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..
17 #include <linux/cpu.h>
18 #include <linux/errno.h>
19 #include <linux/sched.h>
21 #include <linux/kernel.h>
23 #include <linux/elfcore.h>
24 #include <linux/smp.h>
25 #include <linux/slab.h>
26 #include <linux/user.h>
27 #include <linux/interrupt.h>
28 #include <linux/delay.h>
29 #include <linux/module.h>
30 #include <linux/ptrace.h>
31 #include <linux/notifier.h>
32 #include <linux/kprobes.h>
33 #include <linux/kdebug.h>
34 #include <linux/prctl.h>
35 #include <linux/uaccess.h>
37 #include <linux/ftrace.h>
39 #include <asm/pgtable.h>
40 #include <asm/processor.h>
42 #include <asm/fpu-internal.h>
43 #include <asm/mmu_context.h>
44 #include <asm/prctl.h>
46 #include <asm/proto.h>
49 #include <asm/syscalls.h>
50 #include <asm/debugreg.h>
51 #include <asm/switch_to.h>
53 asmlinkage
extern void ret_from_fork(void);
55 __visible
DEFINE_PER_CPU(unsigned long, rsp_scratch
);
57 /* Prints also some state that isn't saved in the pt_regs */
58 void __show_regs(struct pt_regs
*regs
, int all
)
60 unsigned long cr0
= 0L, cr2
= 0L, cr3
= 0L, cr4
= 0L, fs
, gs
, shadowgs
;
61 unsigned long d0
, d1
, d2
, d3
, d6
, d7
;
62 unsigned int fsindex
, gsindex
;
63 unsigned int ds
, cs
, es
;
65 printk(KERN_DEFAULT
"RIP: %04lx:[<%016lx>] ", regs
->cs
& 0xffff, regs
->ip
);
66 printk_address(regs
->ip
);
67 printk(KERN_DEFAULT
"RSP: %04lx:%016lx EFLAGS: %08lx\n", regs
->ss
,
68 regs
->sp
, regs
->flags
);
69 printk(KERN_DEFAULT
"RAX: %016lx RBX: %016lx RCX: %016lx\n",
70 regs
->ax
, regs
->bx
, regs
->cx
);
71 printk(KERN_DEFAULT
"RDX: %016lx RSI: %016lx RDI: %016lx\n",
72 regs
->dx
, regs
->si
, regs
->di
);
73 printk(KERN_DEFAULT
"RBP: %016lx R08: %016lx R09: %016lx\n",
74 regs
->bp
, regs
->r8
, regs
->r9
);
75 printk(KERN_DEFAULT
"R10: %016lx R11: %016lx R12: %016lx\n",
76 regs
->r10
, regs
->r11
, regs
->r12
);
77 printk(KERN_DEFAULT
"R13: %016lx R14: %016lx R15: %016lx\n",
78 regs
->r13
, regs
->r14
, regs
->r15
);
80 asm("movl %%ds,%0" : "=r" (ds
));
81 asm("movl %%cs,%0" : "=r" (cs
));
82 asm("movl %%es,%0" : "=r" (es
));
83 asm("movl %%fs,%0" : "=r" (fsindex
));
84 asm("movl %%gs,%0" : "=r" (gsindex
));
86 rdmsrl(MSR_FS_BASE
, fs
);
87 rdmsrl(MSR_GS_BASE
, gs
);
88 rdmsrl(MSR_KERNEL_GS_BASE
, shadowgs
);
98 printk(KERN_DEFAULT
"FS: %016lx(%04x) GS:%016lx(%04x) knlGS:%016lx\n",
99 fs
, fsindex
, gs
, gsindex
, shadowgs
);
100 printk(KERN_DEFAULT
"CS: %04x DS: %04x ES: %04x CR0: %016lx\n", cs
, ds
,
102 printk(KERN_DEFAULT
"CR2: %016lx CR3: %016lx CR4: %016lx\n", cr2
, cr3
,
112 /* Only print out debug registers if they are in their non-default state. */
113 if ((d0
== 0) && (d1
== 0) && (d2
== 0) && (d3
== 0) &&
114 (d6
== DR6_RESERVED
) && (d7
== 0x400))
117 printk(KERN_DEFAULT
"DR0: %016lx DR1: %016lx DR2: %016lx\n", d0
, d1
, d2
);
118 printk(KERN_DEFAULT
"DR3: %016lx DR6: %016lx DR7: %016lx\n", d3
, d6
, d7
);
122 void release_thread(struct task_struct
*dead_task
)
125 if (dead_task
->mm
->context
.size
) {
126 pr_warn("WARNING: dead process %s still has LDT? <%p/%d>\n",
128 dead_task
->mm
->context
.ldt
,
129 dead_task
->mm
->context
.size
