projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
[PATCH] x86: i8253/i8259A lock cleanup
[linux-2.6-block.git] / arch / i386 / kernel / process.c
CommitLineData
1da177e4
LT		 1/*
2 * linux/arch/i386/kernel/process.c
3 *
4 * Copyright (C) 1995 Linus Torvalds
5 *
6 * Pentium III FXSR, SSE support
7 * Gareth Hughes <gareth@valinux.com>, May 2000
8 */
9
10/*
11 * This file handles the architecture-dependent parts of process handling..
12 */
13
14#include <stdarg.h>
15
f3705136 16#include <linux/cpu.h>
1da177e4
LT		 17#include <linux/errno.h>
18#include <linux/sched.h>
19#include <linux/fs.h>
20#include <linux/kernel.h>
21#include <linux/mm.h>
22#include <linux/elfcore.h>
23#include <linux/smp.h>
24#include <linux/smp_lock.h>
25#include <linux/stddef.h>
26#include <linux/slab.h>
27#include <linux/vmalloc.h>
28#include <linux/user.h>
29#include <linux/a.out.h>
30#include <linux/interrupt.h>
31#include <linux/config.h>
32#include <linux/utsname.h>
33#include <linux/delay.h>
34#include <linux/reboot.h>
35#include <linux/init.h>
36#include <linux/mc146818rtc.h>
37#include <linux/module.h>
38#include <linux/kallsyms.h>
39#include <linux/ptrace.h>
40#include <linux/random.h>
b94cce92 41#include <linux/kprobes.h>
1da177e4
LT		 42
43#include <asm/uaccess.h>
44#include <asm/pgtable.h>
45#include <asm/system.h>
46#include <asm/io.h>
47#include <asm/ldt.h>
48#include <asm/processor.h>
49#include <asm/i387.h>
50#include <asm/irq.h>
51#include <asm/desc.h>
52#ifdef CONFIG_MATH_EMULATION
53#include <asm/math_emu.h>
54#endif
55
56#include <linux/irq.h>
57#include <linux/err.h>
58
f3705136
ZM		 59#include <asm/tlbflush.h>
60#include <asm/cpu.h>
61
1da177e4
LT		 62asmlinkage void ret_from_fork(void) __asm__("ret_from_fork");
63
64static int hlt_counter;
65
66unsigned long boot_option_idle_override = 0;
67EXPORT_SYMBOL(boot_option_idle_override);
68
69/*
70 * Return saved PC of a blocked thread.
71 */
72unsigned long thread_saved_pc(struct task_struct *tsk)
73{
74 return ((unsigned long *)tsk->thread.esp)[3];
75}
76
77/*
78 * Powermanagement idle function, if any..
79 */
80void (*pm_idle)(void);
129f6946 81EXPORT_SYMBOL(pm_idle);
1da177e4
LT		 82static DEFINE_PER_CPU(unsigned int, cpu_idle_state);
83
84void disable_hlt(void)
85{
86 hlt_counter++;
87}
88
89EXPORT_SYMBOL(disable_hlt);
90
91void enable_hlt(void)
92{
93 hlt_counter--;
94}
95
96EXPORT_SYMBOL(enable_hlt);
97
98/*
99 * We use this if we don't have any better
100 * idle routine..
101 */
102void default_idle(void)
103{
104 if (!hlt_counter && boot_cpu_data.hlt_works_ok) {
105 local_irq_disable();
106 if (!need_resched())
107 safe_halt();
108 else
109 local_irq_enable();
110 } else {
111 cpu_relax();
112 }
113}
129f6946
AD		 114#ifdef CONFIG_APM_MODULE
115EXPORT_SYMBOL(default_idle);
116#endif
1da177e4
LT		 117
118/*
119 * On SMP it's slightly faster (but much more power-consuming!)
120 * to poll the ->work.need_resched flag instead of waiting for the
121 * cross-CPU IPI to arrive. Use this option with caution.
