1 // SPDX-License-Identifier: GPL-2.0 2 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 3 4 #include <linux/errno.h> 5 #include <linux/kernel.h> 6 #include <linux/mm.h> 7 #include <linux/smp.h> 8 #include <linux/prctl.h> 9 #include <linux/slab.h> 10 #include <linux/sched.h> 11 #include <linux/sched/idle.h> 12 #include <linux/sched/debug.h> 13 #include <linux/sched/task.h> 14 #include <linux/sched/task_stack.h> 15 #include <linux/init.h> 16 #include <linux/export.h> 17 #include <linux/pm.h> 18 #include <linux/tick.h> 19 #include <linux/random.h> 20 #include <linux/user-return-notifier.h> 21 #include <linux/dmi.h> 22 #include <linux/utsname.h> 23 #include <linux/stackprotector.h> 24 #include <linux/cpuidle.h> 25 #include <linux/acpi.h> 26 #include <linux/elf-randomize.h> 27 #include <trace/events/power.h> 28 #include <linux/hw_breakpoint.h> 29 #include <asm/cpu.h> 30 #include <asm/apic.h> 31 #include <linux/uaccess.h> 32 #include <asm/mwait.h> 33 #include <asm/fpu/internal.h> 34 #include <asm/debugreg.h> 35 #include <asm/nmi.h> 36 #include <asm/tlbflush.h> 37 #include <asm/mce.h> 38 #include <asm/vm86.h> 39 #include <asm/switch_to.h> 40 #include <asm/desc.h> 41 #include <asm/prctl.h> 42 #include <asm/spec-ctrl.h> 43 #include <asm/io_bitmap.h> 44 #include <asm/proto.h> 45 46 #include "process.h" 47 48 /* 49 * per-CPU TSS segments. Threads are completely 'soft' on Linux, 50 * no more per-task TSS's. The TSS size is kept cacheline-aligned 51 * so they are allowed to end up in the .data..cacheline_aligned 52 * section. Since TSS's are completely CPU-local, we want them 53 * on exact cacheline boundaries, to eliminate cacheline ping-pong. 54 */ 55 __visible DEFINE_PER_CPU_PAGE_ALIGNED(struct tss_struct, cpu_tss_rw) = { 56 .x86_tss = { 57 /* 58 * .sp0 is only used when entering ring 0 from a lower 59 * privilege level. Since the init task never runs anything 60 * but ring 0 code, there is no need for a valid value here. 61 * Poison it. 62 */ 63 .sp0 = (1UL << (BITS_PER_LONG-1)) + 1, 64 65 /* 66 * .sp1 is cpu_current_top_of_stack. The init task never 67 * runs user code, but cpu_current_top_of_stack should still 68 * be well defined before the first context switch. 69 */ 70 .sp1 = TOP_OF_INIT_STACK, 71 72 #ifdef CONFIG_X86_32 73 .ss0 = __KERNEL_DS, 74 .ss1 = __KERNEL_CS, 75 #endif 76 .io_bitmap_base = IO_BITMAP_OFFSET_INVALID, 77 }, 78 }; 79 EXPORT_PER_CPU_SYMBOL(cpu_tss_rw); 80 81 DEFINE_PER_CPU(bool, __tss_limit_invalid); 82 EXPORT_PER_CPU_SYMBOL_GPL(__tss_limit_invalid); 83 84 /* 85 * this gets called so that we can store lazy state into memory and copy the 86 * current task into the new thread. 87 */ 88 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src) 89 { 90 memcpy(dst, src, arch_task_struct_size); 91 #ifdef CONFIG_VM86 92 dst->thread.vm86 = NULL; 93 #endif 94 95 return fpu__copy(dst, src); 96 } 97 98 /* 99 * Free thread data structures etc.. 100 */ 101 void exit_thread(struct task_struct *tsk) 102 { 103 struct thread_struct *t = &tsk->thread; 104 struct fpu *fpu = &t->fpu; 105 106 if (test_thread_flag(TIF_IO_BITMAP)) 107 io_bitmap_exit(tsk); 108 109 free_vm86(t); 110 111 fpu__drop(fpu); 112 } 113 114 static int set_new_tls(struct task_struct *p, unsigned long tls) 115 { 116 struct user_desc __user *utls = (struct user_desc __user *)tls; 117 118 if (in_ia32_syscall()) 119 return do_set_thread_area(p, -1, utls, 0); 120 else 121 return do_set_thread_area_64(p, ARCH_SET_FS, tls); 122 } 123 124 int