
54 #include <asm/math_emu.h>
57 #include <linux/err.h>
59 #include <asm/tlbflush.h>
62 #include <asm/syscalls.h>
65 asmlinkage
void ret_from_fork(void) __asm__("ret_from_fork");
67 DEFINE_PER_CPU(struct task_struct
*, current_task
) = &init_task
;
68 EXPORT_PER_CPU_SYMBOL(current_task
);
71 * Return saved PC of a blocked thread.
73 unsigned long thread_saved_pc(struct task_struct
*tsk
)
75 return ((unsigned long *)tsk
->thread
.sp
)[3];
79 static inline void play_dead(void)
86 * The idle thread. There's no useful work to be
87 * done, so just try to conserve power and have a
88 * low exit latency (ie sit in a loop waiting for
89 * somebody to say that they'd like to reschedule)
93 int cpu
= smp_processor_id();
96 * If we're the non-boot CPU, nothing set the stack canary up
97 * for us. CPU0 already has it initialized but no harm in
98 * doing it again. This is a good place for updating it, as
99 * we wont ever return from this function (so the invalid
100 * canaries already on the stack wont ever trigger).
102 boot_init_stack_canary();
104 current_thread_info()->status
|= TS_POLLING
;
106 /* endless idle loop with no priority at all */
108 tick_nohz_stop_sched_tick(1);
109 while (!need_resched()) {
114 if (cpu_is_offline(cpu
))
118 /* Don't trace irqs off for idle */
119 stop_critical_timings();
121 start_critical_timings();
123 tick_nohz_restart_sched_tick();
124 preempt_enable_no_resched();
130 void __show_regs(struct pt_regs
*regs
, int all
)
132 unsigned long cr0
= 0L, cr2
= 0L, cr3
= 0L, cr4
= 0L;
133 unsigned long d0
, d1
, d2
, d3
, d6
, d7
;
135 unsigned short ss
, gs
;
138 if (user_mode_vm(regs
)) {
140 ss
= regs
->ss
& 0xffff;
141 gs
= get_user_gs(regs
);
143 sp
= (unsigned long) (®s
->sp
);
150 board
= dmi_get_system_info(DMI_PRODUCT_NAME
);
153 printk("Pid: %d, comm: %s %s (%s %.*s) %s\n",
154 task_pid_nr(current
), current
->comm
,
155 print_tainted(), init_utsname()->release
,
156 (int)strcspn(init_utsname()->version
, " "),
157 init_utsname()->version
, board
);
159 printk("EIP: %04x:[<%08lx>] EFLAGS: %08lx CPU: %d\n",
160 (u16
)regs
->cs
, regs
->ip
, regs
->flags
,
162 print_symbol("EIP is at %s\n", regs
->ip
);
164 printk("EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
165 regs
->ax
, regs
->bx
, regs
->cx
, regs
->dx
);
166 printk("ESI: %08lx EDI: %08lx EBP: %08lx ESP: %08lx\n",
167 regs
->si
, regs
->di
, regs
->bp
, sp
);
168 printk(" DS: %04x ES: %04x FS: %04x GS: %04x SS: %04x\n",
169 (u16
)regs
->ds
, (u16
)regs
->es
, (u16
)regs
->fs
, gs
, ss
);
177 cr4
= read_cr4_safe();
178 printk("CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n",
185 printk("DR0: %08lx DR1: %08lx DR2: %08lx DR3: %08lx\n",
190 printk("DR6: %08lx DR7: %08lx\n",
194 void show_regs(struct pt_regs
*regs
)
196 __show_regs(regs
, 1);
197 show_trace(NULL
, regs
, ®s
->sp
, regs
->bp
);
201 * This gets run with %bx containing the
202 * function to call, and %dx containing
205 extern void kernel_thread_helper(void);
208 * Create a kernel thread
210 int kernel_thread(int (*fn
)(void *), void *arg
, unsigned long flags
)
214 memset(®s
, 0, sizeof(regs
));
216 regs
.bx
= (unsigned long) fn
;
217 regs
.dx
= (unsigned long) arg
;
221 regs
.fs
= __KERNEL_PERCPU
;
222 regs
.gs
= __KERNEL_STACK_CANARY
;
224 regs
.ip
= (unsigned long) kernel_thread_helper
;
225 regs
.cs
= __KERNEL_CS
| get_kernel_rpl();
226 regs
.flags
= X86_EFLAGS_IF
| X86_EFLAGS_SF
| X86_EFLAGS_PF
| 0x2;
228 /* Ok, create the new process.. */
229 return do_fork(flags
| CLONE_VM
| CLONE_UNTRACED
, 0, ®s
, 0, NULL
, NULL
);
231 EXPORT_SYMBOL(kernel_thread
);
234 * Free current thread data structures etc..
