x86/amd-iommu: Add per IOMMU reference counting
[linux/fpc-iii.git] / arch / x86 / kernel / process_64.c
blobeb62cbcaa490ad553ef5d70b6751a2288d25089e
1 /*
2 * Copyright (C) 1995 Linus Torvalds
4 * Pentium III FXSR, SSE support
5 * Gareth Hughes <gareth@valinux.com>, May 2000
7 * X86-64 port
8 * Andi Kleen.
10 * CPU hotplug support - ashok.raj@intel.com
14 * This file handles the architecture-dependent parts of process handling..
17 #include <linux/stackprotector.h>
18 #include <linux/cpu.h>
19 #include <linux/errno.h>
20 #include <linux/sched.h>
21 #include <linux/fs.h>
22 #include <linux/kernel.h>
23 #include <linux/mm.h>
24 #include <linux/elfcore.h>
25 #include <linux/smp.h>
26 #include <linux/slab.h>
27 #include <linux/user.h>
28 #include <linux/interrupt.h>
29 #include <linux/utsname.h>
30 #include <linux/delay.h>
31 #include <linux/module.h>
32 #include <linux/ptrace.h>
33 #include <linux/notifier.h>
34 #include <linux/kprobes.h>
35 #include <linux/kdebug.h>
36 #include <linux/tick.h>
37 #include <linux/prctl.h>
38 #include <linux/uaccess.h>
39 #include <linux/io.h>
40 #include <linux/ftrace.h>
41 #include <linux/dmi.h>
43 #include <asm/pgtable.h>
44 #include <asm/system.h>
45 #include <asm/processor.h>
46 #include <asm/i387.h>
47 #include <asm/mmu_context.h>
48 #include <asm/prctl.h>
49 #include <asm/desc.h>
50 #include <asm/proto.h>
51 #include <asm/ia32.h>
52 #include <asm/idle.h>
53 #include <asm/syscalls.h>
54 #include <asm/ds.h>
56 asmlinkage extern void ret_from_fork(void);
58 DEFINE_PER_CPU(unsigned long, old_rsp);
59 static DEFINE_PER_CPU(unsigned char, is_idle);
61 unsigned long kernel_thread_flags = CLONE_VM | CLONE_UNTRACED;
63 static ATOMIC_NOTIFIER_HEAD(idle_notifier);
65 void idle_notifier_register(struct notifier_block *n)
67 atomic_notifier_chain_register(&idle_notifier, n);
69 EXPORT_SYMBOL_GPL(idle_notifier_register);
71 void idle_notifier_unregister(struct notifier_block *n)
73 atomic_notifier_chain_unregister(&idle_notifier, n);
75 EXPORT_SYMBOL_GPL(idle_notifier_unregister);
77 void enter_idle(void)
79 percpu_write(is_idle, 1);
80 atomic_notifier_call_chain(&idle_notifier, IDLE_START, NULL);
83 static void __exit_idle(void)
85 if (x86_test_and_clear_bit_percpu(0, is_idle) == 0)
86 return;
87 atomic_notifier_call_chain(&idle_notifier, IDLE_END, NULL);
90 /* Called from interrupts to signify idle end */
91 void exit_idle(void)
93 /* idle loop has pid 0 */
94 if (current->pid)
95 return;
96 __exit_idle();
99 #ifndef CONFIG_SMP
100 static inline void play_dead(void)
102 BUG();
104 #endif
107 * The idle thread. There's no useful work to be
108 * done, so just try to conserve power and have a
109 * low exit latency (ie sit in a loop waiting for
110 * somebody to say that they'd like to reschedule)
112 void cpu_idle(void)
114 current_thread_info()->status |= TS_POLLING;
117 * If we're the non-boot CPU, nothing set the stack canary up
118 * for us. CPU0 already has it initialized but no harm in
119 * doing it again. This is a good place for updating it, as
120 * we wont ever return from this function (so the invalid
121 * canaries already on the stack wont ever trigger).
123 boot_init_stack_canary();
125 /* endless idle loop with no priority at all */
126 while (1) {
127 tick_nohz_stop_sched_tick(1);
128 while (!need_resched()) {
130 rmb();
132 if (cpu_is_offline(smp_processor_id()))
133 play_dead();
135 * Idle routines should keep interrupts disabled
136 * from here on, until they go to idle.
137 * Otherwise, idle callbacks can misfire.