copy_thread(unsigned long clone_flags, unsigned long sp, unsigned long arg, 125 struct task_struct *p, unsigned long tls) 126 { 127 struct inactive_task_frame *frame; 128 struct fork_frame *fork_frame; 129 struct pt_regs *childregs; 130 int ret = 0; 131 132 childregs = task_pt_regs(p); 133 fork_frame = container_of(childregs, struct fork_frame, regs); 134 frame = &fork_frame->frame; 135 136 frame->bp = 0; 137 frame->ret_addr = (unsigned long) ret_from_fork; 138 p->thread.sp = (unsigned long) fork_frame; 139 p->thread.io_bitmap = NULL; 140 memset(p->thread.ptrace_bps, 0, sizeof(p->thread.ptrace_bps)); 141 142 #ifdef CONFIG_X86_64 143 savesegment(gs, p->thread.gsindex); 144 p->thread.gsbase = p->thread.gsindex ? 0 : current->thread.gsbase; 145 savesegment(fs, p->thread.fsindex); 146 p->thread.fsbase = p->thread.fsindex ? 0 : current->thread.fsbase; 147 savesegment(es, p->thread.es); 148 savesegment(ds, p->thread.ds); 149 #else 150 p->thread.sp0 = (unsigned long) (childregs + 1); 151 /* 152 * Clear all status flags including IF and set fixed bit. 64bit 153 * does not have this initialization as the frame does not contain 154 * flags. The flags consistency (especially vs. AC) is there 155 * ensured via objtool, which lacks 32bit support. 156 */ 157 frame->flags = X86_EFLAGS_FIXED; 158 #endif 159 160 /* Kernel thread ? */ 161 if (unlikely(p->flags & PF_KTHREAD)) { 162 memset(childregs, 0, sizeof(struct pt_regs)); 163 kthread_frame_init(frame, sp, arg); 164 return 0; 165 } 166 167 frame->bx = 0; 168 *childregs = *current_pt_regs(); 169 childregs->ax = 0; 170 if (sp) 171 childregs->sp = sp; 172 173 #ifdef CONFIG_X86_32 174 task_user_gs(p) = get_user_gs(current_pt_regs()); 175 #endif 176 177 /* Set a new TLS for the child thread? */ 178 if (clone_flags & CLONE_SETTLS) 179 ret = set_new_tls(p, tls); 180 181 if (!ret && unlikely(test_tsk_thread_flag(current, TIF_IO_BITMAP))) 182 io_bitmap_share(p); 183 184 return ret; 185 } 186 187 void flush_thread(void) 188 { 189 struct task_struct *tsk = current; 190 191 flush_ptrace_hw_breakpoint(tsk); 192 memset(tsk->thread.tls_array, 0, sizeof(tsk->thread.tls_array)); 193 194 fpu__clear_all(&tsk->thread.fpu); 195 } 196 197 void disable_TSC(void) 198 { 199 preempt_disable(); 200 if (!test_and_set_thread_flag(TIF_NOTSC)) 201 /* 202 * Must flip the CPU state synchronously with 203 * TIF_NOTSC in the current running context. 204 */ 205 cr4_set_bits(X86_CR4_TSD); 206 preempt_enable(); 207 } 208 209 static void enable_TSC(void) 210 { 211 preempt_disable(); 212 if (test_and_clear_thread_flag(TIF_NOTSC)) 213 /* 214 * Must flip the CPU state synchronously with 215 * TIF_NOTSC in the current running context. 216 */ 217 cr4_clear_bits(X86_CR4_TSD); 218 preempt_enable(); 219 } 220 221 int get_tsc_mode(unsigned long adr) 222 { 223 unsigned int val; 224 225 if (test_thread_flag(TIF_NOTSC)) 226 val = PR_TSC_SIGSEGV; 227 else 228 val = PR_TSC_ENABLE; 229 230 return put_user(val, (unsigned int __user *)adr); 231 } 232 233 int set_tsc_mode(unsigned int val) 234 { 235 if (val == PR_TSC_SIGSEGV) 236 disable_TSC(); 237 else if (val == PR_TSC_ENABLE) 238 enable_TSC(); 239 else 240 return -EINVAL; 241 242 return 0; 243 } 244 245 DEFINE_PER_CPU(u64, msr_misc_features_shadow); 246 247 static void set_cpuid_faulting(bool on) 248 { 249 u64 msrval; 250 251 msrval = this_cpu_read(msr_misc_features_shadow); 252 msrval &= ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT; 253 msrval |= (on << MSR_MISC_FEATURES_ENABLES_CPUID_FAULT_BIT); 254 this_cpu_write(msr_misc_features_shadow, msrval); 255 wrmsrl(MSR_MISC_FEATURES_ENABLES, msrval); 256 } 257 258 static void disable_cpuid(void) 259 { 260 preempt_disable(); 261 if (!test_and_set_thread_flag(TIF_NOCPUID)) { 262 /* 263 * Must flip the CPU state synchronously with 264 * TIF_NOCPUID in the current running context. 