]>
git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - arch/x86/kernel/process_64.c
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>
40 #include <linux/uaccess.h>
42 #include <linux/ftrace.h>
44 #include <asm/pgtable.h>
45 #include <asm/system.h>
46 #include <asm/processor.h>
48 #include <asm/mmu_context.h>
50 #include <asm/prctl.h>
52 #include <asm/proto.h>
55 #include <asm/syscalls.h>
58 asmlinkage
extern void ret_from_fork(void);
60 DEFINE_PER_CPU(struct task_struct
*, current_task
) = &init_task
;
61 EXPORT_PER_CPU_SYMBOL(current_task
);
63 unsigned long kernel_thread_flags
= CLONE_VM
| CLONE_UNTRACED
;
65 static ATOMIC_NOTIFIER_HEAD(idle_notifier
);
67 void idle_notifier_register(struct notifier_block
*n
)
69 atomic_notifier_chain_register(&idle_notifier
, n
);
71 EXPORT_SYMBOL_GPL(idle_notifier_register
);
73 void idle_notifier_unregister(struct notifier_block
*n
)
75 atomic_notifier_chain_unregister(&idle_notifier
, n
);
77 EXPORT_SYMBOL_GPL(idle_notifier_unregister
);
82 atomic_notifier_call_chain(&idle_notifier
, IDLE_START
, NULL
);
85 static void __exit_idle(void)
87 if (test_and_clear_bit_pda(0, isidle
) == 0)
89 atomic_notifier_call_chain(&idle_notifier
, IDLE_END
, NULL
);
92 /* Called from interrupts to signify idle end */
95 /* idle loop has pid 0 */
102 static inline void play_dead(void)
109 * The idle thread. There's no useful work to be
110 * done, so just try to conserve power and have a
111 * low exit latency (ie sit in a loop waiting for
112 * somebody to say that they'd like to reschedule)
116 current_thread_info()->status
|= TS_POLLING
;
117 /* endless idle loop with no priority at all */
119 tick_nohz_stop_sched_tick(1);
120 while (!need_resched()) {
124 if (cpu_is_offline(smp_processor_id()))
127 * Idle routines should keep interrupts disabled
128 * from here on, until they go to idle.
129 * Otherwise, idle callbacks can misfire.
133 /* Don't trace irqs off for idle */
134 stop_critical_timings();
136 start_critical_timings();
137 /* In many cases the interrupt that ended idle
138 has already called exit_idle. But some idle
139 loops can be woken up without interrupt. */
143 tick_nohz_restart_sched_tick();
144 preempt_enable_no_resched();
150 /* Prints also some state that isn't saved in the pt_regs */
151 void __show_regs(struct pt_regs
*regs
, int all
)
153 unsigned long cr0
= 0L, cr2
= 0L, cr3
= 0L, cr4
= 0L, fs
, gs
, shadowgs
;
154 unsigned long d0
, d1
, d2
, d3
, d6
, d7
;
155 unsigned int fsindex
, gsindex
;
156 unsigned int ds
, cs
, es
;
160 printk(KERN_INFO
"Pid: %d, comm: %.20s %s %s %.*s\n",
161 current
->pid
, current
->comm
, print_tainted(),
162 init_utsname()->release
,
163 (int)strcspn(init_utsname()->version
, " "),
164 init_utsname()->version
);
165 printk(KERN_INFO
"RIP: %04lx:[<%016lx>] ", regs
->cs