122 */
123static void poll_idle (void)
124{
125 int oldval;
126
127 local_irq_enable();
128
129 /*
130 * Deal with another CPU just having chosen a thread to
131 * run here:
132 */
133 oldval = test_and_clear_thread_flag(TIF_NEED_RESCHED);
134
135 if (!oldval) {
136 set_thread_flag(TIF_POLLING_NRFLAG);
137 asm volatile(
138 "2:"
139 "testl %0, %1;"
140 "rep; nop;"
141 "je 2b;"
142 : : "i"(_TIF_NEED_RESCHED), "m" (current_thread_info()->flags));
143
144 clear_thread_flag(TIF_POLLING_NRFLAG);
145 } else {
146 set_need_resched();
147 }
148}
149
f3705136
ZM		 150#ifdef CONFIG_HOTPLUG_CPU
151#include <asm/nmi.h>
152/* We don't actually take CPU down, just spin without interrupts. */
153static inline void play_dead(void)
154{
e1367daf
LS		 155 /* This must be done before dead CPU ack */
156 cpu_exit_clear();
157 wbinvd();
158 mb();
f3705136
ZM		 159 /* Ack it */
160 __get_cpu_var(cpu_state) = CPU_DEAD;
161
e1367daf
LS		 162 /*
163 * With physical CPU hotplug, we should halt the cpu
164 */
f3705136 165 local_irq_disable();
e1367daf
LS		 166 while (1)
167 __asm__ __volatile__("hlt":::"memory");
f3705136
ZM		 168}
169#else
170static inline void play_dead(void)
171{
172 BUG();
173}
174#endif /* CONFIG_HOTPLUG_CPU */
175
1da177e4
LT		 176/*
177 * The idle thread. There's no useful work to be
178 * done, so just try to conserve power and have a
179 * low exit latency (ie sit in a loop waiting for
180 * somebody to say that they'd like to reschedule)
181 */
f3705136 182void cpu_idle(void)
1da177e4 183{
f3705136
ZM		 184 int cpu = raw_smp_processor_id();
185
1da177e4
LT		 186 /* endless idle loop with no priority at all */
187 while (1) {
188 while (!need_resched()) {
189 void (*idle)(void);
190
191 if (__get_cpu_var(cpu_idle_state))
192 __get_cpu_var(cpu_idle_state) = 0;
193
194 rmb();
195 idle = pm_idle;
196
197 if (!idle)
198 idle = default_idle;
199
f3705136
ZM		 200 if (cpu_is_offline(cpu))
201 play_dead();
202
1da177e4
LT		 203 __get_cpu_var(irq_stat).idle_timestamp = jiffies;
204 idle();
205 }
206 schedule();
207 }
208}
209
210void cpu_idle_wait(void)
211{
212 unsigned int cpu, this_cpu = get_cpu();
213 cpumask_t map;
214
215 set_cpus_allowed(current, cpumask_of_cpu(this_cpu));
216 put_cpu();
217
218 cpus_clear(map);
219 for_each_online_cpu(cpu) {
220 per_cpu(cpu_idle_state, cpu) = 1;
221 cpu_set(cpu, map);
222 }
223
224 __get_cpu_var(cpu_idle_state) = 0;
225
226 wmb();
227 do {
228 ssleep(1);
229 for_each_online_cpu(cpu) {
230 if (cpu_isset(cpu, map) && !per_cpu(cpu_idle_state, cpu))
231 cpu_clear(cpu, map);
232 }
233 cpus_and(map, map, cpu_online_map);
234 } while (!cpus_empty(map));
235}
236EXPORT_SYMBOL_GPL(cpu_idle_wait);
237
238/*
239 * This uses new MONITOR/MWAIT instructions on P4 processors with PNI,
240 * which can obviate IPI to trigger checking of need_resched.
241 * We execute MONITOR against need_resched and enter optimized wait state
242 * through MWAIT. Whenever someone changes need_resched, we would be woken
243 * up from MWAIT (without an IPI).
244 */
245static void mwait_idle(void)
246{
247 local_irq_enable();
248
249 if (!need_resched()) {
250 set_thread_flag(TIF_POLLING_NRFLAG);
251 do {
252 __monitor((void *)&current_thread_info()->flags, 0, 0);
253 if (need_resched())
254 break;
255 __mwait(0, 0);
256 } while (!need_resched());
257 clear_thread_flag(TIF_POLLING_NRFLAG);
258 }
259}
260
0bb3184d 261void __devinit select_idle_routine(const struct cpuinfo_x86 *c)
1da177e4
LT		 262{
263 if (cpu_has(c, X86_FEATURE_MWAIT)) {
264 printk("monitor/mwait feature present.\n");
265 /*
266 * Skip, if setup has overridden idle.