139 local_irq_disable();
140 enter_idle();
141 /* Don't trace irqs off for idle */
142 stop_critical_timings();
143 pm_idle();
144 start_critical_timings();
145 /* In many cases the interrupt that ended idle
146 has already called exit_idle. But some idle
147 loops can be woken up without interrupt. */
148 __exit_idle();
151 tick_nohz_restart_sched_tick();
152 preempt_enable_no_resched();
153 schedule();
154 preempt_disable();
158 /* Prints also some state that isn't saved in the pt_regs */
159 void __show_regs(struct pt_regs *regs, int all)
161 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L, fs, gs, shadowgs;
162 unsigned long d0, d1, d2, d3, d6, d7;
163 unsigned int fsindex, gsindex;
164 unsigned int ds, cs, es;
165 const char *board;
167 printk("\n");
168 print_modules();
169 board = dmi_get_system_info(DMI_PRODUCT_NAME);
170 if (!board)
171 board = "";
172 printk(KERN_INFO "Pid: %d, comm: %.20s %s %s %.*s %s\n",
173 current->pid, current->comm, print_tainted(),
174 init_utsname()->release,
175 (int)strcspn(init_utsname()->version, " "),
176 init_utsname()->version, board);
177 printk(KERN_INFO "RIP: %04lx:[<%016lx>] ", regs->cs & 0xffff, regs->ip);
178 printk_address(regs->ip, 1);
179 printk(KERN_INFO "RSP: %04lx:%016lx EFLAGS: %08lx\n", regs->ss,
180 regs->sp, regs->flags);
181 printk(KERN_INFO "RAX: %016lx RBX: %016lx RCX: %016lx\n",
182 regs->ax, regs->bx, regs->cx);
183 printk(KERN_INFO "RDX: %016lx RSI: %016lx RDI: %016lx\n",
184 regs->dx, regs->si, regs->di);
185 printk(KERN_INFO "RBP: %016lx R08: %016lx R09: %016lx\n",
186 regs->bp, regs->r8, regs->r9);
187 printk(KERN_INFO "R10: %016lx R11: %016lx R12: %016lx\n",
188 regs->r10, regs->r11, regs->r12);
189 printk(KERN_INFO "R13: %016lx R14: %016lx R15: %016lx\n",
190 regs->r13, regs->r14, regs->r15);
192 asm("movl %%ds,%0" : "=r" (ds));
193 asm("movl %%cs,%0" : "=r" (cs));
194 asm("movl %%es,%0" : "=r" (es));
195 asm("movl %%fs,%0" : "=r" (fsindex));
196 asm("movl %%gs,%0" : "=r" (gsindex));
198 rdmsrl(MSR_FS_BASE, fs);
199 rdmsrl(MSR_GS_BASE, gs);
200 rdmsrl(MSR_KERNEL_GS_BASE, shadowgs);
202 if (!all)
203 return;
205 cr0 = read_cr0();
206 cr2 = read_cr2();
207 cr3 = read_cr3();
208 cr4 = read_cr4();
210 printk(KERN_INFO "FS: %016lx(%04x) GS:%016lx(%04x) knlGS:%016lx\n",
211 fs, fsindex, gs, gsindex, shadowgs);
212 printk(KERN_INFO "CS: %04x DS: %04x ES: %04x CR0: %016lx\n", cs, ds,
213 es, cr0);
214 printk(KERN_INFO "CR2: %016lx CR3: %016lx CR4: %016lx\n", cr2, cr3,
215 cr4);
217 get_debugreg(d0, 0);
218 get_debugreg(d1, 1);
219 get_debugreg(d2, 2);
220 printk(KERN_INFO "DR0: %016lx DR1: %016lx DR2: %016lx\n", d0, d1, d2);
221 get_debugreg(d3, 3);
222 get_debugreg(d6, 6);
223 get_debugreg(d7, 7);
224 printk(KERN_INFO "DR3: %016lx DR6: %016lx DR7: %016lx\n", d3, d6, d7);
227 void show_regs(struct pt_regs *regs)
229 printk(KERN_INFO "CPU %d:", smp_processor_id());
230 __show_regs(regs, 1);
231 show_trace(NULL, regs, (void *)(regs + 1), regs->bp);
234 void release_thread(struct task_struct *dead_task)
236 if (dead_task->mm) {
237 if (dead_task->mm->context.size) {
238 printk("WARNING: dead process %8s still has LDT? <%p/%d>\n",
239 dead_task->comm,
240 dead_task->mm->context.ldt,
241 dead_task->mm->context.size);
242 BUG();
247 static inline void set_32bit_tls(struct task_struct *t, int tls, u32 addr)
249 struct user_desc ud = {
250 .base_addr = addr,
251 .limit = 0xfffff,
252 .seg_32bit = 1,
253 .limit_in_pages = 1,
254 .useable = 1,
256 struct desc_struct *desc = t->thread.tls_array;
257 desc += tls;
258 fill_ldt(desc, &ud);
261 static inline u32 read_32bit_tls(struct task_struct *t, int tls)
263 return get_desc_base(&t->thread.tls_array[tls]);
267 * This gets called before we allocate a new thread and copy
268 * the current task into it.