& 0xffff, regs
->ip
);
166 printk_address(regs
->ip
, 1);
167 printk(KERN_INFO
"RSP: %04lx:%016lx EFLAGS: %08lx\n", regs
->ss
,
168 regs
->sp
, regs
->flags
);
169 printk(KERN_INFO
"RAX: %016lx RBX: %016lx RCX: %016lx\n",
170 regs
->ax
, regs
->bx
, regs
->cx
);
171 printk(KERN_INFO
"RDX: %016lx RSI: %016lx RDI: %016lx\n",
172 regs
->dx
, regs
->si
, regs
->di
);
173 printk(KERN_INFO
"RBP: %016lx R08: %016lx R09: %016lx\n",
174 regs
->bp
, regs
->r8
, regs
->r9
);
175 printk(KERN_INFO
"R10: %016lx R11: %016lx R12: %016lx\n",
176 regs
->r10
, regs
->r11
, regs
->r12
);
177 printk(KERN_INFO
"R13: %016lx R14: %016lx R15: %016lx\n",
178 regs
->r13
, regs
->r14
, regs
->r15
);
180 asm("movl %%ds,%0" : "=r" (ds
));
181 asm("movl %%cs,%0" : "=r" (cs
));
182 asm("movl %%es,%0" : "=r" (es
));
183 asm("movl %%fs,%0" : "=r" (fsindex
));
184 asm("movl %%gs,%0" : "=r" (gsindex
));
186 rdmsrl(MSR_FS_BASE
, fs
);
187 rdmsrl(MSR_GS_BASE
, gs
);
188 rdmsrl(MSR_KERNEL_GS_BASE
, shadowgs
);
198 printk(KERN_INFO
"FS: %016lx(%04x) GS:%016lx(%04x) knlGS:%016lx\n",
199 fs
, fsindex
, gs
, gsindex
, shadowgs
);
200 printk(KERN_INFO
"CS: %04x DS: %04x ES: %04x CR0: %016lx\n", cs
, ds
,
202 printk(KERN_INFO
"CR2: %016lx CR3: %016lx CR4: %016lx\n", cr2
, cr3
,
208 printk(KERN_INFO
"DR0: %016lx DR1: %016lx DR2: %016lx\n", d0
, d1
, d2
);
212 printk(KERN_INFO
"DR3: %016lx DR6: %016lx DR7: %016lx\n", d3
, d6
, d7
);
215 void show_regs(struct pt_regs
*regs
)
217 printk(KERN_INFO
"CPU %d:", smp_processor_id());
218 __show_regs(regs
, 1);
219 show_trace(NULL
, regs
, (void *)(regs
+ 1), regs
->bp
);
223 * Free current thread data structures etc..
225 void exit_thread(void)
227 struct task_struct
*me
= current
;
228 struct thread_struct
*t
= &me
->thread
;
230 if (me
->thread
.io_bitmap_ptr
) {
231 struct tss_struct
*tss
= &per_cpu(init_tss
, get_cpu());
233 kfree(t
->io_bitmap_ptr
);
234 t
->io_bitmap_ptr
= NULL
;
235 clear_thread_flag(TIF_IO_BITMAP
);
237 * Careful, clear this in the TSS too:
239 memset(tss
->io_bitmap
, 0xff, t
->io_bitmap_max
);
240 t
->io_bitmap_max
= 0;
244 ds_exit_thread(current
);
247 void flush_thread(void)
249 struct task_struct
*tsk
= current
;
251 if (test_tsk_thread_flag(tsk
, TIF_ABI_PENDING
)) {
252 clear_tsk_thread_flag(tsk
, TIF_ABI_PENDING
);
253 if (test_tsk_thread_flag(tsk
, TIF_IA32
)) {
254 clear_tsk_thread_flag(tsk
, TIF_IA32
);
256 set_tsk_thread_flag(tsk
, TIF_IA32
);
257 current_thread_info()->status
|= TS_COMPAT
;
260 clear_tsk_thread_flag(tsk
, TIF_DEBUG
);
262 tsk
->thread
.debugreg0
= 0;
263 tsk
->thread
.debugreg1
= 0;
264 tsk
->thread
.debugreg2
= 0;
265 tsk
->thread
.debugreg3
= 0;
266 tsk
->thread
.debugreg6
= 0;
267 tsk
->thread
.debugreg7
= 0;
268 memset(tsk
->thread
.tls_array
, 0, sizeof(tsk
->thread
.tls_array
));
270 * Forget coprocessor state..
