1 /* 2 * Copyright (C) 1995 Linus Torvalds 3 * 4 * Pentium III FXSR, SSE support 5 * Gareth Hughes <gareth@valinux.com>, May 2000 6 */ 7 8 /* 9 * This file handles the architecture-dependent parts of process handling.. 10 */ 11 12 #include <linux/stackprotector.h> 13 #include <linux/cpu.h> 14 #include <linux/errno.h> 15 #include <linux/sched.h> 16 #include <linux/fs.h> 17 #include <linux/kernel.h> 18 #include <linux/mm.h> 19 #include <linux/elfcore.h> 20 #include <linux/smp.h> 21 #include <linux/stddef.h> 22 #include <linux/slab.h> 23 #include <linux/vmalloc.h> 24 #include <linux/user.h> 25 #include <linux/interrupt.h> 26 #include <linux/delay.h> 27 #include <linux/reboot.h> 28 #include <linux/init.h> 29 #include <linux/mc146818rtc.h> 30 #include <linux/module.h> 31 #include <linux/kallsyms.h> 32 #include <linux/ptrace.h> 33 #include <linux/personality.h> 34 #include <linux/tick.h> 35 #include <linux/percpu.h> 36 #include <linux/prctl.h> 37 #include <linux/ftrace.h> 38 #include <linux/uaccess.h> 39 #include <linux/io.h> 40 #include <linux/kdebug.h> 41 42 #include <asm/pgtable.h> 43 #include <asm/system.h> 44 #include <asm/ldt.h> 45 #include <asm/processor.h> 46 #include <asm/i387.h> 47 #include <asm/desc.h> 48 #ifdef CONFIG_MATH_EMULATION 49 #include <asm/math_emu.h> 50 #endif 51 52 #include <linux/err.h> 53 54 #include <asm/tlbflush.h> 55 #include <asm/cpu.h> 56 #include <asm/idle.h> 57 #include <asm/syscalls.h> 58 #include <asm/ds.h> 59 #include <asm/debugreg.h> 60 61 asmlinkage void ret_from_fork(void) __asm__("ret_from_fork"); 62 63 /* 64 * Return saved PC of a blocked thread. 65 */ 66 unsigned long thread_saved_pc(struct task_struct *tsk) 67 { 68 return ((unsigned long *)tsk->thread.sp)[3]; 69 } 70 71 #ifndef CONFIG_SMP 72 static inline void play_dead(void) 73 { 74 BUG(); 75 } 76 #endif 77 78 /* 79 * The idle thread. There's no useful work to be 80 * done, so just try to conserve power and have a 81 * low exit latency (ie sit in a loop waiting for 82 * somebody to say that they'd like to reschedule) 83 */ 84 void cpu_idle(void) 85 { 86 int cpu = smp_processor_id(); 87 88 /* 89 * If we're the non-boot CPU, nothing set the stack canary up 90 * for us. CPU0 already has it initialized but no harm in 91 * doing it again. This is a good place for updating it, as 92 * we wont ever return from this function (so the invalid 93 * canaries already on the stack wont ever trigger). 94 */ 95 boot_init_stack_canary(); 96 97 current_thread_info()->status |= TS_POLLING; 98 99 /* endless idle loop with no priority at all */ 100 while (1) { 101 tick_nohz_stop_sched_tick(1); 102 while (!need_resched()) { 103 104 check_pgt_cache(); 105 rmb(); 106 107 if (cpu_is_offline(cpu)) 108 play_dead(); 109 110 local_irq_disable(); 111 /* Don't trace irqs off for idle */ 112 stop_critical_timings(); 113 pm_idle(); 114 start_critical_timings(); 115 } 116 tick_nohz_restart_sched_tick(); 117 preempt_enable_no_resched(); 118 schedule(); 119 preempt_disable(); 120 } 121 } 122 123 void __show_regs(struct pt_regs *regs, int all) 124 { 125 unsigned long cr0 = 0L, cr2 = 0L, cr3 = 0L, cr4 = 0L; 126 unsigned long d0, d1, d2, d3, d6, d7; 127 unsigned long sp; 128 unsigned short ss, gs; 129 130 if (user_mode_vm(regs)) { 131 sp = regs->sp; 132 ss = regs->ss & 0xffff; 133 gs = get_user_gs(regs); 134 } else { 135 sp = kernel_stack_pointer(regs); 136 savesegment(ss, ss); 137 savesegment(gs, gs); 138 } 139 