1 /* 2 * Copyright (C) 1995 Linus Torvalds 3 * 4 * Pentium III FXSR, SSE support 5 * Gareth Hughes <gareth@valinux.com>, May 2000 6 * 7 * X86-64 port 8 * Andi Kleen. 9 * 10 * CPU hotplug support - ashok.raj@intel.com 11 */ 12 13 /* 14 * This file handles the architecture-dependent parts of process handling.. 15 */ 16 17 #include <linux/cpu.h> 18 #include <linux/errno.h> 19 #include <linux/sched.h> 20 #include <linux/fs.h> 21 #include <linux/kernel.h> 22 #include <linux/mm.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> 36 #include <linux/io.h> 37 #include <linux/ftrace.h> 38 39 #include <asm/pgtable.h> 40 #include <asm/processor.h> 41 #include <asm/i387.h> 42 #include <asm/fpu-internal.h> 43 #include <asm/mmu_context.h> 44 #include <asm/prctl.h> 45 #include <asm/desc.h> 46 #include <asm/proto.h> 47 #include <asm/ia32.h> 48 #include <asm/idle.h> 49 #include <asm/syscalls.h> 50 #include <asm/debugreg.h> 51 #include <asm/switch_to.h> 52 53 asmlinkage extern void ret_from_fork(void); 54 55 DEFINE_PER_CPU(unsigned long, old_rsp); 56 57 /* Prints also some state that isn't saved in the pt_regs */ 58 void __show_regs(struct pt_regs *regs, int all) 59 { 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; 64 65 printk(KERN_DEFAULT "RIP: %04lx:[<%016lx>] ", regs->cs & 0xffff, regs->ip); 66 printk_address(regs->ip, 1); 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); 79 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)); 85 86 rdmsrl(MSR_FS_BASE, fs); 87 rdmsrl(MSR_GS_BASE, gs); 88 rdmsrl(MSR_KERNEL_GS_BASE, shadowgs); 89 90 if (!all) 91 return; 92 93 cr0 = read_cr0(); 94 cr2 = read_cr2(); 95 cr3 = read_cr3(); 96 cr4 = read_cr4(); 97 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, 101 es, cr0); 102 printk(KERN_DEFAULT "CR2: %016lx CR3: %016lx CR4: %016lx\n", cr2, cr3, 103 cr4); 104 105 get_debugreg(d0, 0); 106 get_debugreg(d1, 1); 107 get_debugreg(d2, 2); 108 printk(KERN_DEFAULT "DR0: %016lx DR1: %016lx DR2: %016lx\n", d0, d1, d2); 109 get_debugreg(d3, 3); 110 get_debugreg(d6, 6); 111 get_debugreg(d7, 7); 112 printk(KERN_DEFAULT "DR3: %016lx DR6: %016lx DR7: %016lx\n", d3, d6, d7); 113 } 114 115 void release_thread(struct task_struct *dead_task) 116 { 117 if (dead_task->mm) { 118 if (dead_task->mm->context.size) { 119 pr_warn("WARNING: dead process %s still has LDT? <%p/%d>\n", 120 dead_task->comm, 121 dead_task->mm->context.ldt, 122 dead_task->mm->context.size); 123 BUG(); 124 } 125 } 126 } 127 128 static inline void set_32bit_tls(struct task_struct *t, int tls, u32 addr) 129 { 130 struct user_desc ud = { 131 .base_addr = addr, 132 .limit = 0xfffff, 133 .seg_32bit = 1, 134 .limit_in_pages = 1, 135 .useable = 1, 136 }; 137 struct desc_struct *desc = t->thread.tls_array; 138 desc += tls; 139 fill_ldt(desc, &ud); 140 } 141 142 static inline u32 read_32bit_tls(struct task_struct *t, int tls) 143 { 144 return get_desc_base(&t->thread.tls_array[tls]); 145 } 146 147 int copy_thread(unsigned long clone_flags, unsigned long sp, 148 unsigned long arg, struct task_struct *p) 149 { 150 int err; 151 struct pt_regs *childregs; 152 struct task_struct *me = current; 153 154 p->thread.sp0 = (unsigned long)task_stack_page(p) + THREAD_SIZE; 155 childregs = task_pt_regs(p); 156 p->thread.sp = (unsigned long) childregs; 157 p->thread.usersp = me->thread.usersp; 158 set_tsk_thread_flag(p, TIF_FORK); 159 p->fpu_counter = 0; 160 p->thread.io_bitmap_ptr = NULL; 161 162 savesegment(gs, p->thread.gsindex); 163 p->thread.gs = p->thread.gsindex ? 