272 tsk
->fpu_counter
= 0;
277 void release_thread(struct task_struct
*dead_task
)
280 if (dead_task
->mm
->context
.size
) {
281 printk("WARNING: dead process %8s still has LDT? <%p/%d>\n",
283 dead_task
->mm
->context
.ldt
,
284 dead_task
->mm
->context
.size
);
290 static inline void set_32bit_tls(struct task_struct
*t
, int tls
, u32 addr
)
292 struct user_desc ud
= {
299 struct desc_struct
*desc
= t
->thread
.tls_array
;
304 static inline u32
read_32bit_tls(struct task_struct
*t
, int tls
)
306 return get_desc_base(&t
->thread
.tls_array
[tls
]);
310 * This gets called before we allocate a new thread and copy
311 * the current task into it.
313 void prepare_to_copy(struct task_struct
*tsk
)
318 int copy_thread(int nr
, unsigned long clone_flags
, unsigned long sp
,
319 unsigned long unused
,
320 struct task_struct
*p
, struct pt_regs
*regs
)
323 struct pt_regs
*childregs
;
324 struct task_struct
*me
= current
;
326 childregs
= ((struct pt_regs
*)
327 (THREAD_SIZE
+ task_stack_page(p
))) - 1;
333 childregs
->sp
= (unsigned long)childregs
;
335 p
->thread
.sp
= (unsigned long) childregs
;
336 p
->thread
.sp0
= (unsigned long) (childregs
+1);
337 p
->thread
.usersp
= me
->thread
.usersp
;
339 set_tsk_thread_flag(p
, TIF_FORK
);
341 p
->thread
.fs
= me
->thread
.fs
;
342 p
->thread
.gs
= me
->thread
.gs
;
344 savesegment(gs
, p
->thread
.gsindex
);
345 savesegment(fs
, p
->thread
.fsindex
);
346 savesegment(es
, p
->thread
.es
);
347 savesegment(ds
, p
->thread
.ds
);
349 if (unlikely(test_tsk_thread_flag(me
, TIF_IO_BITMAP
))) {
350 p
->thread
.io_bitmap_ptr
= kmalloc(IO_BITMAP_BYTES
, GFP_KERNEL
);
351 if (!p
->thread
.io_bitmap_ptr
) {
352 p
->thread
.io_bitmap_max
= 0;
355 memcpy(p
->thread
.io_bitmap_ptr
, me
->thread
.io_bitmap_ptr
,
357 set_tsk_thread_flag(p
, TIF_IO_BITMAP
);
361 * Set a new TLS for the child thread?
363 if (clone_flags
& CLONE_SETTLS
) {
364 #ifdef CONFIG_IA32_EMULATION
365 if (test_thread_flag(TIF_IA32
))
366 err
= do_set_thread_area(p
, -1,
367 (struct user_desc __user
*)childregs
->si
, 0);
370 err
= do_arch_prctl(p
, ARCH_SET_FS
, childregs
->r8
);
375 ds_copy_thread(p
, me
);
377 clear_tsk_thread_flag(p
, TIF_DEBUGCTLMSR
);
378 p
->thread
.debugctlmsr
= 0;
382 if (err
&& p
->thread
.io_bitmap_ptr
) {
383 kfree(p
->thread
.io_bitmap_ptr
);
384 p
->thread
.io_bitmap_max
= 0;
390 start_thread(struct pt_regs
*regs
, unsigned long new_ip
, unsigned long new_sp
)
398 write_pda(oldrsp
, new_sp
);
399 regs
->cs
= __USER_CS
;
400 regs
->ss
= __USER_DS
;
404 * Free the old FP and other extended state
406 free_thread_xstate(current
);
408 EXPORT_SYMBOL_GPL(start_thread
);
410 static void hard_disable_TSC(void)
412 write_cr4(read_cr4() | X86_CR4_TSD
);
415 void disable_TSC(void)
418 if (!test_and_set_thread_flag(TIF_NOTSC
))
420 * Must flip the CPU state synchronously with
421 * TIF_NOTSC in the current running context.