236 void exit_thread(void)
238 /* The process may have allocated an io port bitmap... nuke it. */
239 if (unlikely(test_thread_flag(TIF_IO_BITMAP
))) {
240 struct task_struct
*tsk
= current
;
241 struct thread_struct
*t
= &tsk
->thread
;
243 struct tss_struct
*tss
= &per_cpu(init_tss
, cpu
);
245 kfree(t
->io_bitmap_ptr
);
246 t
->io_bitmap_ptr
= NULL
;
247 clear_thread_flag(TIF_IO_BITMAP
);
249 * Careful, clear this in the TSS too:
251 memset(tss
->io_bitmap
, 0xff, tss
->io_bitmap_max
);
252 t
->io_bitmap_max
= 0;
253 tss
->io_bitmap_owner
= NULL
;
254 tss
->io_bitmap_max
= 0;
255 tss
->x86_tss
.io_bitmap_base
= INVALID_IO_BITMAP_OFFSET
;
259 ds_exit_thread(current
);
262 void flush_thread(void)
264 struct task_struct
*tsk
= current
;
266 tsk
->thread
.debugreg0
= 0;
267 tsk
->thread
.debugreg1
= 0;
268 tsk
->thread
.debugreg2
= 0;
269 tsk
->thread
.debugreg3
= 0;
270 tsk
->thread
.debugreg6
= 0;
271 tsk
->thread
.debugreg7
= 0;
272 memset(tsk
->thread
.tls_array
, 0, sizeof(tsk
->thread
.tls_array
));
273 clear_tsk_thread_flag(tsk
, TIF_DEBUG
);
275 * Forget coprocessor state..
277 tsk
->fpu_counter
= 0;
282 void release_thread(struct task_struct
*dead_task
)
284 BUG_ON(dead_task
->mm
);
285 release_vm86_irqs(dead_task
);
289 * This gets called before we allocate a new thread and copy
290 * the current task into it.
292 void prepare_to_copy(struct task_struct
*tsk
)
297 int copy_thread(int nr
, unsigned long clone_flags
, unsigned long sp
,
298 unsigned long unused
,
299 struct task_struct
*p
, struct pt_regs
*regs
)
301 struct pt_regs
*childregs
;
302 struct task_struct
*tsk
;
305 childregs
= task_pt_regs(p
);
310 p
->thread
.sp
= (unsigned long) childregs
;
311 p
->thread
.sp0
= (unsigned long) (childregs
+1);
313 p
->thread
.ip
= (unsigned long) ret_from_fork
;
315 task_user_gs(p
) = get_user_gs(regs
);
318 if (unlikely(test_tsk_thread_flag(tsk
, TIF_IO_BITMAP
))) {
319 p
->thread
.io_bitmap_ptr
= kmemdup(tsk
->thread
.io_bitmap_ptr
,
320 IO_BITMAP_BYTES
, GFP_KERNEL
);
321 if (!p
->thread
.io_bitmap_ptr
) {
322 p
->thread
.io_bitmap_max
= 0;
325 set_tsk_thread_flag(p
, TIF_IO_BITMAP
);
331 * Set a new TLS for the child thread?