0 : me->thread.gs; 164 savesegment(fs, p->thread.fsindex); 165 p->thread.fs = p->thread.fsindex ? 0 : me->thread.fs; 166 savesegment(es, p->thread.es); 167 savesegment(ds, p->thread.ds); 168 memset(p->thread.ptrace_bps, 0, sizeof(p->thread.ptrace_bps)); 169 170 if (unlikely(p->flags & PF_KTHREAD)) { 171 /* kernel thread */ 172 memset(childregs, 0, sizeof(struct pt_regs)); 173 childregs->sp = (unsigned long)childregs; 174 childregs->ss = __KERNEL_DS; 175 childregs->bx = sp; /* function */ 176 childregs->bp = arg; 177 childregs->orig_ax = -1; 178 childregs->cs = __KERNEL_CS | get_kernel_rpl(); 179 childregs->flags = X86_EFLAGS_IF | X86_EFLAGS_BIT1; 180 return 0; 181 } 182 *childregs = *current_pt_regs(); 183 184 childregs->ax = 0; 185 if (sp) 186 childregs->sp = sp; 187 188 err = -ENOMEM; 189 memset(p->thread.ptrace_bps, 0, sizeof(p->thread.ptrace_bps)); 190 191 if (unlikely(test_tsk_thread_flag(me, TIF_IO_BITMAP))) { 192 p->thread.io_bitmap_ptr = kmemdup(me->thread.io_bitmap_ptr, 193 IO_BITMAP_BYTES, GFP_KERNEL); 194 if (!p->thread.io_bitmap_ptr) { 195 p->thread.io_bitmap_max = 0; 196 return -ENOMEM; 197 } 198 set_tsk_thread_flag(p, TIF_IO_BITMAP); 199 } 200 201 /* 202 * Set a new TLS for the child thread? 203 */ 204 if (clone_flags & CLONE_SETTLS) { 205 #ifdef CONFIG_IA32_EMULATION 206 if (test_thread_flag(TIF_IA32)) 207 err = do_set_thread_area(p, -1, 208 (struct user_desc __user *)childregs->si, 0); 209 else 210 #endif 211 err = do_arch_prctl(p, ARCH_SET_FS, childregs->r8); 212 if (err) 213 goto out; 214 } 215 err = 0; 216 out: 217 if (err && p->thread.io_bitmap_ptr) { 218 kfree(p->thread.io_bitmap_ptr); 219 p->thread.io_bitmap_max = 0; 220 } 221 222 return err; 223 } 224 225 static void 226 start_thread_common(struct pt_regs *regs, unsigned long new_ip, 227 unsigned long new_sp, 228 unsigned int _cs, unsigned int _ss, unsigned int _ds) 229 { 230 loadsegment(fs, 0); 231 loadsegment(es, _ds); 232 loadsegment(ds, _ds); 233 load_gs_index(0); 234 current->thread.usersp = new_sp; 235 regs->ip = new_ip; 236 regs->sp = new_sp; 237 this_cpu_write(old_rsp, new_sp); 238 regs->cs = _cs; 239 regs->ss = _ss; 240 regs->flags = X86_EFLAGS_IF; 241 } 242 243 void 244 start_thread(struct pt_regs *regs, unsigned long new_ip, unsigned long new_sp) 245 { 246 start_thread_common(regs, new_ip, new_sp, 247 __USER_CS, __USER_DS, 0); 248 } 249 250 #ifdef CONFIG_IA32_EMULATION 251 void start_thread_ia32(struct pt_regs *regs, u32 new_ip, u32 new_sp) 252 { 253 start_thread_common(regs, new_ip, new_sp, 254 test_thread_flag(TIF_X32) 255 ? __USER_CS : __USER32_CS, 256 __USER_DS, __USER_DS); 257 } 258 #endif 259 260 /* 261 * switch_to(x,y) should switch tasks from x to y. 262 * 263 * This could still be optimized: 264 * - fold all the options into a flag word and test it with a single test. 265 * - could test fs/gs bitsliced 266 * 267 * Kprobes not supported here. Set the probe on schedule instead. 268 * Function graph tracer not supported too. 269 */ 270 __notrace_funcgraph struct task_struct * 271 __switch_to(struct task_struct *prev_p, struct task_struct *next_p) 272 { 273 struct thread_struct *prev = &prev_p->thread; 274 struct thread_struct *next = &next_p->thread; 275 int cpu = smp_processor_id(); 276 struct tss_struct *tss = &per_cpu(init_tss, cpu); 277 unsigned fsindex, gsindex; 278 fpu_switch_t fpu; 279 280 fpu = switch_fpu_prepare(prev_p, next_p, cpu); 281 282 /* 283 * Reload esp0, LDT and the page table pointer: 284 */ 285 load_sp0(tss, next); 286 287 /* 288 * Switch DS and ES. 289 * This won't pick up thread selector changes, but I guess that is ok. 290 */ 291 savesegment(es, prev->es); 292 if (unlikely(next->es | prev->es)) 293 loadsegment(es, next->es); 294 295 savesegment(ds, prev->ds); 296 if (unlikely(next->ds | prev->ds)) 297 loadsegment(ds, next->ds); 298 299 300 /* We must save %fs and %gs before load_TLS() because 301 * %fs and %gs may be cleared by load_TLS(). 