1 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 2 3 #include <linux/errno.h> 4 #include <linux/kernel.h> 5 #include <linux/mm.h> 6 #include <linux/smp.h> 7 #include <linux/prctl.h> 8 #include <linux/slab.h> 9 #include <linux/sched.h> 10 #include <linux/sched/idle.h> 11 #include <linux/sched/debug.h> 12 #include <linux/sched/task.h> 13 #include <linux/sched/task_stack.h> 14 #include <linux/init.h> 15 #include <linux/export.h> 16 #include <linux/pm.h> 17 #include <linux/tick.h> 18 #include <linux/random.h> 19 #include <linux/user-return-notifier.h> 20 #include <linux/dmi.h> 21 #include <linux/utsname.h> 22 #include <linux/stackprotector.h> 23 #include <linux/tick.h> 24 #include <linux/cpuidle.h> 25 #include <trace/events/power.h> 26 #include <linux/hw_breakpoint.h> 27 #include <asm/cpu.h> 28 #include <asm/apic.h> 29 #include <asm/syscalls.h> 30 #include <linux/uaccess.h> 31 #include <asm/mwait.h> 32 #include <asm/fpu/internal.h> 33 #include <asm/debugreg.h> 34 #include <asm/nmi.h> 35 #include <asm/tlbflush.h> 36 #include <asm/mce.h> 37 #include <asm/vm86.h> 38 #include <asm/switch_to.h> 39 #include <asm/desc.h> 40 #include <asm/prctl.h> 41 42 /* 43 * per-CPU TSS segments. Threads are completely 'soft' on Linux, 44 * no more per-task TSS's. The TSS size is kept cacheline-aligned 45 * so they are allowed to end up in the .data..cacheline_aligned 46 * section. Since TSS's are completely CPU-local, we want them 47 * on exact cacheline boundaries, to eliminate cacheline ping-pong. 48 */ 49 __visible DEFINE_PER_CPU_SHARED_ALIGNED(struct tss_struct, cpu_tss) = { 50 .x86_tss = { 51 .sp0 = TOP_OF_INIT_STACK, 52 #ifdef CONFIG_X86_32 53 .ss0 = __KERNEL_DS, 54 .ss1 = __KERNEL_CS, 55 .io_bitmap_base = INVALID_IO_BITMAP_OFFSET, 56 #endif 57 }, 58 #ifdef CONFIG_X86_32 59 /* 60 * Note that the .io_bitmap member must be extra-big. This is because 61 * the CPU will access an additional byte beyond the end of the IO 62 * permission bitmap. The extra byte must be all 1 bits, and must 63 * be within the limit. 64 */ 65 .io_bitmap = { [0 ... IO_BITMAP_LONGS] = ~0 }, 66 #endif 67 #ifdef CONFIG_X86_32 68 .SYSENTER_stack_canary = STACK_END_MAGIC, 69 #endif 70 }; 71 EXPORT_PER_CPU_SYMBOL(cpu_tss); 72 73 DEFINE_PER_CPU(bool, __tss_limit_invalid); 74 EXPORT_PER_CPU_SYMBOL_GPL(__tss_limit_invalid); 75 76 /* 77 * this gets called so that we can store lazy state into memory and copy the 78 * current task into the new thread. 79 */ 80 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src) 81 { 82 memcpy(dst, src, arch_task_struct_size); 83 #ifdef CONFIG_VM86 84 dst->thread.vm86 = NULL; 85 #endif 86 87 return fpu__copy(&dst->thread.fpu, &src->thread.fpu); 88 } 89 90 /* 91 * Free current thread data structures etc.. 92 */ 93 void exit_thread(struct task_struct *tsk) 94 { 95 struct thread_struct *t = &tsk->thread; 96 unsigned long *bp = t->io_bitmap_ptr; 97 struct fpu *fpu = &t->fpu; 98 99 if (bp) { 100 struct tss_struct *tss = &per_cpu(cpu_tss, get_cpu()); 101 102 t->io_bitmap_ptr = NULL; 103 clear_thread_flag(TIF_IO_BITMAP); 104 /* 105 * Careful, clear this in the TSS too: 106 */ 107 memset(tss->io_bitmap, 0xff, t->io_bitmap_max); 108 t->io_bitmap_max = 0; 109 put_cpu(); 110 kfree(bp); 111 } 112 113 free_vm86(t); 114 115 fpu__drop(fpu); 116 } 117 118 void flush_thread(void) 119 { 120 struct task_struct *tsk = current; 121 122 flush_ptrace_hw_breakpoint(tsk); 123 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array)); 124 125 fpu__clear(&tsk->thread.fpu); 126 } 127 128 void disable_TSC(void) 129 { 130 preempt_disable(); 131 if (!test_and_set_thread_flag(TIF_NOTSC)) 132 /* 133 * Must flip the CPU state synchronously with 134 * TIF_NOTSC in the current running context. 