);
135 static inline void set_32bit_tls(struct task_struct
*t
, int tls
, u32 addr
)
137 struct user_desc ud
= {
144 struct desc_struct
*desc
= t
->thread
.tls_array
;
149 static inline u32
read_32bit_tls(struct task_struct
*t
, int tls
)
151 return get_desc_base(&t
->thread
.tls_array
[tls
]);
154 int copy_thread(unsigned long clone_flags
, unsigned long sp
,
155 unsigned long arg
, struct task_struct
*p
)
158 struct pt_regs
*childregs
;
159 struct task_struct
*me
= current
;
161 p
->thread
.sp0
= (unsigned long)task_stack_page(p
) + THREAD_SIZE
;
162 childregs
= task_pt_regs(p
);
163 p
->thread
.sp
= (unsigned long) childregs
;
164 set_tsk_thread_flag(p
, TIF_FORK
);
165 p
->thread
.io_bitmap_ptr
= NULL
;
167 savesegment(gs
, p
->thread
.gsindex
);
168 p
->thread
.gs
= p
->thread
.gsindex
? 0 : me
->thread
.gs
;
169 savesegment(fs
, p
->thread
.fsindex
);
170 p
->thread
.fs
= p
->thread
.fsindex
? 0 : me
->thread
.fs
;
171 savesegment(es
, p
->thread
.es
);
172 savesegment(ds
, p
->thread
.ds
);
173 memset(p
->thread
.ptrace_bps
, 0, sizeof(p
->thread
.ptrace_bps
));
175 if (unlikely(p
->flags
& PF_KTHREAD
)) {
177 memset(childregs
, 0, sizeof(struct pt_regs
));
178 childregs
->sp
= (unsigned long)childregs
;
179 childregs
->ss
= __KERNEL_DS
;
180 childregs
->bx
= sp
; /* function */
182 childregs
->orig_ax
= -1;
183 childregs
->cs
= __KERNEL_CS
| get_kernel_rpl();
184 childregs
->flags
= X86_EFLAGS_IF
| X86_EFLAGS_FIXED
;
187 *childregs
= *current_pt_regs();
194 if (unlikely(test_tsk_thread_flag(me
, TIF_IO_BITMAP
))) {
195 p
->thread
.io_bitmap_ptr
= kmemdup(me
->thread
.io_bitmap_ptr
,
196 IO_BITMAP_BYTES
, GFP_KERNEL
);
197 if (!p
->thread
.io_bitmap_ptr
) {
198 p
->thread
.io_bitmap_max
= 0;
201 set_tsk_thread_flag(p
, TIF_IO_BITMAP
);
205 * Set a new TLS for the child thread?
207 if (clone_flags
& CLONE_SETTLS
) {
208 #ifdef CONFIG_IA32_EMULATION
210 err
= do_set_thread_area(p
, -1,
211 (struct user_desc __user
*)childregs
->si
, 0);
214 err
= do_arch_prctl(p
, ARCH_SET_FS
, childregs
->r8
);
220 if (err
&& p
->thread
.io_bitmap_ptr
) {
221 kfree(p
->thread
.io_bitmap_ptr
);
222 p
->thread
.io_bitmap_max
= 0;
229 start_thread_common(struct pt_regs
*regs
, unsigned long new_ip
,
230 unsigned long new_sp
,
231 unsigned int _cs
, unsigned int _ss
, unsigned int _ds
)
234 loadsegment(es
, _ds
);
235 loadsegment(ds
, _ds
);
241 regs
->flags
= X86_EFLAGS_IF
;
246 start_thread(struct pt_regs
*regs
, unsigned long new_ip
, unsigned long new_sp
)
248 start_thread_common(regs
, new_ip
, new_sp
,
249 __USER_CS
, __USER_DS
, 0);
252 #ifdef CONFIG_IA32_EMULATION
253 void start_thread_ia32(struct pt_regs
*regs
, u32 new_ip
, u32 new_sp
)
255 start_thread_common(regs
, new_ip
, new_sp
,
256 test_thread_flag(TIF_X32
)
257 ? __USER_CS
: __USER32_CS
,
258 __USER_DS
, __USER_DS
);
263 * switch_to(x,y) should switch tasks from x to y.
265 * This could still be optimized:
266 * - fold all the options into a flag word and test it with a single test.
267 * - could test fs/gs bitsliced
269 * Kprobes not supported here. Set the probe on schedule instead.
270 * Function graph tracer not supported too.