270 void prepare_to_copy(struct task_struct *tsk)
272 unlazy_fpu(tsk);
275 int copy_thread(unsigned long clone_flags, unsigned long sp,
276 unsigned long unused,
277 struct task_struct *p, struct pt_regs *regs)
279 int err;
280 struct pt_regs *childregs;
281 struct task_struct *me = current;
283 childregs = ((struct pt_regs *)
284 (THREAD_SIZE + task_stack_page(p))) - 1;
285 *childregs = *regs;
287 childregs->ax = 0;
288 childregs->sp = sp;
289 if (sp == ~0UL)
290 childregs->sp = (unsigned long)childregs;
292 p->thread.sp = (unsigned long) childregs;
293 p->thread.sp0 = (unsigned long) (childregs+1);
294 p->thread.usersp = me->thread.usersp;
296 set_tsk_thread_flag(p, TIF_FORK);
298 p->thread.fs = me->thread.fs;
299 p->thread.gs = me->thread.gs;
301 savesegment(gs, p->thread.gsindex);
302 savesegment(fs, p->thread.fsindex);
303 savesegment(es, p->thread.es);
304 savesegment(ds, p->thread.ds);
306 if (unlikely(test_tsk_thread_flag(me, TIF_IO_BITMAP))) {
307 p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL);
308 if (!p->thread.io_bitmap_ptr) {
309 p->thread.io_bitmap_max = 0;
310 return -ENOMEM;
312 memcpy(p->thread.io_bitmap_ptr, me->thread.io_bitmap_ptr,
313 IO_BITMAP_BYTES);
314 set_tsk_thread_flag(p, TIF_IO_BITMAP);
318 * Set a new TLS for the child thread?
320 if (clone_flags & CLONE_SETTLS) {
321 #ifdef CONFIG_IA32_EMULATION
322 if (test_thread_flag(TIF_IA32))
323 err = do_set_thread_area(p, -1,
324 (struct user_desc __user *)childregs->si, 0);
325 else
326 #endif
327 err = do_arch_prctl(p, ARCH_SET_FS, childregs->r8);
328 if (err)
329 goto out;
332 clear_tsk_thread_flag(p, TIF_DS_AREA_MSR);
333 p->thread.ds_ctx = NULL;
335 clear_tsk_thread_flag(p, TIF_DEBUGCTLMSR);
336 p->thread.debugctlmsr = 0;
338 err = 0;
339 out:
340 if (err && p->thread.io_bitmap_ptr) {
341 kfree(p->thread.io_bitmap_ptr);
342 p->thread.io_bitmap_max = 0;
344 return err;
347 void
348 start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp)
350 loadsegment(fs, 0);
351 loadsegment(es, 0);
352 loadsegment(ds, 0);
353 load_gs_index(0);
354 regs->ip = new_ip;
355 regs->sp = new_sp;
356 percpu_write(old_rsp, new_sp);
357 regs->cs = __USER_CS;
358 regs->ss = __USER_DS;
359 regs->flags = 0x200;
360 set_fs(USER_DS);
362 * Free the old FP and other extended state
364 free_thread_xstate(current);
366 EXPORT_SYMBOL_GPL(start_thread);
369 * switch_to(x,y) should switch tasks from x to y.
371 * This could still be optimized:
372 * - fold all the options into a flag word and test it with a single test.
373 * - could test fs/gs bitsliced
375 * Kprobes not supported here. Set the probe on schedule instead.
376 * Function graph tracer not supported too.
