1 // SPDX-License-Identifier: GPL-2.0-only
3 * Based on arch/arm/kernel/process.c
5 * Original Copyright (C) 1995 Linus Torvalds
6 * Copyright (C) 1996-2000 Russell King - Converted to ARM.
7 * Copyright (C) 2012 ARM Ltd.
12 #include <linux/compat.h>
13 #include <linux/efi.h>
14 #include <linux/elf.h>
15 #include <linux/export.h>
16 #include <linux/sched.h>
17 #include <linux/sched/debug.h>
18 #include <linux/sched/task.h>
19 #include <linux/sched/task_stack.h>
20 #include <linux/kernel.h>
21 #include <linux/lockdep.h>
22 #include <linux/mman.h>
24 #include <linux/nospec.h>
25 #include <linux/stddef.h>
26 #include <linux/sysctl.h>
27 #include <linux/unistd.h>
28 #include <linux/user.h>
29 #include <linux/delay.h>
30 #include <linux/reboot.h>
31 #include <linux/interrupt.h>
32 #include <linux/init.h>
33 #include <linux/cpu.h>
34 #include <linux/elfcore.h>
36 #include <linux/tick.h>
37 #include <linux/utsname.h>
38 #include <linux/uaccess.h>
39 #include <linux/random.h>
40 #include <linux/hw_breakpoint.h>
41 #include <linux/personality.h>
42 #include <linux/notifier.h>
43 #include <trace/events/power.h>
44 #include <linux/percpu.h>
45 #include <linux/thread_info.h>
46 #include <linux/prctl.h>
48 #include <asm/alternative.h>
49 #include <asm/arch_gicv3.h>
50 #include <asm/compat.h>
51 #include <asm/cpufeature.h>
52 #include <asm/cacheflush.h>
54 #include <asm/fpsimd.h>
55 #include <asm/mmu_context.h>
57 #include <asm/processor.h>
58 #include <asm/pointer_auth.h>
59 #include <asm/stacktrace.h>
61 #if defined(CONFIG_STACKPROTECTOR) && !defined(CONFIG_STACKPROTECTOR_PER_TASK)
62 #include <linux/stackprotector.h>
63 unsigned long __stack_chk_guard __read_mostly;
64 EXPORT_SYMBOL(__stack_chk_guard);
68 * Function pointers to optional machine specific functions
70 void (*pm_power_off)(void);
71 EXPORT_SYMBOL_GPL(pm_power_off);
73 void (*arm_pm_restart)(enum reboot_mode reboot_mode, const char *cmd);
75 static void __cpu_do_idle(void)
81 static void __cpu_do_idle_irqprio(void)
84 unsigned long daif_bits;
86 daif_bits = read_sysreg(daif);
87 write_sysreg(daif_bits | PSR_I_BIT, daif);
90 * Unmask PMR before going idle to make sure interrupts can
94 gic_write_pmr(GIC_PRIO_IRQON | GIC_PRIO_PSR_I_SET);
99 write_sysreg(daif_bits, daif);
105 * Idle the processor (wait for interrupt).
107 * If the CPU supports priority masking we must do additional work to
108 * ensure that interrupts are not masked at the PMR (because the core will
109 * not wake up if we block the wake up signal in the interrupt controller).
111 void cpu_do_idle(void)
113 if (system_uses_irq_prio_masking())
114 __cpu_do_idle_irqprio();
120 * This is our default idle handler.
122 void arch_cpu_idle(void)
125 * This should do all the clock switching and wait for interrupt
132 #ifdef CONFIG_HOTPLUG_CPU
133 void arch_cpu_idle_dead(void)
140 * Called by kexec, immediately prior to machine_kexec().
142 * This must completely disable all secondary CPUs; simply causing those CPUs
143 * to execute e.g. a RAM-based pin loop is not sufficient. This allows the
144 * kexec'd kernel to use any and all RAM as it sees fit, without having to
145 * avoid any code or data used by any SW CPU pin loop. The CPU hotplug
146 * functionality embodied in smpt_shutdown_nonboot_cpus() to achieve this.
148 void machine_shutdown(void)
150 smp_shutdown_nonboot_cpus(reboot_cpu);
154 * Halting simply requires that the secondary CPUs stop performing any
155 * activity (executing tasks, handling interrupts). smp_send_stop()
158 void machine_halt(void)
166 * Power-off simply requires that the secondary CPUs stop performing any
167 * activity (executing tasks, handling interrupts). smp_send_stop()
168 * achieves this. When the system power is turned off, it will take all CPUs
171 void machine_power_off(void)
180 * Restart requires that the secondary CPUs stop performing any activity
181 * while the primary CPU resets the system. Systems with multiple CPUs must
182 * provide a HW restart implementation, to ensure that all CPUs reset at once.
183 * This is required so that any code running after reset on the primary CPU
184 * doesn't have to co-ordinate with other CPUs to ensure they aren't still
185 * executing pre-reset code, and using RAM that the primary CPU's code wishes
186 * to use. Implementing such co-ordination would be essentially impossible.
