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 (rcu_pending(cpu
))
115 rcu_check_callbacks(cpu
, 0);
117 if (cpu_is_offline(cpu
))
121 /* Don't trace irqs off for idle */
122 stop_critical_timings();
124 start_critical_timings();
126 tick_nohz_restart_sched_tick();
127 preempt_enable_no_resched();
133 void __show_regs(struct pt_regs
*regs
, int all
)
135 unsigned long cr0
= 0L, cr2
= 0L, cr3
= 0L, cr4
= 0L;
136 unsigned long d0
, d1
, d2
, d3
, d6
, d7
;
138 unsigned short ss
, gs
;
141 if (user_mode_vm(regs
)) {
143 ss
= regs
->ss
& 0xffff;
144 gs
= get_user_gs(regs
);
146 sp
= (unsigned long) (®s
->sp
);
153 board
= dmi_get_system_info(DMI_PRODUCT_NAME
);
156 printk("Pid: %d, comm: %s %s (%s %.*s) %s\n",
157 task_pid_nr(current
), current
->comm
,
158 print_tainted(), init_utsname()->release
,
159 (int)strcspn(init_utsname()->version
, " "),
160 init_utsname()->version
, board
);
162 printk("EIP: %04x:[<%08lx>] EFLAGS: %08lx CPU: %d\n",
163 (u16
)regs
->cs
, regs
->ip
, regs
->flags
,
165 print_symbol("EIP is at %s\n", regs
->ip
);
167 printk("EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
168 regs
->ax
, regs
->bx
, regs
->cx
, regs
->dx
);
169 printk("ESI: %08lx EDI: %08lx EBP: %08lx ESP: %08lx\n",
170 regs
->si
, regs
->di
, regs
->bp
, sp
);
171 printk(" DS: %04x ES: %04x FS: %04x GS: %04x SS: %04x\n",
172 (u16
)regs
->ds
, (u16
)regs
->es
, (u16
)regs
->fs
, gs
, ss
);
180 cr4
= read_cr4_safe();
181 printk("CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n",
188 printk("DR0: %08lx DR1: %08lx DR2: %08lx DR3: %08lx\n",
193 printk("DR6: %08lx DR7: %08lx\n",
197 void show_regs(struct pt_regs
*regs
)
199 __show_regs(regs
, 1);
200 show_trace(NULL
, regs
, ®s
->sp
, regs
->bp
);
204 * This gets run with %bx containing the
205 * function to call, and %dx containing
208 extern void kernel_thread_helper(void);
211 * Create a kernel thread
213 int kernel_thread(int (*fn
)(void *), void *arg
, unsigned long flags
)
217 memset(®s
, 0, sizeof(regs
));
219 regs
.bx
= (unsigned long) fn
;
220 regs
.dx
= (unsigned long) arg
;
224 regs
.fs
= __KERNEL_PERCPU
;
225 regs
.gs
= __KERNEL_STACK_CANARY
;
227 regs
.ip
= (unsigned long) kernel_thread_helper
;
228 regs
.cs
= __KERNEL_CS
| get_kernel_rpl();
229 regs
.flags
= X86_EFLAGS_IF
| X86_EFLAGS_SF
| X86_EFLAGS_PF
| 0x2;
231 /* Ok, create the new process.. */
232 return do_fork(flags
| CLONE_VM
| CLONE_UNTRACED
, 0, ®s
, 0, NULL
, NULL
);
234 EXPORT_SYMBOL(kernel_thread
);
237 * Free current thread data structures etc..
239 void exit_thread(void)
241 /* The process may have allocated an io port bitmap... nuke it. */
242 if (unlikely(test_thread_flag(TIF_IO_BITMAP
))) {
243 struct task_struct
*tsk
= current
;
244 struct thread_struct
*t
= &tsk
->thread
;
246 struct tss_struct
*tss
= &per_cpu(init_tss
, cpu
);
248 kfree(t
->io_bitmap_ptr
);
249 t
->io_bitmap_ptr
= NULL
;
250 clear_thread_flag(TIF_IO_BITMAP
);
252 * Careful, clear this in the TSS too:
254 memset(tss
->io_bitmap
, 0xff, tss
->io_bitmap_max
);
255 t
->io_bitmap_max
= 0;
256 tss
->io_bitmap_owner
= NULL
;
257 tss
->io_bitmap_max
= 0;
258 tss
->x86_tss
.io_bitmap_base
= INVALID_IO_BITMAP_OFFSET
;
262 ds_exit_thread(current
);
265 void flush_thread(void)
267 struct task_struct
*tsk
= current
;
269 tsk
->thread
.debugreg0
= 0;
270 tsk
->thread
.debugreg1
= 0;
271 tsk
->thread
.debugreg2
= 0;
272 tsk
->thread
.debugreg3
= 0;
273 tsk
->thread
.debugreg6
= 0;
274 tsk
->thread
.debugreg7
= 0;
275 memset(tsk
->thread
.tls_array
, 0, sizeof(tsk
->thread
.tls_array
));
276 clear_tsk_thread_flag(tsk
, TIF_DEBUG
);
278 * Forget coprocessor state..