378 __notrace_funcgraph struct task_struct *
379 __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
381 struct thread_struct *prev = &prev_p->thread;
382 struct thread_struct *next = &next_p->thread;
383 int cpu = smp_processor_id();
384 struct tss_struct *tss = &per_cpu(init_tss, cpu);
385 unsigned fsindex, gsindex;
386 bool preload_fpu;
389 * If the task has used fpu the last 5 timeslices, just do a full
390 * restore of the math state immediately to avoid the trap; the
391 * chances of needing FPU soon are obviously high now
393 preload_fpu = tsk_used_math(next_p) && next_p->fpu_counter > 5;
395 /* we're going to use this soon, after a few expensive things */
396 if (preload_fpu)
397 prefetch(next->xstate);
400 * Reload esp0, LDT and the page table pointer:
402 load_sp0(tss, next);
405 * Switch DS and ES.
406 * This won't pick up thread selector changes, but I guess that is ok.
408 savesegment(es, prev->es);
409 if (unlikely(next->es | prev->es))
410 loadsegment(es, next->es);
412 savesegment(ds, prev->ds);
413 if (unlikely(next->ds | prev->ds))
414 loadsegment(ds, next->ds);
417 /* We must save %fs and %gs before load_TLS() because
418 * %fs and %gs may be cleared by load_TLS().
420 * (e.g. xen_load_tls())
422 savesegment(fs, fsindex);
423 savesegment(gs, gsindex);
425 load_TLS(next, cpu);
427 /* Must be after DS reload */
428 unlazy_fpu(prev_p);
430 /* Make sure cpu is ready for new context */
431 if (preload_fpu)
432 clts();
435 * Leave lazy mode, flushing any hypercalls made here.
436 * This must be done before restoring TLS segments so
437 * the GDT and LDT are properly updated, and must be
438 * done before math_state_restore, so the TS bit is up
439 * to date.
441 arch_end_context_switch(next_p);
444 * Switch FS and GS.
446 * Segment register != 0 always requires a reload. Also
447 * reload when it has changed. When prev process used 64bit
448 * base always reload to avoid an information leak.
450 if (unlikely(fsindex | next->fsindex | prev->fs)) {
451 loadsegment(fs, next->fsindex);
453 * Check if the user used a selector != 0; if yes
454 * clear 64bit base, since overloaded base is always
455 * mapped to the Null selector
457 if (fsindex)
458 prev->fs = 0;
460 /* when next process has a 64bit base use it */
461 if (next->fs)
462 wrmsrl(MSR_FS_BASE, next->fs);
463 prev->fsindex = fsindex;
465 if (unlikely(gsindex | next->gsindex | prev->gs)) {
466 load_gs_index(next->gsindex);
467 if (gsindex)
468 prev->gs = 0;
470 if (next->gs)
471 wrmsrl(MSR_KERNEL_GS_BASE, next->gs);
472 prev->gsindex = gsindex;
475 * Switch the PDA and FPU contexts.
477 prev->usersp = percpu_read(old_rsp);
478 percpu_write(old_rsp, next->usersp);
479 percpu_write(current_task, next_p);
481 percpu_write(kernel_stack,
482 (unsigned long)task_stack_page(next_p) +
483 THREAD_SIZE - KERNEL_STACK_OFFSET);
486 * Now maybe reload the debug registers and handle I/O bitmaps
488 if (unlikely(task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT ||
489 task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV))
490 __switch_to_xtra(prev_p, next_p, tss);
493 * Preload the FPU context, now that we've determined that the
494 * task is likely to be using it.
496 if (preload_fpu)
497 __math_state_restore();
498 return prev_p;
502 * sys_execve() executes a new program.
504 asmlinkage
505 long sys_execve(char __user *name, char __user * __user *argv,
506 char __user * __user *envp, struct pt_regs *regs)
508 long error;
509 char *filename;
511 filename = getname(name);
512 error = PTR_ERR(filename);
513 if (IS_ERR(filename))
514 return error;
515 error = do_execve(filename, argv, envp, regs);
516 putname(filename);
517 return error;
520 void set_personality_64bit(void)
522 /* inherit personality from parent */
524 /* Make sure to be in 64bit mode */
525 clear_thread_flag(TIF_IA32);
527 /* TBD: overwrites user setup. Should have two bits.