135 */ 136 cr4_set_bits(X86_CR4_TSD); 137 preempt_enable(); 138 } 139 140 static void enable_TSC(void) 141 { 142 preempt_disable(); 143 if (test_and_clear_thread_flag(TIF_NOTSC)) 144 /* 145 * Must flip the CPU state synchronously with 146 * TIF_NOTSC in the current running context. 147 */ 148 cr4_clear_bits(X86_CR4_TSD); 149 preempt_enable(); 150 } 151 152 int get_tsc_mode(unsigned long adr) 153 { 154 unsigned int val; 155 156 if (test_thread_flag(TIF_NOTSC)) 157 val = PR_TSC_SIGSEGV; 158 else 159 val = PR_TSC_ENABLE; 160 161 return put_user(val, (unsigned int __user *)adr); 162 } 163 164 int set_tsc_mode(unsigned int val) 165 { 166 if (val == PR_TSC_SIGSEGV) 167 disable_TSC(); 168 else if (val == PR_TSC_ENABLE) 169 enable_TSC(); 170 else 171 return -EINVAL; 172 173 return 0; 174 } 175 176 DEFINE_PER_CPU(u64, msr_misc_features_shadow); 177 178 static void set_cpuid_faulting(bool on) 179 { 180 u64 msrval; 181 182 msrval = this_cpu_read(msr_misc_features_shadow); 183 msrval &= ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT; 184 msrval |= (on << MSR_MISC_FEATURES_ENABLES_CPUID_FAULT_BIT); 185 this_cpu_write(msr_misc_features_shadow, msrval); 186 wrmsrl(MSR_MISC_FEATURES_ENABLES, msrval); 187 } 188 189 static void disable_cpuid(void) 190 { 191 preempt_disable(); 192 if (!test_and_set_thread_flag(TIF_NOCPUID)) { 193 /* 194 * Must flip the CPU state synchronously with 195 * TIF_NOCPUID in the current running context. 196 */ 197 set_cpuid_faulting(true); 198 } 199 preempt_enable(); 200 } 201 202 static void enable_cpuid(void) 203 { 204 preempt_disable(); 205 if (test_and_clear_thread_flag(TIF_NOCPUID)) { 206 /* 207 * Must flip the CPU state synchronously with 208 * TIF_NOCPUID in the current running context. 209 */ 210 set_cpuid_faulting(false); 211 } 212 preempt_enable(); 213 } 214 215 static int get_cpuid_mode(void) 216 { 217 return !test_thread_flag(TIF_NOCPUID); 218 } 219 220 static int set_cpuid_mode(struct task_struct *task, unsigned long cpuid_enabled) 221 { 222 if (!static_cpu_has(X86_FEATURE_CPUID_FAULT)) 223 return -ENODEV; 224 225 if (cpuid_enabled) 226 enable_cpuid(); 227 else 228 disable_cpuid(); 229 230 return 0; 231 } 232 233 /* 234 * Called immediately after a successful exec. 235 */ 236 void arch_setup_new_exec(void) 237 { 238 /* If cpuid was previously disabled for this task, re-enable it. */ 239 if (test_thread_flag(TIF_NOCPUID)) 240 enable_cpuid(); 241 } 242 243 static inline void switch_to_bitmap(struct tss_struct *tss, 244 struct thread_struct *prev, 245 struct thread_struct *next, 246 unsigned long tifp, unsigned long tifn) 247 { 248 if (tifn & _TIF_IO_BITMAP) { 249 /* 250 * Copy the relevant range of the IO bitmap. 251 * Normally this is 128 bytes or less: 252 */ 253 memcpy(tss->io_bitmap, next->io_bitmap_ptr, 254 max(prev->io_bitmap_max, next->io_bitmap_max)); 255 /* 256 * Make sure that the TSS limit is correct for the CPU 257 * to notice the IO bitmap. 