427 static void hard_enable_TSC(void)
429 write_cr4(read_cr4() & ~X86_CR4_TSD
);
432 static void enable_TSC(void)
435 if (test_and_clear_thread_flag(TIF_NOTSC
))
437 * Must flip the CPU state synchronously with
438 * TIF_NOTSC in the current running context.
444 int get_tsc_mode(unsigned long adr
)
448 if (test_thread_flag(TIF_NOTSC
))
449 val
= PR_TSC_SIGSEGV
;
453 return put_user(val
, (unsigned int __user
*)adr
);
456 int set_tsc_mode(unsigned int val
)
458 if (val
== PR_TSC_SIGSEGV
)
460 else if (val
== PR_TSC_ENABLE
)
469 * This special macro can be used to load a debugging register
471 #define loaddebug(thread, r) set_debugreg(thread->debugreg ## r, r)
473 static inline void __switch_to_xtra(struct task_struct
*prev_p
,
474 struct task_struct
*next_p
,
475 struct tss_struct
*tss
)
477 struct thread_struct
*prev
, *next
;
479 prev
= &prev_p
->thread
,
480 next
= &next_p
->thread
;
482 if (test_tsk_thread_flag(next_p
, TIF_DS_AREA_MSR
) ||
483 test_tsk_thread_flag(prev_p
, TIF_DS_AREA_MSR
))
484 ds_switch_to(prev_p
, next_p
);
485 else if (next
->debugctlmsr
!= prev
->debugctlmsr
)
486 update_debugctlmsr(next
->debugctlmsr
);
488 if (test_tsk_thread_flag(next_p
, TIF_DEBUG
)) {
498 if (test_tsk_thread_flag(prev_p
, TIF_NOTSC
) ^
499 test_tsk_thread_flag(next_p
, TIF_NOTSC
)) {
500 /* prev and next are different */
501 if (test_tsk_thread_flag(next_p
, TIF_NOTSC
))
507 if (test_tsk_thread_flag(next_p
, TIF_IO_BITMAP
)) {
509 * Copy the relevant range of the IO bitmap.
510 * Normally this is 128 bytes or less:
512 memcpy(tss
->io_bitmap
, next
->io_bitmap_ptr
,
513 max(prev
->io_bitmap_max
, next
->io_bitmap_max
));
514 } else if (test_tsk_thread_flag(prev_p
, TIF_IO_BITMAP
)) {
516 * Clear any possible leftover bits:
518 memset(tss
->io_bitmap
, 0xff, prev
->io_bitmap_max
);
523 * switch_to(x,y) should switch tasks from x to y.
525 * This could still be optimized:
526 * - fold all the options into a flag word and test it with a single test.
527 * - could test fs/gs bitsliced
529 * Kprobes not supported here. Set the probe on schedule instead.
530 * Function graph tracer not supported too.
532 __notrace_funcgraph
struct task_struct
*
533 __switch_to(struct task_struct
*prev_p
, struct task_struct
*next_p
)
535 struct thread_struct
*prev
= &prev_p
->thread
;
536 struct thread_struct
*next
= &next_p
->thread
;
537 int cpu
= smp_processor_id();
538 struct tss_struct
*tss
= &per_cpu(init_tss
, cpu
);
539 unsigned fsindex
, gsindex
;
541 /* we're going to use this soon, after a few expensive things */
542 if (next_p
->fpu_counter
> 5)
543 prefetch(next
->xstate
);
546 * Reload esp0, LDT and the page table pointer:
552 * This won't pick up thread selector changes, but I guess that is ok.
554 savesegment(es
, prev
->es
);
555 if (unlikely(next
->es
| prev
->es
))
556 loadsegment(es
, next
->es
);
558 savesegment(ds
, prev
->ds
);
559 if (unlikely(next
->ds
| prev
->ds
))
560 loadsegment(ds
, next
->ds
);
563 /* We must save %fs and %gs before load_TLS() because
564 * %fs and %gs may be cleared by load_TLS().