272 __visible __notrace_funcgraph
struct task_struct
*
273 __switch_to(struct task_struct
*prev_p
, struct task_struct
*next_p
)
275 struct thread_struct
*prev
= &prev_p
->thread
;
276 struct thread_struct
*next
= &next_p
->thread
;
277 int cpu
= smp_processor_id();
278 struct tss_struct
*tss
= &per_cpu(cpu_tss
, cpu
);
279 unsigned fsindex
, gsindex
;
282 fpu
= switch_fpu_prepare(prev_p
, next_p
, cpu
);
284 /* We must save %fs and %gs before load_TLS() because
285 * %fs and %gs may be cleared by load_TLS().
287 * (e.g. xen_load_tls())
289 savesegment(fs
, fsindex
);
290 savesegment(gs
, gsindex
);
293 * Load TLS before restoring any segments so that segment loads
294 * reference the correct GDT entries.
299 * Leave lazy mode, flushing any hypercalls made here. This
300 * must be done after loading TLS entries in the GDT but before
301 * loading segments that might reference them, and and it must
302 * be done before math_state_restore, so the TS bit is up to
305 arch_end_context_switch(next_p
);
309 * Reading them only returns the selectors, but writing them (if
310 * nonzero) loads the full descriptor from the GDT or LDT. The
311 * LDT for next is loaded in switch_mm, and the GDT is loaded
314 * We therefore need to write new values to the segment
315 * registers on every context switch unless both the new and old
318 * Note that we don't need to do anything for CS and SS, as
319 * those are saved and restored as part of pt_regs.
321 savesegment(es
, prev
->es
);
322 if (unlikely(next
->es
| prev
->es
))
323 loadsegment(es
, next
->es
);
325 savesegment(ds
, prev
->ds
);
326 if (unlikely(next
->ds
| prev
->ds
))
327 loadsegment(ds
, next
->ds
);
332 * These are even more complicated than FS and GS: they have
333 * 64-bit bases are that controlled by arch_prctl. Those bases
334 * only differ from the values in the GDT or LDT if the selector
337 * Loading the segment register resets the hidden base part of
338 * the register to 0 or the value from the GDT / LDT. If the
339 * next base address zero, writing 0 to the segment register is
340 * much faster than using wrmsr to explicitly zero the base.
342 * The thread_struct.fs and thread_struct.gs values are 0
343 * if the fs and gs bases respectively are not overridden
344 * from the values implied by fsindex and gsindex. They
345 * are nonzero, and store the nonzero base addresses, if
346 * the bases are overridden.
348 * (fs != 0 && fsindex != 0) || (gs != 0 && gsindex != 0) should
351 * Therefore we need to reload the segment registers if either
352 * the old or new selector is nonzero, and we need to override
353 * the base address if next thread expects it to be overridden.
355 * This code is unnecessarily slow in the case where the old and
356 * new indexes are zero and the new base is nonzero -- it will
357 * unnecessarily write 0 to the selector before writing the new
360 * Note: This all depends on arch_prctl being the only way that
361 * user code can override the segment base. Once wrfsbase and
362 * wrgsbase are enabled, most of this code will need to change.
364 if (unlikely(fsindex
| next
->fsindex
| prev
->fs
)) {
365 loadsegment(fs
, next
->fsindex
);
368 * If user code wrote a nonzero value to FS, then it also
369 * cleared the overridden base address.
371 * XXX: if user code wrote 0 to FS and cleared the base
372 * address itself, we won't notice and we'll incorrectly
373 * restore the prior base address next time we reschdule
380 wrmsrl(MSR_FS_BASE
, next
->fs
);
381 prev
->fsindex
= fsindex
;
383 if (unlikely(gsindex
| next
->gsindex
| prev
->gs
)) {
384 load_gs_index(next
->gsindex
);
386 /* This works (and fails) the same way as fsindex above. */
391 wrmsrl(MSR_KERNEL_GS_BASE
, next
->gs
);
392 prev
->gsindex
= gsindex
;
394 switch_fpu_finish(next_p
, fpu
);
397 * Switch the PDA and FPU contexts.
399 this_cpu_write(current_task
, next_p
);
402 * If it were not for PREEMPT_ACTIVE we could guarantee that the
403 * preempt_count of all tasks was equal here and this would not be
406 task_thread_info(prev_p
)->saved_preempt_count
= this_cpu_read(__preempt_count
);
407 this_cpu_write(__preempt_count
, task_thread_info(next_p
)->saved_preempt_count
);
409 /* Reload esp0 and ss1. This changes current_thread_info(). */
412 this_cpu_write(kernel_stack
,
413 (unsigned long)task_stack_page(next_p
) + THREAD_SIZE
);
416 * Now maybe reload the debug registers and handle I/O bitmaps
418 if (unlikely(task_thread_info(next_p
)->flags
& _TIF_WORK_CTXSW_NEXT
||
419 task_thread_info(prev_p
)->flags
& _TIF_WORK_CTXSW_PREV
))
420 __switch_to_xtra(prev_p
, next_p
, tss
);
425 void set_personality_64bit(void)
427 /* inherit personality from parent */
429 /* Make sure to be in 64bit mode */
430 clear_thread_flag(TIF_IA32
);
431 clear_thread_flag(TIF_ADDR32
);
432 clear_thread_flag(TIF_X32
);
434 /* Ensure the corresponding mm is not marked. */
436 current
->mm
->context
.ia32_compat
= 0;
438 /* TBD: overwrites user setup. Should have two bits.