280 tsk
->fpu_counter
= 0;
285 void release_thread(struct task_struct
*dead_task
)
287 BUG_ON(dead_task
->mm
);
288 release_vm86_irqs(dead_task
);
292 * This gets called before we allocate a new thread and copy
293 * the current task into it.
295 void prepare_to_copy(struct task_struct
*tsk
)
300 int copy_thread(int nr
, unsigned long clone_flags
, unsigned long sp
,
301 unsigned long unused
,
302 struct task_struct
*p
, struct pt_regs
*regs
)
304 struct pt_regs
*childregs
;
305 struct task_struct
*tsk
;
308 childregs
= task_pt_regs(p
);
313 p
->thread
.sp
= (unsigned long) childregs
;
314 p
->thread
.sp0
= (unsigned long) (childregs
+1);
316 p
->thread
.ip
= (unsigned long) ret_from_fork
;
318 task_user_gs(p
) = get_user_gs(regs
);
321 if (unlikely(test_tsk_thread_flag(tsk
, TIF_IO_BITMAP
))) {
322 p
->thread
.io_bitmap_ptr
= kmemdup(tsk
->thread
.io_bitmap_ptr
,
323 IO_BITMAP_BYTES
, GFP_KERNEL
);
324 if (!p
->thread
.io_bitmap_ptr
) {
325 p
->thread
.io_bitmap_max
= 0;
328 set_tsk_thread_flag(p
, TIF_IO_BITMAP
);
334 * Set a new TLS for the child thread?
336 if (clone_flags
& CLONE_SETTLS
)
337 err
= do_set_thread_area(p
, -1,
338 (struct user_desc __user
*)childregs
->si
, 0);
340 if (err
&& p
->thread
.io_bitmap_ptr
) {
341 kfree(p
->thread
.io_bitmap_ptr
);
342 p
->thread
.io_bitmap_max
= 0;
345 ds_copy_thread(p
, current
);
347 clear_tsk_thread_flag(p
, TIF_DEBUGCTLMSR
);
348 p
->thread
.debugctlmsr
= 0;
354 start_thread(struct pt_regs
*regs
, unsigned long new_ip
, unsigned long new_sp
)
356 set_user_gs(regs
, 0);
359 regs
->ds
= __USER_DS
;
360 regs
->es
= __USER_DS
;
361 regs
->ss
= __USER_DS
;
362 regs
->cs
= __USER_CS
;
366 * Free the old FP and other extended state
368 free_thread_xstate(current
);
370 EXPORT_SYMBOL_GPL(start_thread
);
372 static void hard_disable_TSC(void)
374 write_cr4(read_cr4() | X86_CR4_TSD
);
377 void disable_TSC(void)
380 if (!test_and_set_thread_flag(TIF_NOTSC
))
382 * Must flip the CPU state synchronously with
383 * TIF_NOTSC in the current running context.
389 static void hard_enable_TSC(void)
391 write_cr4(read_cr4() & ~X86_CR4_TSD
);
394 static void enable_TSC(void)
397 if (test_and_clear_thread_flag(TIF_NOTSC
))
399 * Must flip the CPU state synchronously with
400 * TIF_NOTSC in the current running context.