528 But 64bit processes have always behaved this way,
529 so it's not too bad. The main problem is just that
530 32bit childs are affected again. */
531 current->personality &= ~READ_IMPLIES_EXEC;
534 asmlinkage long
535 sys_clone(unsigned long clone_flags, unsigned long newsp,
536 void __user *parent_tid, void __user *child_tid, struct pt_regs *regs)
538 if (!newsp)
539 newsp = regs->sp;
540 return do_fork(clone_flags, newsp, regs, 0, parent_tid, child_tid);
543 unsigned long get_wchan(struct task_struct *p)
545 unsigned long stack;
546 u64 fp, ip;
547 int count = 0;
549 if (!p || p == current || p->state == TASK_RUNNING)
550 return 0;
551 stack = (unsigned long)task_stack_page(p);
552 if (p->thread.sp < stack || p->thread.sp >= stack+THREAD_SIZE)
553 return 0;
554 fp = *(u64 *)(p->thread.sp);
555 do {
556 if (fp < (unsigned long)stack ||
557 fp >= (unsigned long)stack+THREAD_SIZE)
558 return 0;
559 ip = *(u64 *)(fp+8);
560 if (!in_sched_functions(ip))
561 return ip;
562 fp = *(u64 *)fp;
563 } while (count++ < 16);
564 return 0;
567 long do_arch_prctl(struct task_struct *task, int code, unsigned long addr)
569 int ret = 0;
570 int doit = task == current;
571 int cpu;
573 switch (code) {
574 case ARCH_SET_GS:
575 if (addr >= TASK_SIZE_OF(task))
576 return -EPERM;
577 cpu = get_cpu();
578 /* handle small bases via the GDT because that's faster to
579 switch. */
580 if (addr <= 0xffffffff) {
581 set_32bit_tls(task, GS_TLS, addr);
582 if (doit) {
583 load_TLS(&task->thread, cpu);
584 load_gs_index(GS_TLS_SEL);
586 task->thread.gsindex = GS_TLS_SEL;
587 task->thread.gs = 0;
588 } else {
589 task->thread.gsindex = 0;
590 task->thread.gs = addr;
591 if (doit) {
592 load_gs_index(0);
593 ret = checking_wrmsrl(MSR_KERNEL_GS_BASE, addr);
596 put_cpu();
597 break;
598 case ARCH_SET_FS:
599 /* Not strictly needed for fs, but do it for symmetry
600 with gs */
601 if (addr >= TASK_SIZE_OF(task))
602 return -EPERM;
603 cpu = get_cpu();
604 /* handle small bases via the GDT because that's faster to
605 switch. */
606 if (addr <= 0xffffffff) {
607 set_32bit_tls(task, FS_TLS, addr);
608 if (doit) {
609 load_TLS(&task->thread, cpu);
610 loadsegment(fs, FS_TLS_SEL);
612 task->thread.fsindex = FS_TLS_SEL;
613 task->thread.fs = 0;
614 } else {
615 task->thread.fsindex = 0;
616 task->thread.fs = addr;
617 if (doit) {
618 /* set the selector to 0 to not confuse
619 __switch_to */
620 loadsegment(fs, 0);
621 ret = checking_wrmsrl(MSR_FS_BASE, addr);
624 put_cpu();
625 break;
626 case ARCH_GET_FS: {
627 unsigned long base;
628 if (task->thread.fsindex == FS_TLS_SEL)
629 base = read_32bit_tls(task, FS_TLS);
630 else if (doit)
631 rdmsrl(MSR_FS_BASE, base);
632 else
633 base = task->thread.fs;
634 ret = put_user(base, (unsigned long __user *)addr);
635 break;
637 case ARCH_GET_GS: {
638 unsigned long base;
639 unsigned gsindex;
640 if (task->thread.gsindex == GS_TLS_SEL)
641 base = read_32bit_tls(task, GS_TLS);
642 else if (doit) {
643 savesegment(gs, gsindex);
644 if (gsindex)
645 rdmsrl(MSR_KERNEL_GS_BASE, base);
646 else
647 base = task->thread.gs;
648 } else
649 base = task->thread.gs;
650 ret = put_user(base, (unsigned long __user *)addr);
651 break;
654 default:
655 ret = -EINVAL;
656 break;
659 return ret;
662 long sys_arch_prctl(int code, unsigned long addr)
664 return do_arch_prctl(current, code, addr);
667 unsigned long KSTK_ESP(struct task_struct *task)
669 return (test_tsk_thread_flag(task, TIF_IA32)) ?
670 (task_pt_regs(task)->sp) : ((task)->thread.usersp);