267 * One CPU supports mwait => All CPUs supports mwait
268 */
269 if (!pm_idle) {
270 printk("using mwait in idle threads.\n");
271 pm_idle = mwait_idle;
272 }
273 }
274}
275
276static int __init idle_setup (char *str)
277{
278 if (!strncmp(str, "poll", 4)) {
279 printk("using polling idle threads.\n");
280 pm_idle = poll_idle;
281#ifdef CONFIG_X86_SMP
282 if (smp_num_siblings > 1)
283 printk("WARNING: polling idle and HT enabled, performance may degrade.\n");
284#endif
285 } else if (!strncmp(str, "halt", 4)) {
286 printk("using halt in idle threads.\n");
287 pm_idle = default_idle;
288 }
289
290 boot_option_idle_override = 1;
291 return 1;
292}
293
294__setup("idle=", idle_setup);
295
296void show_regs(struct pt_regs * regs)
297{
298 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L;
299
300 printk("\n");
301 printk("Pid: %d, comm: %20s\n", current->pid, current->comm);
302 printk("EIP: %04x:[<%08lx>] CPU: %d\n",0xffff & regs->xcs,regs->eip, smp_processor_id());
303 print_symbol("EIP is at %s\n", regs->eip);
304
717b594a 305 if (user_mode(regs))
1da177e4
LT		 306 printk(" ESP: %04x:%08lx",0xffff & regs->xss,regs->esp);
307 printk(" EFLAGS: %08lx %s (%s)\n",
308 regs->eflags, print_tainted(), system_utsname.release);
309 printk("EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n",
310 regs->eax,regs->ebx,regs->ecx,regs->edx);
311 printk("ESI: %08lx EDI: %08lx EBP: %08lx",
312 regs->esi, regs->edi, regs->ebp);
313 printk(" DS: %04x ES: %04x\n",
314 0xffff & regs->xds,0xffff & regs->xes);
315
316 __asm__("movl %%cr0, %0": "=r" (cr0));
317 __asm__("movl %%cr2, %0": "=r" (cr2));
318 __asm__("movl %%cr3, %0": "=r" (cr3));
319 /* This could fault if %cr4 does not exist */
320 __asm__("1: movl %%cr4, %0 \n"
321 "2: \n"
322 ".section __ex_table,\"a\" \n"
323 ".long 1b,2b \n"
324 ".previous \n"
325 : "=r" (cr4): "0" (0));
326 printk("CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n", cr0, cr2, cr3, cr4);
327 show_trace(NULL, &regs->esp);
328}
329
330/*
331 * This gets run with %ebx containing the
332 * function to call, and %edx containing
333 * the "args".
334 */
335extern void kernel_thread_helper(void);
336__asm__(".section .text\n"
337 ".align 4\n"
338 "kernel_thread_helper:\n\t"
339 "movl %edx,%eax\n\t"
340 "pushl %edx\n\t"
341 "call *%ebx\n\t"
342 "pushl %eax\n\t"
343 "call do_exit\n"
344 ".previous");
345
346/*
347 * Create a kernel thread
348 */
349int kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
350{
351 struct pt_regs regs;
352
353 memset(&regs, 0, sizeof(regs));
354
355 regs.ebx = (unsigned long) fn;
356 regs.edx = (unsigned long) arg;
357
358 regs.xds = __USER_DS;
359 regs.xes = __USER_DS;
360 regs.orig_eax = -1;
361 regs.eip = (unsigned long) kernel_thread_helper;
362 regs.xcs = __KERNEL_CS;
363 regs.eflags = X86_EFLAGS_IF | X86_EFLAGS_SF | X86_EFLAGS_PF | 0x2;
364
365 /* Ok, create the new process.. */
366 return do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0, &regs, 0, NULL, NULL);
367}
129f6946 368EXPORT_SYMBOL(kernel_thread);
1da177e4
LT		 369
370/*
371 * Free current thread data structures etc..
