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, old_rsp);
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 p->thread.usersp = me->thread.usersp;
165 set_tsk_thread_flag(p, TIF_FORK);
166 p->thread.io_bitmap_ptr = NULL;
168 savesegment(gs, p->thread.gsindex);
169 p->thread.gs = p->thread.gsindex ? 0 : me->thread.gs;
170 savesegment(fs, p->thread.fsindex);
171 p->thread.fs = p->thread.fsindex ? 0 : me->thread.fs;
172 savesegment(es, p->thread.es);
173 savesegment(ds, p->thread.ds);
174 memset(p->thread.ptrace_bps, 0, sizeof(p->thread.ptrace_bps));
176 if (unlikely(p->flags & PF_KTHREAD)) {
178 memset(childregs, 0, sizeof(struct pt_regs));
179 childregs->sp = (unsigned long)childregs;
180 childregs->ss = __KERNEL_DS;
181 childregs->bx = sp; /* function */
183 childregs->orig_ax = -1;
184 childregs->cs = __KERNEL_CS | get_kernel_rpl();
185 childregs->flags = X86_EFLAGS_IF | X86_EFLAGS_FIXED;
188 *childregs = *current_pt_regs();
195 memset(p->thread.ptrace_bps, 0, sizeof(p->thread.ptrace_bps));
197 if (unlikely(test_tsk_thread_flag(me, TIF_IO_BITMAP))) {
198 p->thread.io_bitmap_ptr = kmemdup(me->thread.io_bitmap_ptr,
199 IO_BITMAP_BYTES, GFP_KERNEL);
200 if (!p->thread.io_bitmap_ptr) {
201 p->thread.io_bitmap_max = 0;
204 set_tsk_thread_flag(p, TIF_IO_BITMAP);
208 * Set a new TLS for the child thread?
210 if (clone_flags & CLONE_SETTLS) {
211 #ifdef CONFIG_IA32_EMULATION
212 if (test_thread_flag(TIF_IA32))
213 err = do_set_thread_area(p, -1,
214 (struct user_desc __user *)childregs->si, 0);
217 err = do_arch_prctl(p, ARCH_SET_FS, childregs->r8);
223 if (err && p->thread.io_bitmap_ptr) {
224 kfree(p->thread.io_bitmap_ptr);
225 p->thread.io_bitmap_max = 0;
232 start_thread_common(struct pt_regs *regs, unsigned long new_ip,
233 unsigned long new_sp,
234 unsigned int _cs, unsigned int _ss, unsigned int _ds)
237 loadsegment(es, _ds);
238 loadsegment(ds, _ds);
240 current->thread.usersp = new_sp;
243 this_cpu_write(old_rsp, new_sp);
246 regs->flags = X86_EFLAGS_IF;
250 start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp)
252 start_thread_common(regs, new_ip, new_sp,
253 __USER_CS, __USER_DS, 0);
256 #ifdef CONFIG_IA32_EMULATION
257 void start_thread_ia32(struct pt_regs *regs, u32 new_ip, u32 new_sp)
259 start_thread_common(regs, new_ip, new_sp,
260 test_thread_flag(TIF_X32)
261 ? __USER_CS : __USER32_CS,
262 __USER_DS, __USER_DS);
267 * switch_to(x,y) should switch tasks from x to y.
269 * This could still be optimized:
270 * - fold all the options into a flag word and test it with a single test.
271 * - could test fs/gs bitsliced
273 * Kprobes not supported here. Set the probe on schedule instead.
274 * Function graph tracer not supported too.
276 __visible __notrace_funcgraph struct task_struct *
277 __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
279 struct thread_struct *prev = &prev_p->thread;
280 struct thread_struct *next = &next_p->thread;
281 int cpu = smp_processor_id();
282 struct tss_struct *tss = &per_cpu(init_tss, cpu);
283 unsigned fsindex, gsindex;
286 fpu = switch_fpu_prepare(prev_p, next_p, cpu);
289 * Reload esp0, LDT and the page table pointer:
295 * This won't pick up thread selector changes, but I guess that is ok.
297 savesegment(es, prev->es);
298 if (unlikely(next->es | prev->es))
299 loadsegment(es, next->es);
301 savesegment(ds, prev->ds);
302 if (unlikely(next->ds | prev->ds))
303 loadsegment(ds, next->ds);
306 /* We must save %fs and %gs before load_TLS() because
307 * %fs and %gs may be cleared by load_TLS().
309 * (e.g. xen_load_tls())
311 savesegment(fs, fsindex);
312 savesegment(gs, gsindex);
317 * Leave lazy mode, flushing any hypercalls made here.
318 * This must be done before restoring TLS segments so
319 * the GDT and LDT are properly updated, and must be
320 * done before math_state_restore, so the TS bit is up
323 arch_end_context_switch(next_p);
328 * Segment register != 0 always requires a reload. Also
329 * reload when it has changed. When prev process used 64bit
330 * base always reload to avoid an information leak.
332 if (unlikely(fsindex | next->fsindex | prev->fs)) {
333 loadsegment(fs, next->fsindex);
335 * Check if the user used a selector != 0; if yes
336 * clear 64bit base, since overloaded base is always
337 * mapped to the Null selector
342 /* when next process has a 64bit base use it */
344 wrmsrl(MSR_FS_BASE, next->fs);
345 prev->fsindex = fsindex;
347 if (unlikely(gsindex | next->gsindex | prev->gs)) {
348 load_gs_index(next->gsindex);
353 wrmsrl(MSR_KERNEL_GS_BASE, next->gs);
354 prev->gsindex = gsindex;
356 switch_fpu_finish(next_p, fpu);
359 * Switch the PDA and FPU contexts.