566 * (e.g. xen_load_tls())
568 savesegment(fs
, fsindex
);
569 savesegment(gs
, gsindex
);
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();
585 * Segment register != 0 always requires a reload. Also
586 * reload when it has changed. When prev process used 64bit
587 * base always reload to avoid an information leak.
589 if (unlikely(fsindex
| next
->fsindex
| prev
->fs
)) {
590 loadsegment(fs
, next
->fsindex
);
592 * Check if the user used a selector != 0; if yes
593 * clear 64bit base, since overloaded base is always
594 * mapped to the Null selector
599 /* when next process has a 64bit base use it */
601 wrmsrl(MSR_FS_BASE
, next
->fs
);
602 prev
->fsindex
= fsindex
;
604 if (unlikely(gsindex
| next
->gsindex
| prev
->gs
)) {
605 load_gs_index(next
->gsindex
);
610 wrmsrl(MSR_KERNEL_GS_BASE
, next
->gs
);
611 prev
->gsindex
= gsindex
;
613 /* Must be after DS reload */
617 * Switch the PDA and FPU contexts.
619 prev
->usersp
= read_pda(oldrsp
);
620 write_pda(oldrsp
, next
->usersp
);
621 percpu_write(current_task
, next_p
);
623 percpu_write(kernel_stack
,
624 (unsigned long)task_stack_page(next_p
) +
625 THREAD_SIZE
- KERNEL_STACK_OFFSET
);
626 #ifdef CONFIG_CC_STACKPROTECTOR
627 write_pda(stack_canary
, next_p
->stack_canary
);
629 * Build time only check to make sure the stack_canary is at
630 * offset 40 in the pda; this is a gcc ABI requirement
632 BUILD_BUG_ON(offsetof(struct x8664_pda
, stack_canary
) != 40);
636 * Now maybe reload the debug registers and handle I/O bitmaps
638 if (unlikely(task_thread_info(next_p
)->flags
& _TIF_WORK_CTXSW_NEXT
||
639 task_thread_info(prev_p
)->flags
& _TIF_WORK_CTXSW_PREV
))
640 __switch_to_xtra(prev_p
, next_p
, tss
);
642 /* If the task has used fpu the last 5 timeslices, just do a full
643 * restore of the math state immediately to avoid the trap; the
644 * chances of needing FPU soon are obviously high now
646 * tsk_used_math() checks prevent calling math_state_restore(),
647 * which can sleep in the case of !tsk_used_math()
649 if (tsk_used_math(next_p
) && next_p
->fpu_counter
> 5)
650 math_state_restore();
655 * sys_execve() executes a new program.
658 long sys_execve(char __user
*name
, char __user
* __user
*argv
,
659 char __user
* __user
*envp
, struct pt_regs
*regs
)
664 filename
= getname(name
);
665 error
= PTR_ERR(filename
);
666 if (IS_ERR(filename
))
668 error
= do_execve(filename
, argv
, envp
, regs
);
673 void set_personality_64bit(void)
675 /* inherit personality from parent */
677 /* Make sure to be in 64bit mode */
678 clear_thread_flag(TIF_IA32
);
680 /* TBD: overwrites user setup. Should have two bits.
681 But 64bit processes have always behaved this way,
682 so it's not too bad. The main problem is just that
683 32bit childs are affected again. */
684 current
->personality
&= ~READ_IMPLIES_EXEC
;
687 asmlinkage
long sys_fork(struct pt_regs
*regs
)
689 return do_fork(SIGCHLD
, regs
->sp
, regs
, 0, NULL
, NULL
);
693 sys_clone(unsigned long clone_flags
, unsigned long newsp
,
694 void __user
*parent_tid
, void __user
*child_tid
, struct pt_regs
*regs
)
698 return do_fork(clone_flags
, newsp
, regs
, 0, parent_tid
, child_tid
);
702 * This is trivial, and on the face of it looks like it
703 * could equally well be done in user mode.