140 show_regs_common(); 141 142 printk("EIP: %04x:[<%08lx>] EFLAGS: %08lx CPU: %d\n", 143 (u16)regs->cs, regs->ip, regs->flags, 144 smp_processor_id()); 145 print_symbol("EIP is at %s\n", regs->ip); 146 147 printk("EAX: %08lx EBX: %08lx ECX: %08lx EDX: %08lx\n", 148 regs->ax, regs->bx, regs->cx, regs->dx); 149 printk("ESI: %08lx EDI: %08lx EBP: %08lx ESP: %08lx\n", 150 regs->si, regs->di, regs->bp, sp); 151 printk(" DS: %04x ES: %04x FS: %04x GS: %04x SS: %04x\n", 152 (u16)regs->ds, (u16)regs->es, (u16)regs->fs, gs, ss); 153 154 if (!all) 155 return; 156 157 cr0 = read_cr0(); 158 cr2 = read_cr2(); 159 cr3 = read_cr3(); 160 cr4 = read_cr4_safe(); 161 printk("CR0: %08lx CR2: %08lx CR3: %08lx CR4: %08lx\n", 162 cr0, cr2, cr3, cr4); 163 164 get_debugreg(d0, 0); 165 get_debugreg(d1, 1); 166 get_debugreg(d2, 2); 167 get_debugreg(d3, 3); 168 printk("DR0: %08lx DR1: %08lx DR2: %08lx DR3: %08lx\n", 169 d0, d1, d2, d3); 170 171 get_debugreg(d6, 6); 172 get_debugreg(d7, 7); 173 printk("DR6: %08lx DR7: %08lx\n", 174 d6, d7); 175 } 176 177 void show_regs(struct pt_regs *regs) 178 { 179 show_registers(regs); 180 show_trace(NULL, regs, ®s->sp, regs->bp); 181 } 182 183 /* 184 * This gets run with %bx containing the 185 * function to call, and %dx containing 186 * the "args". 187 */ 188 extern void kernel_thread_helper(void); 189 190 /* 191 * Create a kernel thread 192 */ 193 int kernel_thread(int (*fn)(void *), void *arg, unsigned long flags) 194 { 195 struct pt_regs regs; 196 197 memset(®s, 0, sizeof(regs)); 198 199 regs.bx = (unsigned long) fn; 200 regs.dx = (unsigned long) arg; 201 202 regs.ds = __USER_DS; 203 regs.es = __USER_DS; 204 regs.fs = __KERNEL_PERCPU; 205 regs.gs = __KERNEL_STACK_CANARY; 206 regs.orig_ax = -1; 207 regs.ip = (unsigned long) kernel_thread_helper; 208 regs.cs = __KERNEL_CS | get_kernel_rpl(); 209 regs.flags = X86_EFLAGS_IF | X86_EFLAGS_SF | X86_EFLAGS_PF | 0x2; 210 211 /* Ok, create the new process.. */ 212 return do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0, ®s, 0, NULL, NULL); 213 } 214 EXPORT_SYMBOL(kernel_thread); 215 216 void release_thread(struct task_struct *dead_task) 217 { 218 BUG_ON(dead_task->mm); 219 release_vm86_irqs(dead_task); 220 } 221 222 /* 223 * This gets called before we allocate a new thread and copy 224 * the current task into it. 225 */ 226 void prepare_to_copy(struct task_struct *tsk) 227 { 228 unlazy_fpu(tsk); 229 } 230 231 int copy_thread(unsigned long clone_flags, unsigned long sp, 232 unsigned long unused, 233 struct task_struct *p, struct pt_regs *regs) 234 { 235 struct pt_regs *childregs; 236 struct task_struct *tsk; 237 int err; 238 239 childregs = task_pt_regs(p); 240 *childregs = *regs; 241 childregs->ax = 0; 242 childregs->sp = sp; 243 244 p->thread.sp = (unsigned long) childregs; 245 p->thread.sp0 = (unsigned long) (childregs+1); 246 247 p->thread.ip = (unsigned long) ret_from_fork; 248 249 task_user_gs(p) = get_user_gs(regs); 250 251 p->thread.io_bitmap_ptr = NULL; 252 tsk = current; 253 err = -ENOMEM; 254 255 memset(p->thread.ptrace_bps, 0, sizeof(p->thread.ptrace_bps)); 256 257 if (unlikely(test_tsk_thread_flag(tsk, TIF_IO_BITMAP))) { 258 p->thread.io_bitmap_ptr = kmemdup(tsk->thread.io_bitmap_ptr, 259 IO_BITMAP_BYTES, GFP_KERNEL); 260 if (!p->thread.io_bitmap_ptr) { 261 p->thread.io_bitmap_max = 0; 262 return -ENOMEM; 263 } 264 set_tsk_thread_flag(p, TIF_IO_BITMAP); 265 } 266 267 err = 0; 268 269 /* 270 * Set a new TLS for the child thread? 