361 prev->usersp = this_cpu_read(old_rsp);
362 this_cpu_write(old_rsp, next->usersp);
363 this_cpu_write(current_task, next_p);
366 * If it were not for PREEMPT_ACTIVE we could guarantee that the
367 * preempt_count of all tasks was equal here and this would not be
370 task_thread_info(prev_p)->saved_preempt_count = this_cpu_read(__preempt_count);
371 this_cpu_write(__preempt_count, task_thread_info(next_p)->saved_preempt_count);
373 this_cpu_write(kernel_stack,
374 (unsigned long)task_stack_page(next_p) +
375 THREAD_SIZE - KERNEL_STACK_OFFSET);
378 * Now maybe reload the debug registers and handle I/O bitmaps
380 if (unlikely(task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT ||
381 task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV))
382 __switch_to_xtra(prev_p, next_p, tss);
387 void set_personality_64bit(void)
389 /* inherit personality from parent */
391 /* Make sure to be in 64bit mode */
392 clear_thread_flag(TIF_IA32);
393 clear_thread_flag(TIF_ADDR32);
394 clear_thread_flag(TIF_X32);
396 /* Ensure the corresponding mm is not marked. */
398 current->mm->context.ia32_compat = 0;
400 /* TBD: overwrites user setup. Should have two bits.
401 But 64bit processes have always behaved this way,
402 so it's not too bad. The main problem is just that
403 32bit childs are affected again. */
404 current->personality &= ~READ_IMPLIES_EXEC;
407 void set_personality_ia32(bool x32)
409 /* inherit personality from parent */
411 /* Make sure to be in 32bit mode */
412 set_thread_flag(TIF_ADDR32);
414 /* Mark the associated mm as containing 32-bit tasks. */
416 clear_thread_flag(TIF_IA32);
417 set_thread_flag(TIF_X32);
419 current->mm->context.ia32_compat = TIF_X32;
420 current->personality &= ~READ_IMPLIES_EXEC;
421 /* is_compat_task() uses the presence of the x32
422 syscall bit flag to determine compat status */
423 current_thread_info()->status &= ~TS_COMPAT;
425 set_thread_flag(TIF_IA32);
426 clear_thread_flag(TIF_X32);
428 current->mm->context.ia32_compat = TIF_IA32;
429 current->personality |= force_personality32;
430 /* Prepare the first "return" to user space */
431 current_thread_info()->status |= TS_COMPAT;
434 EXPORT_SYMBOL_GPL(set_personality_ia32);
436 unsigned long get_wchan(struct task_struct *p)
442 if (!p || p == current || p->state == TASK_RUNNING)
444 stack = (unsigned long)task_stack_page(p);
445 if (p->thread.sp < stack || p->thread.sp >= stack+THREAD_SIZE)
447 fp = *(u64 *)(p->thread.sp);
449 if (fp < (unsigned long)stack ||
450 fp >= (unsigned long)stack+THREAD_SIZE)
453 if (!in_sched_functions(ip))
456 } while (count++ < 16);
460 long do_arch_prctl(struct task_struct *task, int code, unsigned long addr)
463 int doit = task == current;
468 if (addr >= TASK_SIZE_OF(task))
471 /* handle small bases via the GDT because that's faster to
473 if (addr <= 0xffffffff) {
474 set_32bit_tls(task, GS_TLS, addr);
476 load_TLS(&task->thread, cpu);
477 load_gs_index(GS_TLS_SEL);
479 task->thread.gsindex = GS_TLS_SEL;
482 task->thread.gsindex = 0;
483 task->thread.gs = addr;
486 ret = wrmsrl_safe(MSR_KERNEL_GS_BASE, addr);
492 /* Not strictly needed for fs, but do it for symmetry
494 if (addr >= TASK_SIZE_OF(task))
497 /* handle small bases via the GDT because that's faster to
499 if (addr <= 0xffffffff) {
500 set_32bit_tls(task, FS_TLS, addr);
502 load_TLS(&task->thread, cpu);
503 loadsegment(fs, FS_TLS_SEL);
505 task->thread.fsindex = FS_TLS_SEL;
508 task->thread.fsindex = 0;
509 task->thread.fs = addr;
511 /* set the selector to 0 to not confuse
514 ret = wrmsrl_safe(MSR_FS_BASE, addr);
521 if (task->thread.fsindex == FS_TLS_SEL)
522 base = read_32bit_tls(task, FS_TLS);
524 rdmsrl(MSR_FS_BASE, base);
526 base = task->thread.fs;
527 ret = put_user(base, (unsigned long __user *)addr);
533 if (task->thread.gsindex == GS_TLS_SEL)
534 base = read_32bit_tls(task, GS_TLS);
536 savesegment(gs, gsindex);
538 rdmsrl(MSR_KERNEL_GS_BASE, base);
540 base = task->thread.gs;
542 base = task->thread.gs;
543 ret = put_user(base, (unsigned long __user *)addr);
555 long sys_arch_prctl(int code, unsigned long addr)
557 return do_arch_prctl(current, code, addr);
560 unsigned long KSTK_ESP(struct task_struct *task)
562 return (test_tsk_thread_flag(task, TIF_IA32)) ?
563 (task_pt_regs(task)->sp) : ((task)->thread.usersp);