372 */
373void exit_thread(void)
374{
375 struct task_struct *tsk = current;
376 struct thread_struct *t = &tsk->thread;
377
b94cce92
HN		 378 /*
379 * Remove function-return probe instances associated with this task
380 * and put them back on the free list. Do not insert an exit probe for
381 * this function, it will be disabled by kprobe_flush_task if you do.
382 */
383 kprobe_flush_task(tsk);
384
1da177e4
LT		 385 /* The process may have allocated an io port bitmap... nuke it. */
386 if (unlikely(NULL != t->io_bitmap_ptr)) {
387 int cpu = get_cpu();
388 struct tss_struct *tss = &per_cpu(init_tss, cpu);
389
390 kfree(t->io_bitmap_ptr);
391 t->io_bitmap_ptr = NULL;
392 /*
393 * Careful, clear this in the TSS too:
394 */
395 memset(tss->io_bitmap, 0xff, tss->io_bitmap_max);
396 t->io_bitmap_max = 0;
397 tss->io_bitmap_owner = NULL;
398 tss->io_bitmap_max = 0;
399 tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
400 put_cpu();
401 }
402}
403
404void flush_thread(void)
405{
406 struct task_struct *tsk = current;
407
b94cce92
HN		 408 /*
409 * Remove function-return probe instances associated with this task
410 * and put them back on the free list. Do not insert an exit probe for
411 * this function, it will be disabled by kprobe_flush_task if you do.
412 */
413 kprobe_flush_task(tsk);
414
1da177e4
LT		 415 memset(tsk->thread.debugreg, 0, sizeof(unsigned long)*8);
416 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array));
417 /*
418 * Forget coprocessor state..
419 */
420 clear_fpu(tsk);
421 clear_used_math();
422}
423
424void release_thread(struct task_struct *dead_task)
425{
426 if (dead_task->mm) {
427 // temporary debugging check
428 if (dead_task->mm->context.size) {
429 printk("WARNING: dead process %8s still has LDT? <%p/%d>\n",
430 dead_task->comm,
431 dead_task->mm->context.ldt,
432 dead_task->mm->context.size);
433 BUG();
434 }
435 }
436
437 release_vm86_irqs(dead_task);
438}
439
440/*
441 * This gets called before we allocate a new thread and copy
442 * the current task into it.
443 */
444void prepare_to_copy(struct task_struct *tsk)
445{
446 unlazy_fpu(tsk);
447}
448
449int copy_thread(int nr, unsigned long clone_flags, unsigned long esp,
450 unsigned long unused,
451 struct task_struct * p, struct pt_regs * regs)
452{
453 struct pt_regs * childregs;
454 struct task_struct *tsk;
455 int err;
456
457 childregs = ((struct pt_regs *) (THREAD_SIZE + (unsigned long) p->thread_info)) - 1;
5df24082
SS		 458 /*
459 * The below -8 is to reserve 8 bytes on top of the ring0 stack.
460 * This is necessary to guarantee that the entire "struct pt_regs"
461 * is accessable even if the CPU haven't stored the SS/ESP registers
462 * on the stack (interrupt gate does not save these registers
463 * when switching to the same priv ring).
464 * Therefore beware: accessing the xss/esp fields of the
465 * "struct pt_regs" is possible, but they may contain the
466 * completely wrong values.
467 */
f48d9663
AN		 468 childregs = (struct pt_regs *) ((unsigned long) childregs - 8);
469 *childregs = *regs;
470 childregs->eax = 0;
471 childregs->esp = esp;
472
473 p->thread.esp = (unsigned long) childregs;
474 p->thread.esp0 = (unsigned long) (childregs+1);
1da177e4
LT		 475
476 p->thread.eip = (unsigned long) ret_from_fork;
477
478 savesegment(fs,p->thread.fs);
479 savesegment(gs,p->thread.gs);
480
481 tsk = current;
482 if (unlikely(NULL != tsk->thread.io_bitmap_ptr)) {
483 p->thread.io_bitmap_ptr = kmalloc(IO_BITMAP_BYTES, GFP_KERNEL);
484 if (!p->thread.io_bitmap_ptr) {
485 p->thread.io_bitmap_max = 0;
486 return -ENOMEM;
487 }
488 memcpy(p->thread.io_bitmap_ptr, tsk->thread.io_bitmap_ptr,
489 IO_BITMAP_BYTES);
490 }
491
492 /*
493 * Set a new TLS for the child thread?