271 */ 272 if (clone_flags & CLONE_SETTLS) 273 err = do_set_thread_area(p, -1, 274 (struct user_desc __user *)childregs->si, 0); 275 276 if (err && p->thread.io_bitmap_ptr) { 277 kfree(p->thread.io_bitmap_ptr); 278 p->thread.io_bitmap_max = 0; 279 } 280 281 clear_tsk_thread_flag(p, TIF_DS_AREA_MSR); 282 p->thread.ds_ctx = NULL; 283 284 clear_tsk_thread_flag(p, TIF_DEBUGCTLMSR); 285 p->thread.debugctlmsr = 0; 286 287 return err; 288 } 289 290 void 291 start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp) 292 { 293 set_user_gs(regs, 0); 294 regs->fs = 0; 295 set_fs(USER_DS); 296 regs->ds = __USER_DS; 297 regs->es = __USER_DS; 298 regs->ss = __USER_DS; 299 regs->cs = __USER_CS; 300 regs->ip = new_ip; 301 regs->sp = new_sp; 302 /* 303 * Free the old FP and other extended state 304 */ 305 free_thread_xstate(current); 306 } 307 EXPORT_SYMBOL_GPL(start_thread); 308 309 310 /* 311 * switch_to(x,yn) should switch tasks from x to y. 312 * 313 * We fsave/fwait so that an exception goes off at the right time 314 * (as a call from the fsave or fwait in effect) rather than to 315 * the wrong process. Lazy FP saving no longer makes any sense 316 * with modern CPU's, and this simplifies a lot of things (SMP 317 * and UP become the same). 318 * 319 * NOTE! We used to use the x86 hardware context switching. The 320 * reason for not using it any more becomes apparent when you 321 * try to recover gracefully from saved state that is no longer 322 * valid (stale segment register values in particular). With the 323 * hardware task-switch, there is no way to fix up bad state in 324 * a reasonable manner. 325 * 326 * The fact that Intel documents the hardware task-switching to 327 * be slow is a fairly red herring - this code is not noticeably 328 * faster. However, there _is_ some room for improvement here, 329 * so the performance issues may eventually be a valid point. 330 * More important, however, is the fact that this allows us much 331 * more flexibility. 332 * 333 * The return value (in %ax) will be the "prev" task after 334 * the task-switch, and shows up in ret_from_fork in entry.S, 335 * for example. 336 */ 337 __notrace_funcgraph struct task_struct * 338 __switch_to(struct task_struct *prev_p, struct task_struct *next_p) 339 { 340 struct thread_struct *prev = &prev_p->thread, 341 *next = &next_p->thread; 342 int cpu = smp_processor_id(); 343 struct tss_struct *tss = &per_cpu(init_tss, cpu); 344 bool preload_fpu; 345 346 /* never put a printk in __switch_to... printk() calls wake_up*() indirectly */ 347 348 /* 349 * If the task has used fpu the last 5 timeslices, just do a full 350 * restore of the math state immediately to avoid the trap; the 351 * chances of needing FPU soon are obviously high now 352 */ 353 preload_fpu = tsk_used_math(next_p) && next_p->fpu_counter > 5; 354 355 __unlazy_fpu(prev_p); 356 357 /* we're going to use this soon, after a few expensive things */ 358 if (preload_fpu) 359 prefetch(next->xstate); 360 361 /* 362 * Reload esp0. 363 */ 364 load_sp0(tss, next); 365 366 /* 367 * Save away %gs. No need to save %fs, as it was saved on the 368 * stack on entry. No need to save %es and %ds, as those are 369 * always kernel segments while inside the kernel. Doing this 370 * before setting the new TLS descriptors avoids the situation 371 * where we temporarily have non-reloadable segments in %fs 372 * and %gs. This could be an issue if the NMI handler ever 373 * used %fs or %gs (it does not today), or if the kernel is 374 * running inside of a hypervisor layer. 