265 */ 266 set_cpuid_faulting(true); 267 } 268 preempt_enable(); 269 } 270 271 static void enable_cpuid(void) 272 { 273 preempt_disable(); 274 if (test_and_clear_thread_flag(TIF_NOCPUID)) { 275 /* 276 * Must flip the CPU state synchronously with 277 * TIF_NOCPUID in the current running context. 278 */ 279 set_cpuid_faulting(false); 280 } 281 preempt_enable(); 282 } 283 284 static int get_cpuid_mode(void) 285 { 286 return !test_thread_flag(TIF_NOCPUID); 287 } 288 289 static int set_cpuid_mode(struct task_struct *task, unsigned long cpuid_enabled) 290 { 291 if (!boot_cpu_has(X86_FEATURE_CPUID_FAULT)) 292 return -ENODEV; 293 294 if (cpuid_enabled) 295 enable_cpuid(); 296 else 297 disable_cpuid(); 298 299 return 0; 300 } 301 302 /* 303 * Called immediately after a successful exec. 304 */ 305 void arch_setup_new_exec(void) 306 { 307 /* If cpuid was previously disabled for this task, re-enable it. */ 308 if (test_thread_flag(TIF_NOCPUID)) 309 enable_cpuid(); 310 311 /* 312 * Don't inherit TIF_SSBD across exec boundary when 313 * PR_SPEC_DISABLE_NOEXEC is used. 314 */ 315 if (test_thread_flag(TIF_SSBD) && 316 task_spec_ssb_noexec(current)) { 317 clear_thread_flag(TIF_SSBD); 318 task_clear_spec_ssb_disable(current); 319 task_clear_spec_ssb_noexec(current); 320 speculation_ctrl_update(task_thread_info(current)->flags); 321 } 322 } 323 324 #ifdef CONFIG_X86_IOPL_IOPERM 325 static inline void tss_invalidate_io_bitmap(struct tss_struct *tss) 326 { 327 /* 328 * Invalidate the I/O bitmap by moving io_bitmap_base outside the 329 * TSS limit so any subsequent I/O access from user space will 330 * trigger a #GP. 331 * 332 * This is correct even when VMEXIT rewrites the TSS limit 333 * to 0x67 as the only requirement is that the base points 334 * outside the limit. 335 */ 336 tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET_INVALID; 337 } 338 339 static inline void switch_to_bitmap(unsigned long tifp) 340 { 341 /* 342 * Invalidate I/O bitmap if the previous task used it. This prevents 343 * any possible leakage of an active I/O bitmap. 344 * 345 * If the next task has an I/O bitmap it will handle it on exit to 346 * user mode. 347 */ 348 if (tifp & _TIF_IO_BITMAP) 349 tss_invalidate_io_bitmap(this_cpu_ptr(&cpu_tss_rw)); 350 } 351 352 static void tss_copy_io_bitmap(struct tss_struct *tss, struct io_bitmap *iobm) 353 { 354 /* 355 * Copy at least the byte range of the incoming tasks bitmap which 356 * covers the permitted I/O ports. 357 * 358 * If the previous task which used an I/O bitmap had more bits 359 * permitted, then the copy needs to cover those as well so they 360 * get turned off. 361 */ 362 memcpy(tss->io_bitmap.bitmap, iobm->bitmap, 363 max(tss->io_bitmap.prev_max, iobm->max)); 364 365 /* 366 * Store the new max and the sequence number of this bitmap 367 * and a pointer to the bitmap itself. 