188 void machine_restart(char *cmd)
190 /* Disable interrupts first */
195 * UpdateCapsule() depends on the system being reset via
198 if (efi_enabled(EFI_RUNTIME_SERVICES))
199 efi_reboot(reboot_mode, NULL);
201 /* Now call the architecture specific reboot code. */
203 arm_pm_restart(reboot_mode, cmd);
205 do_kernel_restart(cmd);
208 * Whoops - the architecture was unable to reboot.
210 printk("Reboot failed -- System halted\n");
214 #define bstr(suffix, str) [PSR_BTYPE_ ## suffix >> PSR_BTYPE_SHIFT] = str
215 static const char *const btypes[] = {
223 static void print_pstate(struct pt_regs *regs)
225 u64 pstate = regs->pstate;
227 if (compat_user_mode(regs)) {
228 printk("pstate: %08llx (%c%c%c%c %c %s %s %c%c%c)\n",
230 pstate & PSR_AA32_N_BIT ? 'N' : 'n',
231 pstate & PSR_AA32_Z_BIT ? 'Z' : 'z',
232 pstate & PSR_AA32_C_BIT ? 'C' : 'c',
233 pstate & PSR_AA32_V_BIT ? 'V' : 'v',
234 pstate & PSR_AA32_Q_BIT ? 'Q' : 'q',
235 pstate & PSR_AA32_T_BIT ? "T32" : "A32",
236 pstate & PSR_AA32_E_BIT ? "BE" : "LE",
237 pstate & PSR_AA32_A_BIT ? 'A' : 'a',
238 pstate & PSR_AA32_I_BIT ? 'I' : 'i',
239 pstate & PSR_AA32_F_BIT ? 'F' : 'f');
241 const char *btype_str = btypes[(pstate & PSR_BTYPE_MASK) >>
244 printk("pstate: %08llx (%c%c%c%c %c%c%c%c %cPAN %cUAO %cTCO BTYPE=%s)\n",
246 pstate & PSR_N_BIT ? 'N' : 'n',
247 pstate & PSR_Z_BIT ? 'Z' : 'z',
248 pstate & PSR_C_BIT ? 'C' : 'c',
249 pstate & PSR_V_BIT ? 'V' : 'v',
250 pstate & PSR_D_BIT ? 'D' : 'd',
251 pstate & PSR_A_BIT ? 'A' : 'a',
252 pstate & PSR_I_BIT ? 'I' : 'i',
253 pstate & PSR_F_BIT ? 'F' : 'f',
254 pstate & PSR_PAN_BIT ? '+' : '-',
255 pstate & PSR_UAO_BIT ? '+' : '-',
256 pstate & PSR_TCO_BIT ? '+' : '-',
261 void __show_regs(struct pt_regs *regs)
266 if (compat_user_mode(regs)) {
267 lr = regs->compat_lr;
268 sp = regs->compat_sp;
276 show_regs_print_info(KERN_DEFAULT);
279 if (!user_mode(regs)) {
280 printk("pc : %pS\n", (void *)regs->pc);
281 printk("lr : %pS\n", (void *)ptrauth_strip_insn_pac(lr));
283 printk("pc : %016llx\n", regs->pc);
284 printk("lr : %016llx\n", lr);
287 printk("sp : %016llx\n", sp);
289 if (system_uses_irq_prio_masking())
290 printk("pmr_save: %08llx\n", regs->pmr_save);
295 printk("x%-2d: %016llx ", i, regs->regs[i]);
299 pr_cont("x%-2d: %016llx ", i, regs->regs[i]);
307 void show_regs(struct pt_regs * regs)
310 dump_backtrace(regs, NULL, KERN_DEFAULT);
313 static void tls_thread_flush(void)
315 write_sysreg(0, tpidr_el0);
317 if (is_compat_task()) {
318 current->thread.uw.tp_value = 0;
321 * We need to ensure ordering between the shadow state and the
322 * hardware state, so that we don't corrupt the hardware state
323 * with a stale shadow state during context switch.
326 write_sysreg(0, tpidrro_el0);
330 static void flush_tagged_addr_state(void)
332 if (IS_ENABLED(CONFIG_ARM64_TAGGED_ADDR_ABI))
333 clear_thread_flag(TIF_TAGGED_ADDR);
336 void flush_thread(void)
338 fpsimd_flush_thread();
340 flush_ptrace_hw_breakpoint(current);
341 flush_tagged_addr_state();
345 void release_thread(struct task_struct *dead_task)
349 void arch_release_task_struct(struct task_struct *tsk)
351 fpsimd_release_task(tsk);
354 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
357 fpsimd_preserve_current_state();
360 /* We rely on the above assignment to initialize dst's thread_flags: */
361 BUILD_BUG_ON(!IS_ENABLED(CONFIG_THREAD_INFO_IN_TASK));
364 * Detach src's sve_state (if any) from dst so that it does not
365 * get erroneously used or freed prematurely. dst's sve_state
366 * will be allocated on demand later on if dst uses SVE.