375 */ 376 lazy_save_gs(prev->gs); 377 378 /* 379 * Load the per-thread Thread-Local Storage descriptor. 380 */ 381 load_TLS(next, cpu); 382 383 /* 384 * Restore IOPL if needed. In normal use, the flags restore 385 * in the switch assembly will handle this. But if the kernel 386 * is running virtualized at a non-zero CPL, the popf will 387 * not restore flags, so it must be done in a separate step. 388 */ 389 if (get_kernel_rpl() && unlikely(prev->iopl != next->iopl)) 390 set_iopl_mask(next->iopl); 391 392 /* 393 * Now maybe handle debug registers and/or IO bitmaps 394 */ 395 if (unlikely(task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV || 396 task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT)) 397 __switch_to_xtra(prev_p, next_p, tss); 398 399 /* If we're going to preload the fpu context, make sure clts 400 is run while we're batching the cpu state updates. */ 401 if (preload_fpu) 402 clts(); 403 404 /* 405 * Leave lazy mode, flushing any hypercalls made here. 406 * This must be done before restoring TLS segments so 407 * the GDT and LDT are properly updated, and must be 408 * done before math_state_restore, so the TS bit is up 409 * to date. 410 */ 411 arch_end_context_switch(next_p); 412 413 if (preload_fpu) 414 __math_state_restore(); 415 416 /* 417 * Restore %gs if needed (which is common) 418 */ 419 if (prev->gs | next->gs) 420 lazy_load_gs(next->gs); 421 422 percpu_write(current_task, next_p); 423 424 return prev_p; 425 } 426 427 int sys_clone(struct pt_regs *regs) 428 { 429 unsigned long clone_flags; 430 unsigned long newsp; 431 int __user *parent_tidptr, *child_tidptr; 432 433 clone_flags = regs->bx; 434 newsp = regs->cx; 435 parent_tidptr = (int __user *)regs->dx; 436 child_tidptr = (int __user *)regs->di; 437 if (!newsp) 438 newsp = regs->sp; 439 return do_fork(clone_flags, newsp, regs, 0, parent_tidptr, child_tidptr); 440 } 441 442 /* 443 * sys_execve() executes a new program. 444 */ 445 int sys_execve(struct pt_regs *regs) 446 { 447 int error; 448 char *filename; 449 450 filename = getname((char __user *) regs->bx); 451 error = PTR_ERR(filename); 452 if (IS_ERR(filename)) 453 goto out; 454 error = do_execve(filename, 455 (char __user * __user *) regs->cx, 456 (char __user * __user *) regs->dx, 457 regs); 458 if (error == 0) { 459 /* Make sure we don't return using sysenter.. */ 460 set_thread_flag(TIF_IRET); 461 } 462 putname(filename); 463 out: 464 return error; 465 } 466 467 #define top_esp (THREAD_SIZE - sizeof(unsigned long)) 468 #define top_ebp (THREAD_SIZE - 2*sizeof(unsigned long)) 469 470 unsigned long get_wchan(struct task_struct *p) 471 { 472 unsigned long bp, sp, ip; 473 unsigned long stack_page; 474 int count = 0; 475 if (!p || p == current || p->state == TASK_RUNNING) 476 return 0; 477 stack_page = (unsigned long)task_stack_page(p); 478 sp = p->thread.sp; 479 if (!stack_page || sp < stack_page || sp > top_esp+stack_page) 480 return 0; 481 /* include/asm-i386/system.h:switch_to() pushes bp last. */ 482 bp = *(unsigned long *) sp; 483 do { 484 if (bp < stack_page || bp > top_ebp+stack_page) 485 return 0; 486 ip = *(unsigned long *) (bp+4); 487 if (!in_sched_functions(ip)) 488 return ip; 489 bp = *(unsigned long *) bp; 490 } while (count++ < 16); 491 return 0; 492 } 493 494