302 * 303 * (e.g. xen_load_tls()) 304 */ 305 savesegment(fs, fsindex); 306 savesegment(gs, gsindex); 307 308 load_TLS(next, cpu); 309 310 /* 311 * Leave lazy mode, flushing any hypercalls made here. 312 * This must be done before restoring TLS segments so 313 * the GDT and LDT are properly updated, and must be 314 * done before math_state_restore, so the TS bit is up 315 * to date. 316 */ 317 arch_end_context_switch(next_p); 318 319 /* 320 * Switch FS and GS. 321 * 322 * Segment register != 0 always requires a reload. Also 323 * reload when it has changed. When prev process used 64bit 324 * base always reload to avoid an information leak. 325 */ 326 if (unlikely(fsindex | next->fsindex | prev->fs)) { 327 loadsegment(fs, next->fsindex); 328 /* 329 * Check if the user used a selector != 0; if yes 330 * clear 64bit base, since overloaded base is always 331 * mapped to the Null selector 332 */ 333 if (fsindex) 334 prev->fs = 0; 335 } 336 /* when next process has a 64bit base use it */ 337 if (next->fs) 338 wrmsrl(MSR_FS_BASE, next->fs); 339 prev->fsindex = fsindex; 340 341 if (unlikely(gsindex | next->gsindex | prev->gs)) { 342 load_gs_index(next->gsindex); 343 if (gsindex) 344 prev->gs = 0; 345 } 346 if (next->gs) 347 wrmsrl(MSR_KERNEL_GS_BASE, next->gs); 348 prev->gsindex = gsindex; 349 350 switch_fpu_finish(next_p, fpu); 351 352 /* 353 * Switch the PDA and FPU contexts. 354 */ 355 prev->usersp = this_cpu_read(old_rsp); 356 this_cpu_write(old_rsp, next->usersp); 357 this_cpu_write(current_task, next_p); 358 359 this_cpu_write(kernel_stack, 360 (unsigned long)task_stack_page(next_p) + 361 THREAD_SIZE - KERNEL_STACK_OFFSET); 362 363 /* 364 * Now maybe reload the debug registers and handle I/O bitmaps 365 */ 366 if (unlikely(task_thread_info(next_p)->flags & _TIF_WORK_CTXSW_NEXT || 367 task_thread_info(prev_p)->flags & _TIF_WORK_CTXSW_PREV)) 368 __switch_to_xtra(prev_p, next_p, tss); 369 370 return prev_p; 371 } 372 373 void set_personality_64bit(void) 374 { 375 /* inherit personality from parent */ 376 377 /* Make sure to be in 64bit mode */ 378 clear_thread_flag(TIF_IA32); 379 clear_thread_flag(TIF_ADDR32); 380 clear_thread_flag(TIF_X32); 381 382 /* Ensure the corresponding mm is not marked. */ 383 if (current->mm) 384 current->mm->context.ia32_compat = 0; 385 386 /* TBD: overwrites user setup. Should have two bits. 387 But 64bit processes have always behaved this way, 388 so it's not too bad. The main problem is just that 389 32bit childs are affected again. */ 390 current->personality &= ~READ_IMPLIES_EXEC; 391 } 392 393 void set_personality_ia32(bool x32) 394 { 395 /* inherit personality from parent */ 396 397 /* Make sure to be in 32bit mode */ 398 set_thread_flag(TIF_ADDR32); 399 400 /* Mark the associated mm as containing 32-bit tasks. */ 401 if (current->mm) 402 current->mm->context.ia32_compat = 1; 403 404 if (x32) { 405 clear_thread_flag(TIF_IA32); 406 set_thread_flag(TIF_X32); 407 current->personality &= ~READ_IMPLIES_EXEC; 408 /* is_compat_task() uses the presence of the x32 409 syscall bit flag to determine compat status */ 