705 * Not so, for quite unobvious reasons - register pressure.
706 * In user mode vfork() cannot have a stack frame, and if
707 * done by calling the "clone()" system call directly, you
708 * do not have enough call-clobbered registers to hold all
709 * the information you need.
711 asmlinkage
long sys_vfork(struct pt_regs
*regs
)
713 return do_fork(CLONE_VFORK
| CLONE_VM
| SIGCHLD
, regs
->sp
, regs
, 0,
717 unsigned long get_wchan(struct task_struct
*p
)
723 if (!p
|| p
== current
|| p
->state
== TASK_RUNNING
)
725 stack
= (unsigned long)task_stack_page(p
);
726 if (p
->thread
.sp
< stack
|| p
->thread
.sp
>= stack
+THREAD_SIZE
)
728 fp
= *(u64
*)(p
->thread
.sp
);
730 if (fp
< (unsigned long)stack
||
731 fp
>= (unsigned long)stack
+THREAD_SIZE
)
734 if (!in_sched_functions(ip
))
737 } while (count
++ < 16);
741 long do_arch_prctl(struct task_struct
*task
, int code
, unsigned long addr
)
744 int doit
= task
== current
;
749 if (addr
>= TASK_SIZE_OF(task
))
752 /* handle small bases via the GDT because that's faster to
754 if (addr
<= 0xffffffff) {
755 set_32bit_tls(task
, GS_TLS
, addr
);
757 load_TLS(&task
->thread
, cpu
);
758 load_gs_index(GS_TLS_SEL
);
760 task
->thread
.gsindex
= GS_TLS_SEL
;
763 task
->thread
.gsindex
= 0;
764 task
->thread
.gs
= addr
;
767 ret
= checking_wrmsrl(MSR_KERNEL_GS_BASE
, addr
);
773 /* Not strictly needed for fs, but do it for symmetry
775 if (addr
>= TASK_SIZE_OF(task
))
778 /* handle small bases via the GDT because that's faster to
780 if (addr
<= 0xffffffff) {
781 set_32bit_tls(task
, FS_TLS
, addr
);
783 load_TLS(&task
->thread
, cpu
);
784 loadsegment(fs
, FS_TLS_SEL
);
786 task
->thread
.fsindex
= FS_TLS_SEL
;
789 task
->thread
.fsindex
= 0;
790 task
->thread
.fs
= addr
;
792 /* set the selector to 0 to not confuse
795 ret
= checking_wrmsrl(MSR_FS_BASE
, addr
);
802 if (task
->thread
.fsindex
== FS_TLS_SEL
)
803 base
= read_32bit_tls(task
, FS_TLS
);
805 rdmsrl(MSR_FS_BASE
, base
);
807 base
= task
->thread
.fs
;
808 ret
= put_user(base
, (unsigned long __user
*)addr
);
814 if (task
->thread
.gsindex
== GS_TLS_SEL
)
815 base
= read_32bit_tls(task
, GS_TLS
);
817 savesegment(gs
, gsindex
);
819 rdmsrl(MSR_KERNEL_GS_BASE
, base
);
821 base
= task
->thread
.gs
;
823 base
= task
->thread
.gs
;
824 ret
= put_user(base
, (unsigned long __user
*)addr
);
836 long sys_arch_prctl(int code
, unsigned long addr
)
838 return do_arch_prctl(current
, code
, addr
);
841 unsigned long arch_align_stack(unsigned long sp
)
843 if (!(current
->personality
& ADDR_NO_RANDOMIZE
) && randomize_va_space
)
844 sp
-= get_random_int() % 8192;
848 unsigned long arch_randomize_brk(struct mm_struct
*mm
)
850 unsigned long range_end
= mm
->brk
+ 0x02000000;
851 return randomize_range(mm
->brk
, range_end
, 0) ? : mm
->brk
;