494 */
495 if (clone_flags & CLONE_SETTLS) {
496 struct desc_struct *desc;
497 struct user_desc info;
498 int idx;
499
500 err = -EFAULT;
501 if (copy_from_user(&info, (void __user *)childregs->esi, sizeof(info)))
502 goto out;
503 err = -EINVAL;
504 if (LDT_empty(&info))
505 goto out;
506
507 idx = info.entry_number;
508 if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
509 goto out;
510
511 desc = p->thread.tls_array + idx - GDT_ENTRY_TLS_MIN;
512 desc->a = LDT_entry_a(&info);
513 desc->b = LDT_entry_b(&info);
514 }
515
516 err = 0;
517 out:
518 if (err && p->thread.io_bitmap_ptr) {
519 kfree(p->thread.io_bitmap_ptr);
520 p->thread.io_bitmap_max = 0;
521 }
522 return err;
523}
524
525/*
526 * fill in the user structure for a core dump..
527 */
528void dump_thread(struct pt_regs * regs, struct user * dump)
529{
530 int i;
531
532/* changed the size calculations - should hopefully work better. lbt */
533 dump->magic = CMAGIC;
534 dump->start_code = 0;
535 dump->start_stack = regs->esp & ~(PAGE_SIZE - 1);
536 dump->u_tsize = ((unsigned long) current->mm->end_code) >> PAGE_SHIFT;
537 dump->u_dsize = ((unsigned long) (current->mm->brk + (PAGE_SIZE-1))) >> PAGE_SHIFT;
538 dump->u_dsize -= dump->u_tsize;
539 dump->u_ssize = 0;
540 for (i = 0; i < 8; i++)
541 dump->u_debugreg[i] = current->thread.debugreg[i];
542
543 if (dump->start_stack < TASK_SIZE)
544 dump->u_ssize = ((unsigned long) (TASK_SIZE - dump->start_stack)) >> PAGE_SHIFT;
545
546 dump->regs.ebx = regs->ebx;
547 dump->regs.ecx = regs->ecx;
548 dump->regs.edx = regs->edx;
549 dump->regs.esi = regs->esi;
550 dump->regs.edi = regs->edi;
551 dump->regs.ebp = regs->ebp;
552 dump->regs.eax = regs->eax;
553 dump->regs.ds = regs->xds;
554 dump->regs.es = regs->xes;
555 savesegment(fs,dump->regs.fs);
556 savesegment(gs,dump->regs.gs);
557 dump->regs.orig_eax = regs->orig_eax;
558 dump->regs.eip = regs->eip;
559 dump->regs.cs = regs->xcs;
560 dump->regs.eflags = regs->eflags;
561 dump->regs.esp = regs->esp;
562 dump->regs.ss = regs->xss;
563
564 dump->u_fpvalid = dump_fpu (regs, &dump->i387);
565}
129f6946 566EXPORT_SYMBOL(dump_thread);
1da177e4
LT		 567
568/*
569 * Capture the user space registers if the task is not running (in user space)
570 */
571int dump_task_regs(struct task_struct *tsk, elf_gregset_t *regs)
572{
573 struct pt_regs ptregs;
574
575 ptregs = *(struct pt_regs *)
576 ((unsigned long)tsk->thread_info+THREAD_SIZE - sizeof(ptregs));
577 ptregs.xcs &= 0xffff;
578 ptregs.xds &= 0xffff;
579 ptregs.xes &= 0xffff;
580 ptregs.xss &= 0xffff;
581
582 elf_core_copy_regs(regs, &ptregs);
583
584 return 1;
585}
586
587static inline void
588handle_io_bitmap(struct thread_struct *next, struct tss_struct *tss)
589{
590 if (!next->io_bitmap_ptr) {
591 /*
592 * Disable the bitmap via an invalid offset. We still cache
593 * the previous bitmap owner and the IO bitmap contents:
594 */
595 tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET;
596 return;
597 }
598 if (likely(next == tss->io_bitmap_owner)) {
599 /*
600 * Previous owner of the bitmap (hence the bitmap content)
601 * matches the next task, we dont have to do anything but
602 * to set a valid offset in the TSS:
603 */
604 tss->io_bitmap_base = IO_BITMAP_OFFSET;
605 return;
606 }
607 /*
608 * Lazy TSS's I/O bitmap copy. We set an invalid offset here
609 * and we let the task to get a GPF in case an I/O instruction
610 * is performed. The handler of the GPF will verify that the
611 * faulting task has a valid I/O bitmap and, it true, does the
612 * real copy and restart the instruction. This will save us
613 * redundant copies when the currently switched task does not
614 * perform any I/O during its timeslice.