406 int get_tsc_mode(unsigned long adr
)
410 if (test_thread_flag(TIF_NOTSC
))
411 val
= PR_TSC_SIGSEGV
;
415 return put_user(val
, (unsigned int __user
*)adr
);
418 int set_tsc_mode(unsigned int val
)
420 if (val
== PR_TSC_SIGSEGV
)
422 else if (val
== PR_TSC_ENABLE
)
431 __switch_to_xtra(struct task_struct
*prev_p
, struct task_struct
*next_p
,
432 struct tss_struct
*tss
)
434 struct thread_struct
*prev
, *next
;
436 prev
= &prev_p
->thread
;
437 next
= &next_p
->thread
;
439 if (test_tsk_thread_flag(next_p
, TIF_DS_AREA_MSR
) ||
440 test_tsk_thread_flag(prev_p
, TIF_DS_AREA_MSR
))
441 ds_switch_to(prev_p
, next_p
);
442 else if (next
->debugctlmsr
!= prev
->debugctlmsr
)
443 update_debugctlmsr(next
->debugctlmsr
);
445 if (test_tsk_thread_flag(next_p
, TIF_DEBUG
)) {
446 set_debugreg(next
->debugreg0
, 0);
447 set_debugreg(next
->debugreg1
, 1);
448 set_debugreg(next
->debugreg2
, 2);
449 set_debugreg(next
->debugreg3
, 3);
451 set_debugreg(next
->debugreg6
, 6);
452 set_debugreg(next
->debugreg7
, 7);
455 if (test_tsk_thread_flag(prev_p
, TIF_NOTSC
) ^
456 test_tsk_thread_flag(next_p
, TIF_NOTSC
)) {
457 /* prev and next are different */
458 if (test_tsk_thread_flag(next_p
, TIF_NOTSC
))
464 if (!test_tsk_thread_flag(next_p
, TIF_IO_BITMAP
)) {
466 * Disable the bitmap via an invalid offset. We still cache
467 * the previous bitmap owner and the IO bitmap contents:
469 tss
->x86_tss
.io_bitmap_base
= INVALID_IO_BITMAP_OFFSET
;
473 if (likely(next
== tss
->io_bitmap_owner
)) {
475 * Previous owner of the bitmap (hence the bitmap content)
476 * matches the next task, we dont have to do anything but
477 * to set a valid offset in the TSS:
479 tss
->x86_tss
.io_bitmap_base
= IO_BITMAP_OFFSET
;
483 * Lazy TSS's I/O bitmap copy. We set an invalid offset here
484 * and we let the task to get a GPF in case an I/O instruction
485 * is performed. The handler of the GPF will verify that the
486 * faulting task has a valid I/O bitmap and, it true, does the
487 * real copy and restart the instruction. This will save us
488 * redundant copies when the currently switched task does not
489 * perform any I/O during its timeslice.
491 tss
->x86_tss
.io_bitmap_base
= INVALID_IO_BITMAP_OFFSET_LAZY
;
495 * switch_to(x,yn) should switch tasks from x to y.
497 * We fsave/fwait so that an exception goes off at the right time
498 * (as a call from the fsave or fwait in effect) rather than to
499 * the wrong process. Lazy FP saving no longer makes any sense
500 * with modern CPU's, and this simplifies a lot of things (SMP
501 * and UP become the same).
503 * NOTE! We used to use the x86 hardware context switching. The
504 * reason for not using it any more becomes apparent when you
505 * try to recover gracefully from saved state that is no longer
506 * valid (stale segment register values in particular). With the
507 * hardware task-switch, there is no way to fix up bad state in
508 * a reasonable manner.
510 * The fact that Intel documents the hardware task-switching to
511 * be slow is a fairly red herring - this code is not noticeably
512 * faster. However, there _is_ some room for improvement here,
513 * so the performance issues may eventually be a valid point.
514 * More important, however, is the fact that this allows us much
517 * The return value (in %ax) will be the "prev" task after
518 * the task-switch, and shows up in ret_from_fork in entry.S,
521 __notrace_funcgraph
struct task_struct
*
522 __switch_to(struct task_struct
*prev_p
, struct task_struct
*next_p
)
524 struct thread_struct
*prev
= &prev_p
->thread
,
525 *next
= &next_p
->thread
;
526 int cpu
= smp_processor_id();
527 struct tss_struct
*tss
= &per_cpu(init_tss
, cpu
);
529 /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
531 __unlazy_fpu(prev_p
);
534 /* we're going to use this soon, after a few expensive things */
535 if (next_p
->fpu_counter
> 5)
536 prefetch(next
->xstate
);
544 * Save away %gs. No need to save %fs, as it was saved on the
545 * stack on entry. No need to save %es and %ds, as those are
546 * always kernel segments while inside the kernel. Doing this
547 * before setting the new TLS descriptors avoids the situation
548 * where we temporarily have non-reloadable segments in %fs
549 * and %gs. This could be an issue if the NMI handler ever
550 * used %fs or %gs (it does not today), or if the kernel is
551 * running inside of a hypervisor layer.
553 lazy_save_gs(prev
->gs
);
556 * Load the per-thread Thread-Local Storage descriptor.