368 */ 369 tss->io_bitmap.prev_max = iobm->max; 370 tss->io_bitmap.prev_sequence = iobm->sequence; 371 } 372 373 /** 374 * tss_update_io_bitmap - Update I/O bitmap before exiting to usermode 375 */ 376 void native_tss_update_io_bitmap(void) 377 { 378 struct tss_struct *tss = this_cpu_ptr(&cpu_tss_rw); 379 struct thread_struct *t = ¤t->thread; 380 u16 *base = &tss->x86_tss.io_bitmap_base; 381 382 if (!test_thread_flag(TIF_IO_BITMAP)) { 383 tss_invalidate_io_bitmap(tss); 384 return; 385 } 386 387 if (IS_ENABLED(CONFIG_X86_IOPL_IOPERM) && t->iopl_emul == 3) { 388 *base = IO_BITMAP_OFFSET_VALID_ALL; 389 } else { 390 struct io_bitmap *iobm = t->io_bitmap; 391 392 /* 393 * Only copy bitmap data when the sequence number differs. The 394 * update time is accounted to the incoming task. 395 */ 396 if (tss->io_bitmap.prev_sequence != iobm->sequence) 397 tss_copy_io_bitmap(tss, iobm); 398 399 /* Enable the bitmap */ 400 *base = IO_BITMAP_OFFSET_VALID_MAP; 401 } 402 403 /* 404 * Make sure that the TSS limit is covering the IO bitmap. It might have 405 * been cut down by a VMEXIT to 0x67 which would cause a subsequent I/O 406 * access from user space to trigger a #GP because tbe bitmap is outside 407 * the TSS limit. 408 */ 409 refresh_tss_limit(); 410 } 411 #else /* CONFIG_X86_IOPL_IOPERM */ 412 static inline void switch_to_bitmap(unsigned long tifp) { } 413 #endif 414 415 #ifdef CONFIG_SMP 416 417 struct ssb_state { 418 struct ssb_state *shared_state; 419 raw_spinlock_t lock; 420 unsigned int disable_state; 421 unsigned long local_state; 422 }; 423 424 #define LSTATE_SSB 0 425 426 static DEFINE_PER_CPU(struct ssb_state, ssb_state); 427 428 void speculative_store_bypass_ht_init(void) 429 { 430 struct ssb_state *st = this_cpu_ptr(&ssb_state); 431 unsigned int this_cpu = smp_processor_id(); 432 unsigned int cpu; 433 434 st->local_state = 0; 435 436 /* 437 * Shared state setup happens once on the first bringup 438 * of the CPU. It's not destroyed on CPU hotunplug. 439 */ 440 if (st->shared_state) 441 return; 442 443 raw_spin_lock_init(&st->lock); 444 445 /* 446 * Go over HT siblings and check whether one of them has set up the 447 * shared state pointer already. 448 */ 449 for_each_cpu(cpu, topology_sibling_cpumask(this_cpu)) { 450 if (cpu == this_cpu) 451 continue; 452 453 if (!per_cpu(ssb_state, cpu).shared_state) 454 continue; 455 456 /* Link it to the state of the sibling: */ 457 st->shared_state = per_cpu(ssb_state, cpu).shared_state; 458 return; 459 } 460 461 /* 462 * First HT sibling to come up on the core. Link shared state of 463 * the first HT sibling to itself. The siblings on the same core 464 * which come up later will see the shared state pointer and link 465 * themself to the state of this CPU. 466 */ 467 st->shared_state = st; 468 } 469 470 /* 471 * Logic is: First HT sibling enables SSBD for both siblings in the core 472 * and last sibling to disable it, disables it for the whole core. This how 473 * MSR_SPEC_CTRL works in "hardware": 474 * 475 * CORE_SPEC_CTRL = THREAD0_SPEC_CTRL | THREAD1_SPEC_CTRL 476 */ 477 static __always_inline void amd_set_core_ssb_state(unsigned long tifn) 478 { 479 struct ssb_state *st = this_cpu_ptr(&ssb_state); 480 u64 msr = x86_amd_ls_cfg_base; 481 482 if (!static_cpu_has(X86_FEATURE_ZEN)) { 483 msr |= ssbd_tif_to_amd_ls_cfg(tifn); 484 wrmsrl(MSR_AMD64_LS_CFG, msr); 485 return; 486 } 487 488 if (tifn & _TIF_SSBD) { 489 /* 490 * Since this can race with prctl(), block reentry on the 491 * same CPU. 492 */ 493 if (__test_and_set_bit(LSTATE_SSB, &st->local_state)) 494 return; 495 496 msr |= x86_amd_ls_cfg_ssbd_mask; 497 498 raw_spin_lock(&st->shared_state->lock); 499 /* First sibling enables SSBD: */ 500 if (!st->shared_state->disable_state) 501 wrmsrl(MSR_AMD64_LS_CFG, msr); 502 st->shared_state->disable_state++; 503 raw_spin_unlock(&st->shared_state->lock); 504 } else { 505 if (!