615 */
616 tss->io_bitmap_base = INVALID_IO_BITMAP_OFFSET_LAZY;
617}
1da177e4 618
ffaa8bd6
AA		 619/*
620 * This function selects if the context switch from prev to next
621 * has to tweak the TSC disable bit in the cr4.
622 */
623static inline void disable_tsc(struct task_struct *prev_p,
624 struct task_struct *next_p)
625{
626 struct thread_info *prev, *next;
627
628 /*
629 * gcc should eliminate the ->thread_info dereference if
630 * has_secure_computing returns 0 at compile time (SECCOMP=n).
631 */
632 prev = prev_p->thread_info;
633 next = next_p->thread_info;
634
635 if (has_secure_computing(prev) || has_secure_computing(next)) {
636 /* slow path here */
637 if (has_secure_computing(prev) &&
638 !has_secure_computing(next)) {
639 write_cr4(read_cr4() & ~X86_CR4_TSD);
640 } else if (!has_secure_computing(prev) &&
641 has_secure_computing(next))
642 write_cr4(read_cr4() | X86_CR4_TSD);
643 }
644}
645
1da177e4
LT		 646/*
647 * switch_to(x,yn) should switch tasks from x to y.
648 *
649 * We fsave/fwait so that an exception goes off at the right time
650 * (as a call from the fsave or fwait in effect) rather than to
651 * the wrong process. Lazy FP saving no longer makes any sense
652 * with modern CPU's, and this simplifies a lot of things (SMP
653 * and UP become the same).
654 *
655 * NOTE! We used to use the x86 hardware context switching. The
656 * reason for not using it any more becomes apparent when you
657 * try to recover gracefully from saved state that is no longer
658 * valid (stale segment register values in particular). With the
659 * hardware task-switch, there is no way to fix up bad state in
660 * a reasonable manner.
661 *
662 * The fact that Intel documents the hardware task-switching to
663 * be slow is a fairly red herring - this code is not noticeably
664 * faster. However, there _is_ some room for improvement here,
665 * so the performance issues may eventually be a valid point.
666 * More important, however, is the fact that this allows us much
667 * more flexibility.
668 *
669 * The return value (in %eax) will be the "prev" task after
670 * the task-switch, and shows up in ret_from_fork in entry.S,
671 * for example.
672 */
673struct task_struct fastcall * __switch_to(struct task_struct *prev_p, struct task_struct *next_p)
674{
675 struct thread_struct *prev = &prev_p->thread,
676 *next = &next_p->thread;
677 int cpu = smp_processor_id();
678 struct tss_struct *tss = &per_cpu(init_tss, cpu);
679
680 /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */
681
682 __unlazy_fpu(prev_p);
683
684 /*
685 * Reload esp0, LDT and the page table pointer:
686 */
687 load_esp0(tss, next);
688
689 /*
690 * Load the per-thread Thread-Local Storage descriptor.
691 */
692 load_TLS(next, cpu);
693
694 /*
695 * Save away %fs and %gs. No need to save %es and %ds, as
696 * those are always kernel segments while inside the kernel.
697 */
fd51f666
L		 698 asm volatile("mov %%fs,%0":"=m" (prev->fs));
699 asm volatile("mov %%gs,%0":"=m" (prev->gs));
1da177e4
LT		 700
701 /*
702 * Restore %fs and %gs if needed.