561 * Restore IOPL if needed. In normal use, the flags restore
562 * in the switch assembly will handle this. But if the kernel
563 * is running virtualized at a non-zero CPL, the popf will
564 * not restore flags, so it must be done in a separate step.
566 if (get_kernel_rpl() && unlikely(prev
->iopl
!= next
->iopl
))
567 set_iopl_mask(next
->iopl
);
570 * Now maybe handle debug registers and/or IO bitmaps
572 if (unlikely(task_thread_info(prev_p
)->flags
& _TIF_WORK_CTXSW_PREV
||
573 task_thread_info(next_p
)->flags
& _TIF_WORK_CTXSW_NEXT
))
574 __switch_to_xtra(prev_p
, next_p
, tss
);
577 * Leave lazy mode, flushing any hypercalls made here.
578 * This must be done before restoring TLS segments so
579 * the GDT and LDT are properly updated, and must be
580 * done before math_state_restore, so the TS bit is up
583 arch_leave_lazy_cpu_mode();
585 /* If the task has used fpu the last 5 timeslices, just do a full
586 * restore of the math state immediately to avoid the trap; the
587 * chances of needing FPU soon are obviously high now
589 * tsk_used_math() checks prevent calling math_state_restore(),
590 * which can sleep in the case of !tsk_used_math()
592 if (tsk_used_math(next_p
) && next_p
->fpu_counter
> 5)
593 math_state_restore();
596 * Restore %gs if needed (which is common)
598 if (prev
->gs
| next
->gs
)
599 lazy_load_gs(next
->gs
);
601 percpu_write(current_task
, next_p
);
606 ptregscall
int sys_fork(struct pt_regs
*regs
)
608 return do_fork(SIGCHLD
, regs
->sp
, regs
, 0, NULL
, NULL
);
611 ptregscall
int sys_clone(struct pt_regs
*regs
, unsigned long clone_flags
,
612 unsigned long newsp
, int __user
*parent_tidptr
,
613 unsigned long unused
, int __user
*child_tidptr
)
617 return do_fork(clone_flags
, newsp
, regs
, 0, parent_tidptr
, child_tidptr
);
621 * This is trivial, and on the face of it looks like it
622 * could equally well be done in user mode.
624 * Not so, for quite unobvious reasons - register pressure.
625 * In user mode vfork() cannot have a stack frame, and if
626 * done by calling the "clone()" system call directly, you
627 * do not have enough call-clobbered registers to hold all
628 * the information you need.
630 ptregscall
int sys_vfork(struct pt_regs
*regs
)
632 return do_fork(CLONE_VFORK
| CLONE_VM
| SIGCHLD
, regs
->sp
, regs
, 0, NULL
, NULL
);
636 * sys_execve() executes a new program.
638 ptregscall
int sys_execve(struct pt_regs
*regs
, char __user
*u_filename
,
639 char __user
* __user
*argv
,
640 char __user
* __user
*envp
)
645 filename
= getname(u_filename
);
646 error
= PTR_ERR(filename
);
647 if (IS_ERR(filename
))
649 error
= do_execve(filename
, argv
, envp
, regs
);
651 /* Make sure we don't return using sysenter.. */
652 set_thread_flag(TIF_IRET
);
659 #define top_esp (THREAD_SIZE - sizeof(unsigned long))
660 #define top_ebp (THREAD_SIZE - 2*sizeof(unsigned long))
662 unsigned long get_wchan(struct task_struct
*p
)
664 unsigned long bp
, sp
, ip
;
665 unsigned long stack_page
;
667 if (!p
|| p
== current
|| p
->state
== TASK_RUNNING
)
669 stack_page
= (unsigned long)task_stack_page(p
);
671 if (!stack_page
|| sp
< stack_page
|| sp
> top_esp
+stack_page
)
673 /* include/asm-i386/system.h:switch_to() pushes bp last. */
674 bp
= *(unsigned long *) sp
;
676 if (bp
< stack_page
|| bp
> top_ebp
+stack_page
)
678 ip
= *(unsigned long *) (bp
+4);
679 if (!in_sched_functions(ip
))
681 bp
= *(unsigned long *) bp
;
682 } while (count
++ < 16);
686 unsigned long arch_align_stack(unsigned long sp
)
688 if (!(current
->personality
& ADDR_NO_RANDOMIZE
) && randomize_va_space
)
689 sp
-= get_random_int() % 8192;
693 unsigned long arch_randomize_brk(struct mm_struct
*mm
)
695 unsigned long range_end
= mm
->brk
+ 0x02000000;
696 return randomize_range(mm
->brk
, range_end
, 0) ? : mm
->brk
;