__test_and_clear_bit(LSTATE_SSB, &st->local_state)) 506 return; 507 508 raw_spin_lock(&st->shared_state->lock); 509 st->shared_state->disable_state--; 510 if (!st->shared_state->disable_state) 511 wrmsrl(MSR_AMD64_LS_CFG, msr); 512 raw_spin_unlock(&st->shared_state->lock); 513 } 514 } 515 #else 516 static __always_inline void amd_set_core_ssb_state(unsigned long tifn) 517 { 518 u64 msr = x86_amd_ls_cfg_base | ssbd_tif_to_amd_ls_cfg(tifn); 519 520 wrmsrl(MSR_AMD64_LS_CFG, msr); 521 } 522 #endif 523 524 static __always_inline void amd_set_ssb_virt_state(unsigned long tifn) 525 { 526 /* 527 * SSBD has the same definition in SPEC_CTRL and VIRT_SPEC_CTRL, 528 * so ssbd_tif_to_spec_ctrl() just works. 529 */ 530 wrmsrl(MSR_AMD64_VIRT_SPEC_CTRL, ssbd_tif_to_spec_ctrl(tifn)); 531 } 532 533 /* 534 * Update the MSRs managing speculation control, during context switch. 535 * 536 * tifp: Previous task's thread flags 537 * tifn: Next task's thread flags 538 */ 539 static __always_inline void __speculation_ctrl_update(unsigned long tifp, 540 unsigned long tifn) 541 { 542 unsigned long tif_diff = tifp ^ tifn; 543 u64 msr = x86_spec_ctrl_base; 544 bool updmsr = false; 545 546 lockdep_assert_irqs_disabled(); 547 548 /* Handle change of TIF_SSBD depending on the mitigation method. */ 549 if (static_cpu_has(X86_FEATURE_VIRT_SSBD)) { 550 if (tif_diff & _TIF_SSBD) 551 amd_set_ssb_virt_state(tifn); 552 } else if (static_cpu_has(X86_FEATURE_LS_CFG_SSBD)) { 553 if (tif_diff & _TIF_SSBD) 554 amd_set_core_ssb_state(tifn); 555 } else if (static_cpu_has(X86_FEATURE_SPEC_CTRL_SSBD) || 556 static_cpu_has(X86_FEATURE_AMD_SSBD)) { 557 updmsr |= !!(tif_diff & _TIF_SSBD); 558 msr |= ssbd_tif_to_spec_ctrl(tifn); 559 } 560 561 /* Only evaluate TIF_SPEC_IB if conditional STIBP is enabled. */ 562 if (IS_ENABLED(CONFIG_SMP) && 563 static_branch_unlikely(&switch_to_cond_stibp)) { 564 updmsr |= !!(tif_diff & _TIF_SPEC_IB); 565 msr |= stibp_tif_to_spec_ctrl(tifn); 566 } 567 568 if (updmsr) 569 wrmsrl(MSR_IA32_SPEC_CTRL, msr); 570 } 571 572 static unsigned long speculation_ctrl_update_tif(struct task_struct *tsk) 573 { 574 if (test_and_clear_tsk_thread_flag(tsk, TIF_SPEC_FORCE_UPDATE)) { 575 if (task_spec_ssb_disable(tsk)) 576 set_tsk_thread_flag(tsk, TIF_SSBD); 577 else 578 clear_tsk_thread_flag(tsk, TIF_SSBD); 579 580 if (task_spec_ib_disable(tsk)) 581 set_tsk_thread_flag(tsk, TIF_SPEC_IB); 582 else 583 clear_tsk_thread_flag(tsk, TIF_SPEC_IB); 584 } 585 /* Return the updated threadinfo flags*/ 586 return task_thread_info(tsk)->flags; 587 } 588 589 void speculation_ctrl_update(unsigned long tif) 590 { 591 unsigned long flags; 592 593 /* Forced update. Make sure all relevant TIF flags are different */ 594 local_irq_save(flags); 595 __speculation_ctrl_update(~tif, tif); 596 local_irq_restore(flags); 597 } 598 599 /* Called from seccomp/prctl update */ 600 void speculation_ctrl_update_current(void) 601 { 602 preempt_disable(); 603 speculation_ctrl_update(speculation_ctrl_update_tif(current)); 604 preempt_enable(); 605 } 606 607 static inline void cr4_toggle_bits_irqsoff(unsigned long mask) 608 { 609 unsigned long newval, cr4 = this_cpu_read(cpu_tlbstate.cr4); 610 611 newval = cr4 ^ mask; 612 if (newval != cr4) { 613 this_cpu_write(cpu_tlbstate.cr4, newval); 614 __write_cr4(newval); 615 } 616 } 617 618 void __switch_to_xtra(struct task_struct *prev_p, struct task_struct *next_p) 619 { 620 unsigned long tifp, tifn; 621 622 tifn = READ_ONCE(task_thread_info(next_p)->flags); 623 tifp = READ_ONCE(task_thread_info(prev_p)->flags); 624 625 switch_to_bitmap(tifp); 626 627 propagate_user_return_notify(prev_p, next_p); 628 629 if ((tifp & _TIF_BLOCKSTEP || tifn & _TIF_BLOCKSTEP) && 630 arch_has_block_step()) { 631 unsigned long debugctl, msk; 632 633 rdmsrl(MSR_IA32_DEBUGCTLMSR, debugctl); 634 debugctl &= ~DEBUGCTLMSR_BTF; 635 msk = tifn & _TIF_BLOCKSTEP; 636 debugctl |= (msk >> TIF_BLOCKSTEP) << DEBUGCTLMSR_BTF_SHIFT; 637 wrmsrl(MSR_IA32_DEBUGCTLMSR, debugctl); 638 } 639 640 if ((tifp ^ tifn) & _TIF_NOTSC) 641 cr4_toggle_bits_irqsoff(X86_CR4_TSD); 642 643 if ((tifp ^ tifn) & _TIF_NOCPUID) 644 set_cpuid_faulting(!!(tifn & _TIF_NOCPUID)); 645 646 if (likely(!((tifp | tifn) & _TIF_SPEC_FORCE_UPDATE))) { 647 __speculation_ctrl_update(tifp, tifn); 648 } else { 649 speculation_ctrl_update_tif(prev_p); 650 tifn = speculation_ctrl_update_tif(next_p); 651 652 /* Enforce MSR update to ensure consistent state */ 653 __speculation_ctrl_update(~tifn, tifn); 654 } 655 656 if ((tifp ^ tifn) & _TIF_SLD) 657 switch_to_sld(tifn); 658 } 659 660 /* 661 * Idle related variables and functions 662 */ 663 unsigned long boot_option_idle_override = IDLE_NO_OVERRIDE; 664 EXPORT_SYMBOL(boot_option_idle_override); 665 666 static void (*x86_idle)(void); 667 668 #ifndef CONFIG_SMP 669 static inline void play_dead(void) 670 { 671 BUG(); 672 } 673 #endif 674 675 void arch_cpu_idle_enter(void) 676 { 677 tsc_verify_tsc_adjust(false); 678 local_touch_nmi(); 679 } 680 681 void arch_cpu_idle_dead(void) 682 { 683 play_dead(); 684 } 685 686 /* 687 * Called from the generic idle code. 688 */ 689 void arch_cpu_idle(void) 690 { 691 x86_idle(); 692 } 693 694 /* 695 * We use this if we don't have any better idle routine.. 696 */ 697 void __cpuidle default_idle(void) 698 { 699 trace_cpu_idle_rcuidle(1, smp_processor_id()); 700 safe_halt(); 701 trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, smp_processor_id()); 702 } 703 #if defined(CONFIG_APM_MODULE) || defined(CONFIG_HALTPOLL_CPUIDLE_MODULE) 704 EXPORT_SYMBOL(default_idle); 705 #endif 706 707 #ifdef CONFIG_XEN 708 bool xen_set_default_idle(void) 709 { 710 bool ret = !!x86_idle; 711 712 x86_idle = default_idle; 713 714 return ret; 715 } 716 #endif 717 718 void stop_this_cpu(void *dummy) 719 { 720 local_irq_disable(); 721 /* 722 * Remove this CPU: 723 */ 724 set_cpu_online(smp_processor_id(), false); 725 disable_local_APIC(); 726 mcheck_cpu_clear(this_cpu_ptr(&cpu_info)); 727 728 /* 729 * Use wbinvd on processors that support SME. This provides support 730 * for performing a successful kexec when going from SME inactive 731 * to SME active (or vice-versa). The cache must be cleared so that 732 * if there are entries with the same physical address, both with and 733 * without the encryption bit, they don't race each other when flushed 734 * and potentially end up with the wrong entry being committed to 735 * memory. 736 */ 737 if (boot_cpu_has(X86_FEATURE_SME)) 738 native_wbinvd(); 739 for (;;) { 740 /* 741 * Use native_halt() so that memory contents don't change 742 * (stack usage and variables) after possibly issuing the 743 * native_wbinvd() above. 744 */ 745 native_halt(); 746 } 747 } 748 749 /* 750 * AMD Erratum 400 aware idle routine. We handle it the same way as C3 power 751 * states (local apic timer and TSC stop). 