703 */
704 if (unlikely(prev->fs | prev->gs | next->fs | next->gs)) {
705 loadsegment(fs, next->fs);
706 loadsegment(gs, next->gs);
707 }
708
709 /*
710 * Now maybe reload the debug registers
711 */
712 if (unlikely(next->debugreg[7])) {
1cc6f12e
VH		 713 set_debugreg(current->thread.debugreg[0], 0);
714 set_debugreg(current->thread.debugreg[1], 1);
715 set_debugreg(current->thread.debugreg[2], 2);
716 set_debugreg(current->thread.debugreg[3], 3);
1da177e4 717 /* no 4 and 5 */
1cc6f12e
VH		 718 set_debugreg(current->thread.debugreg[6], 6);
719 set_debugreg(current->thread.debugreg[7], 7);
1da177e4
LT		 720 }
721
722 if (unlikely(prev->io_bitmap_ptr || next->io_bitmap_ptr))
723 handle_io_bitmap(next, tss);
724
ffaa8bd6
AA		 725 disable_tsc(prev_p, next_p);
726
1da177e4
LT		 727 return prev_p;
728}
729
730asmlinkage int sys_fork(struct pt_regs regs)
731{
732 return do_fork(SIGCHLD, regs.esp, &regs, 0, NULL, NULL);
733}
734
735asmlinkage int sys_clone(struct pt_regs regs)
736{
737 unsigned long clone_flags;
738 unsigned long newsp;
739 int __user *parent_tidptr, *child_tidptr;
740
741 clone_flags = regs.ebx;
742 newsp = regs.ecx;
743 parent_tidptr = (int __user *)regs.edx;
744 child_tidptr = (int __user *)regs.edi;
745 if (!newsp)
746 newsp = regs.esp;
747 return do_fork(clone_flags, newsp, &regs, 0, parent_tidptr, child_tidptr);
748}
749
750/*
751 * This is trivial, and on the face of it looks like it
752 * could equally well be done in user mode.
753 *
754 * Not so, for quite unobvious reasons - register pressure.
755 * In user mode vfork() cannot have a stack frame, and if
756 * done by calling the "clone()" system call directly, you
757 * do not have enough call-clobbered registers to hold all
758 * the information you need.
759 */
760asmlinkage int sys_vfork(struct pt_regs regs)
761{
762 return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, regs.esp, &regs, 0, NULL, NULL);
763}
764
765/*
766 * sys_execve() executes a new program.
767 */
768asmlinkage int sys_execve(struct pt_regs regs)
769{
770 int error;
771 char * filename;
772
773 filename = getname((char __user *) regs.ebx);
774 error = PTR_ERR(filename);
775 if (IS_ERR(filename))
776 goto out;
777 error = do_execve(filename,
778 (char __user * __user *) regs.ecx,
779 (char __user * __user *) regs.edx,
780 &regs);
781 if (error == 0) {
782 task_lock(current);
783 current->ptrace &= ~PT_DTRACE;
784 task_unlock(current);
785 /* Make sure we don't return using sysenter.. */
786 set_thread_flag(TIF_IRET);
787 }
788 putname(filename);
789out:
790 return error;
791}
792
793#define top_esp (THREAD_SIZE - sizeof(unsigned long))
794#define top_ebp (THREAD_SIZE - 2*sizeof(unsigned long))
795
796unsigned long get_wchan(struct task_struct *p)
797{
798 unsigned long ebp, esp, eip;
799 unsigned long stack_page;
800 int count = 0;
801 if (!p || p == current || p->state == TASK_RUNNING)
802 return 0;
803 stack_page = (unsigned long)p->thread_info;
804 esp = p->thread.esp;
805 if (!stack_page || esp < stack_page || esp > top_esp+stack_page)
806 return 0;
807 /* include/asm-i386/system.h:switch_to() pushes ebp last. */
808 ebp = *(unsigned long *) esp;
809 do {
810 if (ebp < stack_page || ebp > top_ebp+stack_page)
811 return 0;
812 eip = *(unsigned long *) (ebp+4);
813 if (!in_sched_functions(eip))
814 return eip;
815 ebp = *(unsigned long *) ebp;
816 } while (count++ < 16);
817 return 0;
818}
129f6946 819EXPORT_SYMBOL(get_wchan);
1da177e4
LT		 820
821/*
822 * sys_alloc_thread_area: get a yet unused TLS descriptor index.