439 But 64bit processes have always behaved this way,
440 so it's not too bad. The main problem is just that
441 32bit childs are affected again. */
442 current
->personality
&= ~READ_IMPLIES_EXEC
;
445 void set_personality_ia32(bool x32
)
447 /* inherit personality from parent */
449 /* Make sure to be in 32bit mode */
450 set_thread_flag(TIF_ADDR32
);
452 /* Mark the associated mm as containing 32-bit tasks. */
454 clear_thread_flag(TIF_IA32
);
455 set_thread_flag(TIF_X32
);
457 current
->mm
->context
.ia32_compat
= TIF_X32
;
458 current
->personality
&= ~READ_IMPLIES_EXEC
;
459 /* is_compat_task() uses the presence of the x32
460 syscall bit flag to determine compat status */
461 current_thread_info()->status
&= ~TS_COMPAT
;
463 set_thread_flag(TIF_IA32
);
464 clear_thread_flag(TIF_X32
);
466 current
->mm
->context
.ia32_compat
= TIF_IA32
;
467 current
->personality
|= force_personality32
;
468 /* Prepare the first "return" to user space */
469 current_thread_info()->status
|= TS_COMPAT
;
472 EXPORT_SYMBOL_GPL(set_personality_ia32
);
474 unsigned long get_wchan(struct task_struct
*p
)
480 if (!p
|| p
== current
|| p
->state
== TASK_RUNNING
)
482 stack
= (unsigned long)task_stack_page(p
);
483 if (p
->thread
.sp
< stack
|| p
->thread
.sp
>= stack
+THREAD_SIZE
)
485 fp
= *(u64
*)(p
->thread
.sp
);
487 if (fp
< (unsigned long)stack
||
488 fp
>= (unsigned long)stack
+THREAD_SIZE
)
491 if (!in_sched_functions(ip
))
494 } while (count
++ < 16);
498 long do_arch_prctl(struct task_struct
*task
, int code
, unsigned long addr
)
501 int doit
= task
== current
;
506 if (addr
>= TASK_SIZE_OF(task
))
509 /* handle small bases via the GDT because that's faster to
511 if (addr
<= 0xffffffff) {
512 set_32bit_tls(task
, GS_TLS
, addr
);
514 load_TLS(&task
->thread
, cpu
);
515 load_gs_index(GS_TLS_SEL
);
517 task
->thread
.gsindex
= GS_TLS_SEL
;
520 task
->thread
.gsindex
= 0;
521 task
->thread
.gs
= addr
;
524 ret
= wrmsrl_safe(MSR_KERNEL_GS_BASE
, addr
);
530 /* Not strictly needed for fs, but do it for symmetry
532 if (addr
>= TASK_SIZE_OF(task
))
535 /* handle small bases via the GDT because that's faster to
537 if (addr
<= 0xffffffff) {
538 set_32bit_tls(task
, FS_TLS
, addr
);
540 load_TLS(&task
->thread
, cpu
);
541 loadsegment(fs
, FS_TLS_SEL
);
543 task
->thread
.fsindex
= FS_TLS_SEL
;
546 task
->thread
.fsindex
= 0;
547 task
->thread
.fs
= addr
;
549 /* set the selector to 0 to not confuse
552 ret
= wrmsrl_safe(MSR_FS_BASE
, addr
);
559 if (task
->thread
.fsindex
== FS_TLS_SEL
)
560 base
= read_32bit_tls(task
, FS_TLS
);
562 rdmsrl(MSR_FS_BASE
, base
);
564 base
= task
->thread
.fs
;
565 ret
= put_user(base
, (unsigned long __user
*)addr
);
571 if (task
->thread
.gsindex
== GS_TLS_SEL
)
572 base
= read_32bit_tls(task
, GS_TLS
);
574 savesegment(gs
, gsindex
);
576 rdmsrl(MSR_KERNEL_GS_BASE
, base
);
578 base
= task
->thread
.gs
;
580 base
= task
->thread
.gs
;
581 ret
= put_user(base
, (unsigned long __user
*)addr
);
593 long sys_arch_prctl(int code
, unsigned long addr
)
595 return do_arch_prctl(current
, code
, addr
);
598 unsigned long KSTK_ESP(struct task_struct
*task
)
600 return task_pt_regs(task
)->sp
;