258 */ 259 refresh_tss_limit(); 260 } else if (tifp & _TIF_IO_BITMAP) { 261 /* 262 * Clear any possible leftover bits: 263 */ 264 memset(tss->io_bitmap, 0xff, prev->io_bitmap_max); 265 } 266 } 267 268 void __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p, 269 struct tss_struct *tss) 270 { 271 struct thread_struct *prev, *next; 272 unsigned long tifp, tifn; 273 274 prev = &prev_p->thread; 275 next = &next_p->thread; 276 277 tifn = READ_ONCE(task_thread_info(next_p)->flags); 278 tifp = READ_ONCE(task_thread_info(prev_p)->flags); 279 switch_to_bitmap(tss, prev, next, tifp, tifn); 280 281 propagate_user_return_notify(prev_p, next_p); 282 283 if ((tifp & _TIF_BLOCKSTEP || tifn & _TIF_BLOCKSTEP) && 284 arch_has_block_step()) { 285 unsigned long debugctl, msk; 286 287 rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctl); 288 debugctl &= ~DEBUGCTLMSR_BTF; 289 msk = tifn & _TIF_BLOCKSTEP; 290 debugctl |= (msk >> TIF_BLOCKSTEP) << DEBUGCTLMSR_BTF_SHIFT; 291 wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctl); 292 } 293 294 if ((tifp ^ tifn) & _TIF_NOTSC) 295 cr4_toggle_bits(X86_CR4_TSD); 296 297 if ((tifp ^ tifn) & _TIF_NOCPUID) 298 set_cpuid_faulting(!!(tifn & _TIF_NOCPUID)); 299 } 300 301 /* 302 * Idle related variables and functions 303 */ 304 unsigned long boot_option_idle_override = IDLE_NO_OVERRIDE; 305 EXPORT_SYMBOL(boot_option_idle_override); 306 307 static void (*x86_idle)(void); 308 309 #ifndef CONFIG_SMP 310 static inline void play_dead(void) 311 { 312 BUG(); 313 } 314 #endif 315 316 void arch_cpu_idle_enter(void) 317 { 318 tsc_verify_tsc_adjust(false); 319 local_touch_nmi(); 320 } 321 322 void arch_cpu_idle_dead(void) 323 { 324 play_dead(); 325 } 326 327 /* 328 * Called from the generic idle code. 329 */ 330 void arch_cpu_idle(void) 331 { 332 x86_idle(); 333 } 334 335 /* 336 * We use this if we don't have any better idle routine.. 337 */ 338 void __cpuidle default_idle(void) 339 { 340 trace_cpu_idle_rcuidle(1, smp_processor_id()); 341 safe_halt(); 342 trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, smp_processor_id()); 343 } 344 #ifdef CONFIG_APM_MODULE 345 EXPORT_SYMBOL(default_idle); 346 #endif 347 348 #ifdef CONFIG_XEN 349 bool xen_set_default_idle(void) 350 { 351 bool ret = !!x86_idle; 352 353 x86_idle = default_idle; 354 355 return ret; 356 } 357 #endif 358 359 void stop_this_cpu(void *dummy) 360 { 361 local_irq_disable(); 362 /* 363 * Remove this CPU: 364 */ 365 set_cpu_online(smp_processor_id(), false); 366 disable_local_APIC(); 367 mcheck_cpu_clear(this_cpu_ptr(&cpu_info)); 368 369 for (;;) { 370 /* 371 * Use wbinvd followed by hlt to stop the processor. This 372 * provides support for kexec on a processor that supports 373 * SME. With kexec, going from SME inactive to SME active 374 * requires clearing cache entries so that addresses without 375 * the encryption bit set don't corrupt the same physical 376 * address that has the encryption bit set when caches are 377 * flushed. To achieve this a wbinvd is performed followed by 378 * a hlt. Even if the processor is not in the kexec/SME 379 * scenario this only adds a wbinvd to a halting processor. 380 */ 381 asm volatile("wbinvd; hlt" : : : "memory"); 382 } 383 } 384 385 /* 386 * AMD Erratum 400 aware idle routine. We handle it the same way as C3 power 387 * states (local apic timer and TSC stop). 388 */ 389 static void amd_e400_idle(void) 390 { 391 /* 392 * We cannot use static_cpu_has_bug() here because X86_BUG_AMD_APIC_C1E 393 * gets set after static_cpu_has() places have been converted via 394 * alternatives. 