823 */
824static int get_free_idx(void)
825{
826 struct thread_struct *t = &current->thread;
827 int idx;
828
829 for (idx = 0; idx < GDT_ENTRY_TLS_ENTRIES; idx++)
830 if (desc_empty(t->tls_array + idx))
831 return idx + GDT_ENTRY_TLS_MIN;
832 return -ESRCH;
833}
834
835/*
836 * Set a given TLS descriptor:
837 */
838asmlinkage int sys_set_thread_area(struct user_desc __user *u_info)
839{
840 struct thread_struct *t = &current->thread;
841 struct user_desc info;
842 struct desc_struct *desc;
843 int cpu, idx;
844
845 if (copy_from_user(&info, u_info, sizeof(info)))
846 return -EFAULT;
847 idx = info.entry_number;
848
849 /*
850 * index -1 means the kernel should try to find and
851 * allocate an empty descriptor:
852 */
853 if (idx == -1) {
854 idx = get_free_idx();
855 if (idx < 0)
856 return idx;
857 if (put_user(idx, &u_info->entry_number))
858 return -EFAULT;
859 }
860
861 if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
862 return -EINVAL;
863
864 desc = t->tls_array + idx - GDT_ENTRY_TLS_MIN;
865
866 /*
867 * We must not get preempted while modifying the TLS.
868 */
869 cpu = get_cpu();
870
871 if (LDT_empty(&info)) {
872 desc->a = 0;
873 desc->b = 0;
874 } else {
875 desc->a = LDT_entry_a(&info);
876 desc->b = LDT_entry_b(&info);
877 }
878 load_TLS(t, cpu);
879
880 put_cpu();
881
882 return 0;
883}
884
885/*
886 * Get the current Thread-Local Storage area:
887 */
888
889#define GET_BASE(desc) ( \
890 (((desc)->a >> 16) & 0x0000ffff) | \
891 (((desc)->b << 16) & 0x00ff0000) | \
892 ( (desc)->b & 0xff000000) )
893
894#define GET_LIMIT(desc) ( \
895 ((desc)->a & 0x0ffff) | \
896 ((desc)->b & 0xf0000) )
897
898#define GET_32BIT(desc) (((desc)->b >> 22) & 1)
899#define GET_CONTENTS(desc) (((desc)->b >> 10) & 3)
900#define GET_WRITABLE(desc) (((desc)->b >> 9) & 1)
901#define GET_LIMIT_PAGES(desc) (((desc)->b >> 23) & 1)
902#define GET_PRESENT(desc) (((desc)->b >> 15) & 1)
903#define GET_USEABLE(desc) (((desc)->b >> 20) & 1)
904
905asmlinkage int sys_get_thread_area(struct user_desc __user *u_info)
906{
907 struct user_desc info;
908 struct desc_struct *desc;
909 int idx;
910
911 if (get_user(idx, &u_info->entry_number))
912 return -EFAULT;
913 if (idx < GDT_ENTRY_TLS_MIN || idx > GDT_ENTRY_TLS_MAX)
914 return -EINVAL;
915
916 desc = current->thread.tls_array + idx - GDT_ENTRY_TLS_MIN;
917
918 info.entry_number = idx;
919 info.base_addr = GET_BASE(desc);
920 info.limit = GET_LIMIT(desc);
921 info.seg_32bit = GET_32BIT(desc);
922 info.contents = GET_CONTENTS(desc);
923 info.read_exec_only = !GET_WRITABLE(desc);
924 info.limit_in_pages = GET_LIMIT_PAGES(desc);
925 info.seg_not_present = !GET_PRESENT(desc);
926 info.useable = GET_USEABLE(desc);
927
928 if (copy_to_user(u_info, &info, sizeof(info)))
929 return -EFAULT;
930 return 0;
931}
932
933unsigned long arch_align_stack(unsigned long sp)
934{
935 if (randomize_va_space)
936 sp -= get_random_int() % 8192;
937 return sp & ~0xf;
938}
Linux 6.x block layer and io_uring tree(s)