333 if (clone_flags
& CLONE_SETTLS
)
334 err
= do_set_thread_area(p
, -1,
335 (struct user_desc __user
*)childregs
->si
, 0);
337 if (err
&& p
->thread
.io_bitmap_ptr
) {
338 kfree(p
->thread
.io_bitmap_ptr
);
339 p
->thread
.io_bitmap_max
= 0;
342 ds_copy_thread(p
, current
);
344 clear_tsk_thread_flag(p
, TIF_DEBUGCTLMSR
);
345 p
->thread
.debugctlmsr
= 0;
351 start_thread(struct pt_regs
*regs
, unsigned long new_ip
, unsigned long new_sp
)
353 set_user_gs(regs
, 0);
356 regs
->ds
= __USER_DS
;
357 regs
->es
= __USER_DS
;
358 regs
->ss
= __USER_DS
;
359 regs
->cs
= __USER_CS
;
363 * Free the old FP and other extended state
365 free_thread_xstate(current
);
367 EXPORT_SYMBOL_GPL(start_thread
);
369 static void hard_disable_TSC(void)
371 write_cr4(read_cr4() | X86_CR4_TSD
);
374 void disable_TSC(void)
377 if (!test_and_set_thread_flag(TIF_NOTSC
))
379 * Must flip the CPU state synchronously with
380 * TIF_NOTSC in the current running context.
386 static void hard_enable_TSC(void)
388 write_cr4(read_cr4() & ~X86_CR4_TSD
);
391 static void enable_TSC(void)
394 if (test_and_clear_thread_flag(TIF_NOTSC
))
396 * Must flip the CPU state synchronously with
397 * TIF_NOTSC in the current running context.
403 int get_tsc_mode(unsigned long adr
)
407 if (test_thread_flag(TIF_NOTSC
))
408 val
= PR_TSC_SIGSEGV
;
412 return put_user(val
, (unsigned int __user
*)adr
);
415 int set_tsc_mode(unsigned int val
)
417 if (val
== PR_TSC_SIGSEGV
)
419 else if (val
== PR_TSC_ENABLE
)
428 __switch_to_xtra(struct task_struct
*prev_p
, struct task_struct
*next_p
,
429 struct tss_struct
*tss
)
431 struct thread_struct
*prev
, *next
;
433 prev
= &prev_p
->thread
;
434 next
= &next_p
->thread
;
436 if (test_tsk_thread_flag(next_p
, TIF_DS_AREA_MSR
) ||
437 test_tsk_thread_flag(prev_p
, TIF_DS_AREA_MSR
))
438 ds_switch_to(prev_p
, next_p
);
439 else if (next
->debugctlmsr
!= prev
->debugctlmsr
)
440 update_debugctlmsr(next
->debugctlmsr
);
442 if (test_tsk_thread_flag(next_p
, TIF_DEBUG
)) {
443 set_debugreg(next
->debugreg0
, 0);
444 set_debugreg(next
->debugreg1
, 1);
445 set_debugreg(next
->debugreg2
, 2);
446 set_debugreg(next
->debugreg3
, 3);
448 set_debugreg(next
->debugreg6
, 6);
449 set_debugreg(next
->debugreg7
, 7);
452 if (test_tsk_thread_flag(prev_p
, TIF_NOTSC
) ^
453 test_tsk_thread_flag(next_p
, TIF_NOTSC
)) {
454 /* prev and next are different */
455 if (test_tsk_thread_flag(next_p
, TIF_NOTSC
))
461 if (!test_tsk_thread_flag(next_p
, TIF_IO_BITMAP
)) {
463 * Disable the bitmap via an invalid offset. We still cache
464 * the previous bitmap owner and the IO bitmap contents:
466 tss
->x86_tss
.io_bitmap_base
= INVALID_IO_BITMAP_OFFSET
;
470 if (likely(next
== tss
->io_bitmap_owner
)) {
472 * Previous owner of the bitmap (hence the bitmap content)
473 * matches the next task, we dont have to do anything but
474 * to set a valid offset in the TSS:
476 tss
->x86_tss
.io_bitmap_base
= IO_BITMAP_OFFSET
;
480 * Lazy TSS's I/O bitmap copy. We set an invalid offset here
481 * and we let the task to get a GPF in case an I/O instruction
482 * is performed. The handler of the GPF will verify that the
483 * faulting task has a valid I/O bitmap and, it true, does the
484 * real copy and restart the instruction. This will save us
485 * redundant copies when the currently switched task does not
486 * perform any I/O during its timeslice.