410 current_thread_info()->status &= ~TS_COMPAT; 411 } else { 412 set_thread_flag(TIF_IA32); 413 clear_thread_flag(TIF_X32); 414 current->personality |= force_personality32; 415 /* Prepare the first "return" to user space */ 416 current_thread_info()->status |= TS_COMPAT; 417 } 418 } 419 EXPORT_SYMBOL_GPL(set_personality_ia32); 420 421 unsigned long get_wchan(struct task_struct *p) 422 { 423 unsigned long stack; 424 u64 fp, ip; 425 int count = 0; 426 427 if (!p || p == current || p->state == TASK_RUNNING) 428 return 0; 429 stack = (unsigned long)task_stack_page(p); 430 if (p->thread.sp < stack || p->thread.sp >= stack+THREAD_SIZE) 431 return 0; 432 fp = *(u64 *)(p->thread.sp); 433 do { 434 if (fp < (unsigned long)stack || 435 fp >= (unsigned long)stack+THREAD_SIZE) 436 return 0; 437 ip = *(u64 *)(fp+8); 438 if (!in_sched_functions(ip)) 439 return ip; 440 fp = *(u64 *)fp; 441 } while (count++ < 16); 442 return 0; 443 } 444 445 long do_arch_prctl(struct task_struct *task, int code, unsigned long addr) 446 { 447 int ret = 0; 448 int doit = task == current; 449 int cpu; 450 451 switch (code) { 452 case ARCH_SET_GS: 453 if (addr >= TASK_SIZE_OF(task)) 454 return -EPERM; 455 cpu = get_cpu(); 456 /* handle small bases via the GDT because that's faster to 457 switch. */ 458 if (addr <= 0xffffffff) { 459 set_32bit_tls(task, GS_TLS, addr); 460 if (doit) { 461 load_TLS(&task->thread, cpu); 462 load_gs_index(GS_TLS_SEL); 463 } 464 task->thread.gsindex = GS_TLS_SEL; 465 task->thread.gs = 0; 466 } else { 467 task->thread.gsindex = 0; 468 task->thread.gs = addr; 469 if (doit) { 470 load_gs_index(0); 471 ret = wrmsrl_safe(MSR_KERNEL_GS_BASE, addr); 472 } 473 } 474 put_cpu(); 475 break; 476 case ARCH_SET_FS: 477 /* Not strictly needed for fs, but do it for symmetry 478 with gs */ 479 if (addr >= TASK_SIZE_OF(task)) 480 return -EPERM; 481 cpu = get_cpu(); 482 /* handle small bases via the GDT because that's faster to 483 switch. */ 484 if (addr <= 0xffffffff) { 485 set_32bit_tls(task, FS_TLS, addr); 486 if (doit) { 487 load_TLS(&task->thread, cpu); 488 loadsegment(fs, FS_TLS_SEL); 489 } 490 task->thread.fsindex = FS_TLS_SEL; 491 task->thread.fs = 0; 492 } else { 493 task->thread.fsindex = 0; 494 task->thread.fs = addr; 495 if (doit) { 496 /* set the selector to 0 to not confuse 497 __switch_to */ 498 loadsegment(fs, 0); 499 ret = wrmsrl_safe(MSR_FS_BASE, addr); 500 } 501 } 502 put_cpu(); 503 break; 504 case ARCH_GET_FS: { 505 unsigned long base; 506 if (task->thread.fsindex == FS_TLS_SEL) 507 base = read_32bit_tls(task, FS_TLS); 508 else if (doit) 509 rdmsrl(MSR_FS_BASE, base); 510 else 511 base = task->thread.fs; 512 ret = put_user(base, (unsigned long __user *)addr); 513 break; 514 } 515 case ARCH_GET_GS: { 516 unsigned long base; 517 unsigned gsindex; 518 if (task->thread.gsindex == GS_TLS_SEL) 519 base = read_32bit_tls(task, GS_TLS); 520 else if (doit) { 521 savesegment(gs, gsindex); 522 if (gsindex) 523 rdmsrl(MSR_KERNEL_GS_BASE, base); 524 else 525 base = task->thread.gs; 526 } else 527 base = task->thread.gs; 528 ret = put_user(base, (unsigned long __user *)addr); 529 break; 530 } 531 532 default: 533 ret = -EINVAL; 534 break; 535 } 536 537 return ret; 538 } 539 540 long sys_arch_prctl(int code, unsigned long addr) 541 { 542 return do_arch_prctl(current, code, addr); 543 } 544 545 unsigned long KSTK_ESP(struct task_struct *task) 546 { 547 return (test_tsk_thread_flag(task, TIF_IA32)) ? 548 (task_pt_regs(task)->sp) : ((task)->thread.usersp); 549 } 550