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 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;
66 printk(KERN_DEFAULT "RIP: %04lx:[<%016lx>] ", regs->cs & 0xffff, regs->ip);
67 printk_address(regs->ip, 1);
68 printk(KERN_DEFAULT "RSP: %04lx:%016lx EFLAGS: %08lx\n", regs->ss,
69 regs->sp, regs->flags);
70 printk(KERN_DEFAULT "RAX: %016lx RBX: %016lx RCX: %016lx\n",
71 regs->ax, regs->bx, regs->cx);
72 printk(KERN_DEFAULT "RDX: %016lx RSI: %016lx RDI: %016lx\n",
73 regs->dx, regs->si, regs->di);
74 printk(KERN_DEFAULT "RBP: %016lx R08: %016lx R09: %016lx\n",
75 regs->bp, regs->r8, regs->r9);
76 printk(KERN_DEFAULT "R10: %016lx R11: %016lx R12: %016lx\n",
77 regs->r10, regs->r11, regs->r12);
78 printk(KERN_DEFAULT "R13: %016lx R14: %016lx R15: %016lx\n",
79 regs->r13, regs->r14, regs->r15);
81 asm("movl %%ds,%0" : "=r" (ds));
82 asm("movl %%cs,%0" : "=r" (cs));
83 asm("movl %%es,%0" : "=r" (es));
84 asm("movl %%fs,%0" : "=r" (fsindex));
85 asm("movl %%gs,%0" : "=r" (gsindex));
87 rdmsrl(MSR_FS_BASE, fs);
88 rdmsrl(MSR_GS_BASE, gs);
89 rdmsrl(MSR_KERNEL_GS_BASE, shadowgs);
99 printk(KERN_DEFAULT "FS: %016lx(%04x) GS:%016lx(%04x) knlGS:%016lx\n",
100 fs, fsindex, gs, gsindex, shadowgs);
101 printk(KERN_DEFAULT "CS: %04x DS: %04x ES: %04x CR0: %016lx\n", cs, ds,
103 printk(KERN_DEFAULT "CR2: %016lx CR3: %016lx CR4: %016lx\n", cr2, cr3,
109 printk(KERN_DEFAULT "DR0: %016lx DR1: %016lx DR2: %016lx\n", d0, d1, d2);
113 printk(KERN_DEFAULT "DR3: %016lx DR6: %016lx DR7: %016lx\n", d3, d6, d7);
116 void release_thread(struct task_struct *dead_task)
119 if (dead_task->mm->context.size) {
120 printk("WARNING: dead process %8s still has LDT? <%p/%d>\n",
122 dead_task->mm->context.ldt,
123 dead_task->mm->context.size);
129 static inline void set_32bit_tls(struct task_struct *t, int tls, u32 addr)
131 struct user_desc ud = {
138 struct desc_struct *desc = t->thread.tls_array;
143 static inline u32 read_32bit_tls(struct task_struct *t, int tls)
145 return get_desc_base(&t->thread.tls_array[tls]);
149 * This gets called before we allocate a new thread and copy
150 * the current task into it.
152 void prepare_to_copy(struct task_struct *tsk)
157 int copy_thread(unsigned long clone_flags, unsigned long sp,
158 unsigned long unused,
159 struct task_struct *p, struct pt_regs *regs)
162 struct pt_regs *childregs;
163 struct task_struct *me = current;
165 childregs = ((struct pt_regs *)
166 (THREAD_SIZE + task_stack_page(p))) - 1;
173 childregs->sp = (unsigned long)childregs;
175 p->thread.sp = (unsigned long) childregs;
176 p->thread.sp0 = (unsigned long) (childregs+1);
177 p->thread.usersp = me->thread.usersp;
179 set_tsk_thread_flag(p, TIF_FORK);
182 p->thread.io_bitmap_ptr = NULL;
184 savesegment(gs, p->thread.gsindex);
185 p->thread.gs = p->thread.gsindex ? 0 : me->thread.gs;
186 savesegment(fs, p->thread.fsindex);
187 p->thread.fs = p->thread.fsindex ? 0 : me->thread.fs;
188 savesegment(es, p->thread.es);
189 savesegment(ds, p->thread.ds);
192 memset(p->thread.ptrace_bps, 0, sizeof(p->thread.ptrace_bps));
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
209 if (test_thread_flag(TIF_IA32))
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);
237 current->thread.usersp = new_sp;
240 percpu_write(old_rsp, new_sp);
243 regs->flags = X86_EFLAGS_IF;
245 * Free the old FP and other extended state
247 free_thread_xstate(current);
251 start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp)
253 start_thread_common(regs, new_ip, new_sp,
254 __USER_CS, __USER_DS, 0);
257 #ifdef CONFIG_IA32_EMULATION
258 void start_thread_ia32(struct pt_regs *regs, u32 new_ip, u32 new_sp)
260 start_thread_common(regs, new_ip, new_sp,
261 test_thread_flag(TIF_X32)
262 ? __USER_CS : __USER32_CS,
263 __USER_DS, __USER_DS);
268 * switch_to(x,y) should switch tasks from x to y.