395 */ 396 if (!boot_cpu_has_bug(X86_BUG_AMD_APIC_C1E)) { 397 default_idle(); 398 return; 399 } 400 401 tick_broadcast_enter(); 402 403 default_idle(); 404 405 /* 406 * The switch back from broadcast mode needs to be called with 407 * interrupts disabled. 408 */ 409 local_irq_disable(); 410 tick_broadcast_exit(); 411 local_irq_enable(); 412 } 413 414 /* 415 * Intel Core2 and older machines prefer MWAIT over HALT for C1. 416 * We can't rely on cpuidle installing MWAIT, because it will not load 417 * on systems that support only C1 -- so the boot default must be MWAIT. 418 * 419 * Some AMD machines are the opposite, they depend on using HALT. 420 * 421 * So for default C1, which is used during boot until cpuidle loads, 422 * use MWAIT-C1 on Intel HW that has it, else use HALT. 423 */ 424 static int prefer_mwait_c1_over_halt(const struct cpuinfo_x86 *c) 425 { 426 if (c->x86_vendor != X86_VENDOR_INTEL) 427 return 0; 428 429 if (!cpu_has(c, X86_FEATURE_MWAIT) || static_cpu_has_bug(X86_BUG_MONITOR)) 430 return 0; 431 432 return 1; 433 } 434 435 /* 436 * MONITOR/MWAIT with no hints, used for default C1 state. This invokes MWAIT 437 * with interrupts enabled and no flags, which is backwards compatible with the 438 * original MWAIT implementation. 439 */ 440 static __cpuidle void mwait_idle(void) 441 { 442 if (!current_set_polling_and_test()) { 443 trace_cpu_idle_rcuidle(1, smp_processor_id()); 444 if (this_cpu_has(X86_BUG_CLFLUSH_MONITOR)) { 445 mb(); /* quirk */ 446 clflush((void *)¤t_thread_info()->flags); 447 mb(); /* quirk */ 448 } 449 450 __monitor((void *)¤t_thread_info()->flags, 0, 0); 451 if (!need_resched()) 452 __sti_mwait(0, 0); 453 else 454 local_irq_enable(); 455 trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, smp_processor_id()); 456 } else { 457 local_irq_enable(); 458 } 459 __current_clr_polling(); 460 } 461 462 void select_idle_routine(const struct cpuinfo_x86 *c) 463 { 464 #ifdef CONFIG_SMP 465 if (boot_option_idle_override == IDLE_POLL && smp_num_siblings > 1) 466 pr_warn_once("WARNING: polling idle and HT enabled, performance may degrade\n"); 467 #endif 468 if (x86_idle || boot_option_idle_override == IDLE_POLL) 469 return; 470 471 if (boot_cpu_has_bug(X86_BUG_AMD_E400)) { 472 pr_info("using AMD E400 aware idle routine\n"); 473 x86_idle = amd_e400_idle; 474 } else if (prefer_mwait_c1_over_halt(c)) { 475 pr_info("using mwait in idle threads\n"); 476 x86_idle = mwait_idle; 477 } else 478 x86_idle = default_idle; 479 } 480 481 void amd_e400_c1e_apic_setup(void) 482 { 483 if (boot_cpu_has_bug(X86_BUG_AMD_APIC_C1E)) { 484 pr_info("Switch to broadcast mode on CPU%d\n", smp_processor_id()); 485 local_irq_disable(); 486 tick_broadcast_force(); 487 local_irq_enable(); 488 } 489 } 490 491 void __init arch_post_acpi_subsys_init(void) 492 { 493 u32 lo, hi; 494 495 if (!boot_cpu_has_bug(X86_BUG_AMD_E400)) 496 return; 497 498 /* 499 * AMD E400 detection needs to happen after ACPI has been enabled. If 500 * the machine is affected K8_INTP_C1E_ACTIVE_MASK bits are set in 501 * MSR_K8_INT_PENDING_MSG. 502 */ 503 rdmsr(MSR_K8_INT_PENDING_MSG, lo, hi); 504 if (!