752 */ 753 static void amd_e400_idle(void) 754 { 755 /* 756 * We cannot use static_cpu_has_bug() here because X86_BUG_AMD_APIC_C1E 757 * gets set after static_cpu_has() places have been converted via 758 * alternatives. 759 */ 760 if (!boot_cpu_has_bug(X86_BUG_AMD_APIC_C1E)) { 761 default_idle(); 762 return; 763 } 764 765 tick_broadcast_enter(); 766 767 default_idle(); 768 769 /* 770 * The switch back from broadcast mode needs to be called with 771 * interrupts disabled. 772 */ 773 local_irq_disable(); 774 tick_broadcast_exit(); 775 local_irq_enable(); 776 } 777 778 /* 779 * Intel Core2 and older machines prefer MWAIT over HALT for C1. 780 * We can't rely on cpuidle installing MWAIT, because it will not load 781 * on systems that support only C1 -- so the boot default must be MWAIT. 782 * 783 * Some AMD machines are the opposite, they depend on using HALT. 784 * 785 * So for default C1, which is used during boot until cpuidle loads, 786 * use MWAIT-C1 on Intel HW that has it, else use HALT. 787 */ 788 static int prefer_mwait_c1_over_halt(const struct cpuinfo_x86 *c) 789 { 790 if (c->x86_vendor != X86_VENDOR_INTEL) 791 return 0; 792 793 if (!cpu_has(c, X86_FEATURE_MWAIT) || boot_cpu_has_bug(X86_BUG_MONITOR)) 794 return 0; 795 796 return 1; 797 } 798 799 /* 800 * MONITOR/MWAIT with no hints, used for default C1 state. This invokes MWAIT 801 * with interrupts enabled and no flags, which is backwards compatible with the 802 * original MWAIT implementation. 803 */ 804 static __cpuidle void mwait_idle(void) 805 { 806 if (!current_set_polling_and_test()) { 807 trace_cpu_idle_rcuidle(1, smp_processor_id()); 808 if (this_cpu_has(X86_BUG_CLFLUSH_MONITOR)) { 809 mb(); /* quirk */ 810 clflush((void *)¤t_thread_info()->flags); 811 mb(); /* quirk */ 812 } 813 814 __monitor((void *)¤t_thread_info()->flags, 0, 0); 815 if (!need_resched()) 816 __sti_mwait(0, 0); 817 else 818 local_irq_enable(); 819 trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, smp_processor_id()); 820 } else { 821 local_irq_enable(); 822 } 823 __current_clr_polling(); 824 } 825 826 void select_idle_routine(const struct cpuinfo_x86 *c) 827 { 828 #ifdef CONFIG_SMP 829 if (boot_option_idle_override == IDLE_POLL && smp_num_siblings > 1) 830 pr_warn_once("WARNING: polling idle and HT enabled, performance may degrade\n"); 831 #endif 832 if (x86_idle || boot_option_idle_override == IDLE_POLL) 833 return; 834 835 if (boot_cpu_has_bug(X86_BUG_AMD_E400)) { 836 pr_info("using AMD E400 aware idle routine\n"); 837 x86_idle = amd_e400_idle; 838 } else if (prefer_mwait_c1_over_halt(c)) { 839 pr_info("using mwait in idle threads\n"); 840 x86_idle = mwait_idle; 841 } else 842 x86_idle = default_idle; 843 } 844 845 void amd_e400_c1e_apic_setup(void) 846 { 847 if (boot_cpu_has_bug(X86_BUG_AMD_APIC_C1E)) { 848 pr_info("Switch to broadcast mode on CPU%d\n", smp_processor_id()); 849 local_irq_disable(); 850 tick_broadcast_force(); 851 local_irq_enable(); 852 } 853 } 854 855 void __init arch_post_acpi_subsys_init(void) 856 { 857 u32 lo, hi; 858 859 if (!boot_cpu_has_bug(X86_BUG_AMD_E400)) 860 return; 861 862 /* 863 * AMD E400 detection needs to happen after ACPI has been enabled. If 864 * the machine is affected K8_INTP_C1E_ACTIVE_MASK bits are set in 865 * MSR_K8_INT_PENDING_MSG. 866 */ 867 rdmsr(MSR_K8_INT_PENDING_MSG, lo, hi); 868 if (!