270 * This could still be optimized:
271 * - fold all the options into a flag word and test it with a single test.
272 * - could test fs/gs bitsliced
274 * Kprobes not supported here. Set the probe on schedule instead.
275 * Function graph tracer not supported too.
277 __notrace_funcgraph struct task_struct *
278 __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
280 struct thread_struct *prev = &prev_p->thread;
281 struct thread_struct *next = &next_p->thread;
282 int cpu = smp_processor_id();
283 struct tss_struct *tss = &per_cpu(init_tss, cpu);
284 unsigned fsindex, gsindex;
287 fpu = switch_fpu_prepare(prev_p, next_p, cpu);
290 * Reload esp0, LDT and the page table pointer:
296 * This won't pick up thread selector changes, but I guess that is ok.
298 savesegment(es, prev->es);
299 if (unlikely(next->es | prev->es))
300 loadsegment(es, next->es);
302 savesegment(ds, prev->ds);
303 if (unlikely(next->ds | prev->ds))
304 loadsegment(ds, next->ds);
307 /* We must save %fs and %gs before load_TLS() because
308 * %fs and %gs may be cleared by load_TLS().
310 * (e.g. xen_load_tls())
312 savesegment(fs, fsindex);
313 savesegment(gs, gsindex);
318 * Leave lazy mode, flushing any hypercalls made here.
319 * This must be done before restoring TLS segments so
320 * the GDT and LDT are properly updated, and must be
321 * done before math_state_restore, so the TS bit is up
324 arch_end_context_switch(next_p);
329 * Segment register != 0 always requires a reload. Also
330 * reload when it has changed. When prev process used 64bit
331 * base always reload to avoid an information leak.
333 if (unlikely(fsindex | next->fsindex | prev->fs)) {
334 loadsegment(fs, next->fsindex);
336 * Check if the user used a selector != 0; if yes
337 * clear 64bit base, since overloaded base is always
338 * mapped to the Null selector
343 /* when next process has a 64bit base use it */
345 wrmsrl(MSR_FS_BASE, next->fs);
346 prev->fsindex = fsindex;
348 if (unlikely(gsindex | next->gsindex | prev->gs)) {
349 load_gs_index(next->gsindex);
354 wrmsrl(MSR_KERNEL_GS_BASE, next->gs);
355 prev->gsindex = gsindex;
357 switch_fpu_finish(next_p, fpu);
360 * Switch the PDA and FPU contexts.
362 prev->usersp = percpu_read(old_rsp);
363 percpu_write(old_rsp, next->usersp);
364 percpu_write(current_task, next_p);
366 percpu_write(kernel_stack,
367 (unsigned long)task_stack_page(next_p) +
368 THREAD_SIZE - KERNEL_STACK_OFFSET);
371 * Now maybe reload the debug registers and handle I/O bitmaps
373 if (unlikely(task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT ||
374 task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV))
375 __switch_to_xtra(prev_p, next_p, tss);
380 void set_personality_64bit(void)
382 /* inherit personality from parent */
384 /* Make sure to be in 64bit mode */
385 clear_thread_flag(TIF_IA32);
386 clear_thread_flag(TIF_ADDR32);
387 clear_thread_flag(TIF_X32);
389 /* Ensure the corresponding mm is not marked. */
391 current->mm->context.ia32_compat = 0;
393 /* TBD: overwrites user setup. Should have two bits.