488 tss
->x86_tss
.io_bitmap_base
= INVALID_IO_BITMAP_OFFSET_LAZY
;
492 * switch_to(x,yn) should switch tasks from x to y.
494 * We fsave/fwait so that an exception goes off at the right time
495 * (as a call from the fsave or fwait in effect) rather than to
496 * the wrong process. Lazy FP saving no longer makes any sense
497 * with modern CPU's, and this simplifies a lot of things (SMP
498 * and UP become the same).
500 * NOTE! We used to use the x86 hardware context switching. The
501 * reason for not using it any more becomes apparent when you
502 * try to recover gracefully from saved state that is no longer
503 * valid (stale segment register values in particular). With the
504 * hardware task-switch, there is no way to fix up bad state in
505 * a reasonable manner.
507 * The fact that Intel documents the hardware task-switching to
508 * be slow is a fairly red herring - this code is not noticeably
509 * faster. However, there _is_ some room for improvement here,
510 * so the performance issues may eventually be a valid point.
511 * More important, however, is the fact that this allows us much
514 * The return value (in %ax) will be the "prev" task after
515 * the task-switch, and shows up in ret_from_fork in entry.S,
518 __notrace_funcgraph
struct task_struct
*
519 __switch_to(struct task_struct
*prev_p
, struct task_struct
*next_p
)
521 struct thread_struct
*prev
= &prev_p
->thread
,
522 *next
= &next_p
->thread
;
523 int cpu
= smp_processor_id();
524 struct tss_struct
*tss
= &per_cpu(init_tss
, cpu
);
526 /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
528 __unlazy_fpu(prev_p
);
531 /* we're going to use this soon, after a few expensive things */
532 if (next_p
->fpu_counter
> 5)
533 prefetch(next
->xstate
);
541 * Save away %gs. No need to save %fs, as it was saved on the
542 * stack on entry. No need to save %es and %ds, as those are
543 * always kernel segments while inside the kernel. Doing this
544 * before setting the new TLS descriptors avoids the situation
545 * where we temporarily have non-reloadable segments in %fs
546 * and %gs. This could be an issue if the NMI handler ever
547 * used %fs or %gs (it does not today), or if the kernel is
548 * running inside of a hypervisor layer.
550 lazy_save_gs(prev
->gs
);
553 * Load the per-thread Thread-Local Storage descriptor.
558 * Restore IOPL if needed. In normal use, the flags restore
559 * in the switch assembly will handle this. But if the kernel
560 * is running virtualized at a non-zero CPL, the popf will
561 * not restore flags, so it must be done in a separate step.
563 if (get_kernel_rpl() && unlikely(prev
->iopl
!= next
->iopl
))
564 set_iopl_mask(next
->iopl
);
567 * Now maybe handle debug registers and/or IO bitmaps
569 if (unlikely(task_thread_info(prev_p
)->flags
& _TIF_WORK_CTXSW_PREV
||
570 task_thread_info(next_p
)->flags
& _TIF_WORK_CTXSW_NEXT
))
571 __switch_to_xtra(prev_p
, next_p
, tss
);
574 * Leave lazy mode, flushing any hypercalls made here.