(lo & K8_INTP_C1E_ACTIVE_MASK)) 505 return; 506 507 boot_cpu_set_bug(X86_BUG_AMD_APIC_C1E); 508 509 if (!boot_cpu_has(X86_FEATURE_NONSTOP_TSC)) 510 mark_tsc_unstable("TSC halt in AMD C1E"); 511 pr_info("System has AMD C1E enabled\n"); 512 } 513 514 static int __init idle_setup(char *str) 515 { 516 if (!str) 517 return -EINVAL; 518 519 if (!strcmp(str, "poll")) { 520 pr_info("using polling idle threads\n"); 521 boot_option_idle_override = IDLE_POLL; 522 cpu_idle_poll_ctrl(true); 523 } else if (!strcmp(str, "halt")) { 524 /* 525 * When the boot option of idle=halt is added, halt is 526 * forced to be used for CPU idle. In such case CPU C2/C3 527 * won't be used again. 528 * To continue to load the CPU idle driver, don't touch 529 * the boot_option_idle_override. 530 */ 531 x86_idle = default_idle; 532 boot_option_idle_override = IDLE_HALT; 533 } else if (!strcmp(str, "nomwait")) { 534 /* 535 * If the boot option of "idle=nomwait" is added, 536 * it means that mwait will be disabled for CPU C2/C3 537 * states. In such case it won't touch the variable 538 * of boot_option_idle_override. 539 */ 540 boot_option_idle_override = IDLE_NOMWAIT; 541 } else 542 return -1; 543 544 return 0; 545 } 546 early_param("idle", idle_setup); 547 548 unsigned long arch_align_stack(unsigned long sp) 549 { 550 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space) 551 sp -= get_random_int() % 8192; 552 return sp & ~0xf; 553 } 554 555 unsigned long arch_randomize_brk(struct mm_struct *mm) 556 { 557 return randomize_page(mm->brk, 0x02000000); 558 } 559 560 /* 561 * Called from fs/proc with a reference on @p to find the function 562 * which called into schedule(). This needs to be done carefully 563 * because the task might wake up and we might look at a stack 564 * changing under us. 565 */ 566 unsigned long get_wchan(struct task_struct *p) 567 { 568 unsigned long start, bottom, top, sp, fp, ip, ret = 0; 569 int count = 0; 570 571 if (!p || p == current || p->state == TASK_RUNNING) 572 return 0; 573 574 if (!try_get_task_stack(p)) 575 return 0; 576 577 start = (unsigned long)task_stack_page(p); 578 if (!start) 579 goto out; 580 581 /* 582 * Layout of the stack page: 583 * 584 * ----------- topmax = start + THREAD_SIZE - sizeof(unsigned long) 585 * PADDING 586 * ----------- top = topmax - TOP_OF_KERNEL_STACK_PADDING 587 * stack 588 * ----------- bottom = start 589 * 590 * The tasks stack pointer points at the location where the 591 * framepointer is stored. The data on the stack is: 592 * ... IP FP ... IP FP 593 * 594 * We need to read FP and IP, so we need to adjust the upper 595 * bound by another unsigned long. 596 */ 597 top = start + THREAD_SIZE - TOP_OF_KERNEL_STACK_PADDING; 598 top -= 2 * sizeof(unsigned long); 599 bottom = start; 600 601 sp = READ_ONCE(p->thread.sp); 602 if (sp < bottom || sp > top) 603 goto out; 604 605 fp = READ_ONCE_NOCHECK(((struct inactive_task_frame *)sp)->bp); 606 do { 607 if (fp < bottom || fp > top) 608 goto out; 609 ip = READ_ONCE_NOCHECK(*(unsigned long *)(fp + sizeof(unsigned long))); 610 if (!in_sched_functions(ip)) { 611 ret = ip; 612 goto out; 613 } 614 fp = READ_ONCE_NOCHECK(*(unsigned long *)fp); 615 } while (count++ < 16 && p->state != TASK_RUNNING); 616 617 out: 618 put_task_stack(p); 619 return ret; 620 } 621 622 long do_arch_prctl_common(struct task_struct *task, int option, 623 unsigned long cpuid_enabled) 624 { 625 switch (option) { 626 case ARCH_GET_CPUID: 627 return get_cpuid_mode(); 628 case ARCH_SET_CPUID: 629 return set_cpuid_mode(task, cpuid_enabled); 630 } 631 632 return -EINVAL; 633 } 634