394 But 64bit processes have always behaved this way,
395 so it's not too bad. The main problem is just that
396 32bit childs are affected again. */
397 current->personality &= ~READ_IMPLIES_EXEC;
400 void set_personality_ia32(bool x32)
402 /* inherit personality from parent */
404 /* Make sure to be in 32bit mode */
405 set_thread_flag(TIF_ADDR32);
407 /* Mark the associated mm as containing 32-bit tasks. */
409 current->mm->context.ia32_compat = 1;
412 clear_thread_flag(TIF_IA32);
413 set_thread_flag(TIF_X32);
414 current->personality &= ~READ_IMPLIES_EXEC;
415 /* is_compat_task() uses the presence of the x32
416 syscall bit flag to determine compat status */
417 current_thread_info()->status &= ~TS_COMPAT;
419 set_thread_flag(TIF_IA32);
420 clear_thread_flag(TIF_X32);
421 current->personality |= force_personality32;
422 /* Prepare the first "return" to user space */
423 current_thread_info()->status |= TS_COMPAT;
426 EXPORT_SYMBOL_GPL(set_personality_ia32);
428 unsigned long get_wchan(struct task_struct *p)
434 if (!p || p == current || p->state == TASK_RUNNING)
436 stack = (unsigned long)task_stack_page(p);
437 if (p->thread.sp < stack || p->thread.sp >= stack+THREAD_SIZE)
439 fp = *(u64 *)(p->thread.sp);
441 if (fp < (unsigned long)stack ||
442 fp >= (unsigned long)stack+THREAD_SIZE)
445 if (!in_sched_functions(ip))
448 } while (count++ < 16);
452 long do_arch_prctl(struct task_struct *task, int code, unsigned long addr)
455 int doit = task == current;
460 if (addr >= TASK_SIZE_OF(task))
463 /* handle small bases via the GDT because that's faster to
465 if (addr <= 0xffffffff) {
466 set_32bit_tls(task, GS_TLS, addr);
468 load_TLS(&task->thread, cpu);
469 load_gs_index(GS_TLS_SEL);
471 task->thread.gsindex = GS_TLS_SEL;
474 task->thread.gsindex = 0;
475 task->thread.gs = addr;
478 ret = checking_wrmsrl(MSR_KERNEL_GS_BASE, addr);
484 /* Not strictly needed for fs, but do it for symmetry
486 if (addr >= TASK_SIZE_OF(task))
489 /* handle small bases via the GDT because that's faster to
491 if (addr <= 0xffffffff) {
492 set_32bit_tls(task, FS_TLS, addr);
494 load_TLS(&task->thread, cpu);
495 loadsegment(fs, FS_TLS_SEL);
497 task->thread.fsindex = FS_TLS_SEL;
500 task->thread.fsindex = 0;
501 task->thread.fs = addr;
503 /* set the selector to 0 to not confuse
506 ret = checking_wrmsrl(MSR_FS_BASE, addr);
513 if (task->thread.fsindex == FS_TLS_SEL)
514 base = read_32bit_tls(task, FS_TLS);
516 rdmsrl(MSR_FS_BASE, base);
518 base = task->thread.fs;
519 ret = put_user(base, (unsigned long __user *)addr);
525 if (task->thread.gsindex == GS_TLS_SEL)
526 base = read_32bit_tls(task, GS_TLS);
528 savesegment(gs, gsindex);
530 rdmsrl(MSR_KERNEL_GS_BASE, base);
532 base = task->thread.gs;
534 base = task->thread.gs;
535 ret = put_user(base, (unsigned long __user *)addr);
547 long sys_arch_prctl(int code, unsigned long addr)
549 return do_arch_prctl(current, code, addr);
552 unsigned long KSTK_ESP(struct task_struct *task)
554 return (test_tsk_thread_flag(task, TIF_IA32)) ?
555 (task_pt_regs(task)->sp) : ((task)->thread.usersp);