575 * This must be done before restoring TLS segments so
576 * the GDT and LDT are properly updated, and must be
577 * done before math_state_restore, so the TS bit is up
580 arch_leave_lazy_cpu_mode();
582 /* If the task has used fpu the last 5 timeslices, just do a full
583 * restore of the math state immediately to avoid the trap; the
584 * chances of needing FPU soon are obviously high now
586 * tsk_used_math() checks prevent calling math_state_restore(),
587 * which can sleep in the case of !tsk_used_math()
589 if (tsk_used_math(next_p
) && next_p
->fpu_counter
> 5)
590 math_state_restore();
593 * Restore %gs if needed (which is common)
595 if (prev
->gs
| next
->gs
)
596 lazy_load_gs(next
->gs
);
598 percpu_write(current_task
, next_p
);
603 int sys_fork(struct pt_regs
*regs
)
605 return do_fork(SIGCHLD
, regs
->sp
, regs
, 0, NULL
, NULL
);
608 int sys_clone(struct pt_regs
*regs
)
610 unsigned long clone_flags
;
612 int __user
*parent_tidptr
, *child_tidptr
;
614 clone_flags
= regs
->bx
;
616 parent_tidptr
= (int __user
*)regs
->dx
;
617 child_tidptr
= (int __user
*)regs
->di
;
620 return do_fork(clone_flags
, newsp
, regs
, 0, parent_tidptr
, child_tidptr
);
624 * This is trivial, and on the face of it looks like it
625 * could equally well be done in user mode.
627 * Not so, for quite unobvious reasons - register pressure.
628 * In user mode vfork() cannot have a stack frame, and if
629 * done by calling the "clone()" system call directly, you
630 * do not have enough call-clobbered registers to hold all
631 * the information you need.
633 int sys_vfork(struct pt_regs
*regs
)
635 return do_fork(CLONE_VFORK
| CLONE_VM
| SIGCHLD
, regs
->sp
, regs
, 0, NULL
, NULL
);
639 * sys_execve() executes a new program.
641 int sys_execve(struct pt_regs
*regs
)
646 filename
= getname((char __user
*) regs
->bx
);
647 error
= PTR_ERR(filename
);
648 if (IS_ERR(filename
))
650 error
= do_execve(filename
,
651 (char __user
* __user
*) regs
->cx
,
652 (char __user
* __user
*) regs
->dx
,
655 /* Make sure we don't return using sysenter.. */
656 set_thread_flag(TIF_IRET
);
663 #define top_esp (THREAD_SIZE - sizeof(unsigned long))
664 #define top_ebp (THREAD_SIZE - 2*sizeof(unsigned long))
666 unsigned long get_wchan(struct task_struct
*p
)
668 unsigned long bp
, sp
, ip
;
669 unsigned long stack_page
;
671 if (!p
|| p
== current
|| p
->state
== TASK_RUNNING
)
673 stack_page
= (unsigned long)task_stack_page(p
);
675 if (!stack_page
|| sp
< stack_page
|| sp
> top_esp
+stack_page
)
677 /* include/asm-i386/system.h:switch_to() pushes bp last. */
678 bp
= *(unsigned long *) sp
;
680 if (bp
< stack_page
|| bp
> top_ebp
+stack_page
)
682 ip
= *(unsigned long *) (bp
+4);
683 if (!in_sched_functions(ip
))
685 bp
= *(unsigned long *) bp
;
686 } while (count
++ < 16);
690 unsigned long arch_align_stack(unsigned long sp
)
692 if (!(current
->personality
& ADDR_NO_RANDOMIZE
) && randomize_va_space
)
693 sp
-= get_random_int() % 8192;
697 unsigned long arch_randomize_brk(struct mm_struct
*mm
)
699 unsigned long range_end
= mm
->brk
+ 0x02000000;
700 return randomize_range(mm
->brk
, range_end
, 0) ? : mm
->brk
;