(lo & K8_INTP_C1E_ACTIVE_MASK)) 869 return; 870 871 boot_cpu_set_bug(X86_BUG_AMD_APIC_C1E); 872 873 if (!boot_cpu_has(X86_FEATURE_NONSTOP_TSC)) 874 mark_tsc_unstable("TSC halt in AMD C1E"); 875 pr_info("System has AMD C1E enabled\n"); 876 } 877 878 static int __init idle_setup(char *str) 879 { 880 if (!str) 881 return -EINVAL; 882 883 if (!strcmp(str, "poll")) { 884 pr_info("using polling idle threads\n"); 885 boot_option_idle_override = IDLE_POLL; 886 cpu_idle_poll_ctrl(true); 887 } else if (!strcmp(str, "halt")) { 888 /* 889 * When the boot option of idle=halt is added, halt is 890 * forced to be used for CPU idle. In such case CPU C2/C3 891 * won't be used again. 892 * To continue to load the CPU idle driver, don't touch 893 * the boot_option_idle_override. 894 */ 895 x86_idle = default_idle; 896 boot_option_idle_override = IDLE_HALT; 897 } else if (!strcmp(str, "nomwait")) { 898 /* 899 * If the boot option of "idle=nomwait" is added, 900 * it means that mwait will be disabled for CPU C2/C3 901 * states. In such case it won't touch the variable 902 * of boot_option_idle_override. 903 */ 904 boot_option_idle_override = IDLE_NOMWAIT; 905 } else 906 return -1; 907 908 return 0; 909 } 910 early_param("idle", idle_setup); 911 912 unsigned long arch_align_stack(unsigned long sp) 913 { 914 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space) 915 sp -= get_random_int() % 8192; 916 return sp & ~0xf; 917 } 918 919 unsigned long arch_randomize_brk(struct mm_struct *mm) 920 { 921 return randomize_page(mm->brk, 0x02000000); 922 } 923 924 /* 925 * Called from fs/proc with a reference on @p to find the function 926 * which called into schedule(). This needs to be done carefully 927 * because the task might wake up and we might look at a stack 928 * changing under us. 929 */ 930 unsigned long get_wchan(struct task_struct *p) 931 { 932 unsigned long start, bottom, top, sp, fp, ip, ret = 0; 933 int count = 0; 934 935 if (p == current || p->state == TASK_RUNNING) 936 return 0; 937 938 if (!try_get_task_stack(p)) 939 return 0; 940 941 start = (unsigned long)task_stack_page(p); 942 if (!start) 943 goto out; 944 945 /* 946 * Layout of the stack page: 947 * 948 * ----------- topmax = start + THREAD_SIZE - sizeof(unsigned long) 949 * PADDING 950 * ----------- top = topmax - TOP_OF_KERNEL_STACK_PADDING 951 * stack 952 * ----------- bottom = start 953 * 954 * The tasks stack pointer points at the location where the 955 * framepointer is stored. The data on the stack is: 956 * ... IP FP ... IP FP 957 * 958 * We need to read FP and IP, so we need to adjust the upper 959 * bound by another unsigned long. 960 */ 961 top = start + THREAD_SIZE - TOP_OF_KERNEL_STACK_PADDING; 962 top -= 2 * sizeof(unsigned long); 963 bottom = start; 964 965 sp = READ_ONCE(p->thread.sp); 966 if (sp < bottom || sp > top) 967 goto out; 968 969 fp = READ_ONCE_NOCHECK(((struct inactive_task_frame *)sp)->bp); 970 do { 971 if (fp < bottom || fp > top) 972 goto out; 973 ip = READ_ONCE_NOCHECK(*(unsigned long *)(fp + sizeof(unsigned long))); 974 if (!in_sched_functions(ip)) { 975 ret = ip; 976 goto out; 977 } 978 fp = READ_ONCE_NOCHECK(*(unsigned long *)fp); 979 } while (count++ < 16 && p->state != TASK_RUNNING); 980 981 out: 982 put_task_stack(p); 983 return ret; 984 } 985 986 long do_arch_prctl_common(struct task_struct *task, int option, 987 unsigned long cpuid_enabled) 988 { 989 switch (option) { 990 case ARCH_GET_CPUID: 991 return get_cpuid_mode(); 992 case ARCH_SET_CPUID: 993 return set_cpuid_mode(task, cpuid_enabled); 994 } 995 996 return -EINVAL; 997 } 998