367 * For consistency, also clear TIF_SVE here: this could be done
368 * later in copy_process(), but to avoid tripping up future
369 * maintainers it is best not to leave TIF_SVE and sve_state in
370 * an inconsistent state, even temporarily.
372 dst->thread.sve_state = NULL;
373 clear_tsk_thread_flag(dst, TIF_SVE);
375 /* clear any pending asynchronous tag fault raised by the parent */
376 clear_tsk_thread_flag(dst, TIF_MTE_ASYNC_FAULT);
381 asmlinkage void ret_from_fork(void) asm("ret_from_fork");
383 int copy_thread(unsigned long clone_flags, unsigned long stack_start,
384 unsigned long stk_sz, struct task_struct *p, unsigned long tls)
386 struct pt_regs *childregs = task_pt_regs(p);
388 memset(&p->thread.cpu_context, 0, sizeof(struct cpu_context));
391 * In case p was allocated the same task_struct pointer as some
392 * other recently-exited task, make sure p is disassociated from
393 * any cpu that may have run that now-exited task recently.
394 * Otherwise we could erroneously skip reloading the FPSIMD
397 fpsimd_flush_task_state(p);
399 ptrauth_thread_init_kernel(p);
401 if (likely(!(p->flags & PF_KTHREAD))) {
402 *childregs = *current_pt_regs();
403 childregs->regs[0] = 0;
406 * Read the current TLS pointer from tpidr_el0 as it may be
407 * out-of-sync with the saved value.
409 *task_user_tls(p) = read_sysreg(tpidr_el0);
412 if (is_compat_thread(task_thread_info(p)))
413 childregs->compat_sp = stack_start;
415 childregs->sp = stack_start;
419 * If a TLS pointer was passed to clone, use it for the new
422 if (clone_flags & CLONE_SETTLS)
423 p->thread.uw.tp_value = tls;
425 memset(childregs, 0, sizeof(struct pt_regs));
426 childregs->pstate = PSR_MODE_EL1h;
427 if (IS_ENABLED(CONFIG_ARM64_UAO) &&
428 cpus_have_const_cap(ARM64_HAS_UAO))
429 childregs->pstate |= PSR_UAO_BIT;
431 spectre_v4_enable_task_mitigation(p);
433 if (system_uses_irq_prio_masking())
434 childregs->pmr_save = GIC_PRIO_IRQON;
436 p->thread.cpu_context.x19 = stack_start;
437 p->thread.cpu_context.x20 = stk_sz;
439 p->thread.cpu_context.pc = (unsigned long)ret_from_fork;
440 p->thread.cpu_context.sp = (unsigned long)childregs;
442 ptrace_hw_copy_thread(p);
447 void tls_preserve_current_state(void)
449 *task_user_tls(current) = read_sysreg(tpidr_el0);
452 static void tls_thread_switch(struct task_struct *next)
454 tls_preserve_current_state();
456 if (is_compat_thread(task_thread_info(next)))
457 write_sysreg(next->thread.uw.tp_value, tpidrro_el0);
458 else if (!arm64_kernel_unmapped_at_el0())
459 write_sysreg(0, tpidrro_el0);
461 write_sysreg(*task_user_tls(next), tpidr_el0);
464 /* Restore the UAO state depending on next's addr_limit */
465 void uao_thread_switch(struct task_struct *next)
467 if (IS_ENABLED(CONFIG_ARM64_UAO)) {
468 if (task_thread_info(next)->addr_limit == KERNEL_DS)
469 asm(ALTERNATIVE("nop", SET_PSTATE_UAO(1), ARM64_HAS_UAO));
471 asm(ALTERNATIVE("nop", SET_PSTATE_UAO(0), ARM64_HAS_UAO));
476 * Force SSBS state on context-switch, since it may be lost after migrating
477 * from a CPU which treats the bit as RES0 in a heterogeneous system.
479 static void ssbs_thread_switch(struct task_struct *next)
482 * Nothing to do for kernel threads, but 'regs' may be junk
483 * (e.g. idle task) so check the flags and bail early.
485 if (unlikely(next->flags & PF_KTHREAD))
489 * If all CPUs implement the SSBS extension, then we just need to
490 * context-switch the PSTATE field.
492 if (cpus_have_const_cap(ARM64_SSBS))
495 spectre_v4_enable_task_mitigation(next);
499 * We store our current task in sp_el0, which is clobbered by userspace. Keep a
500 * shadow copy so that we can restore this upon entry from userspace.
502 * This is *only* for exception entry from EL0, and is not valid until we
503 * __switch_to() a user task.
505 DEFINE_PER_CPU(struct task_struct *, __entry_task);
507 static void entry_task_switch(struct task_struct *next)
509 __this_cpu_write(__entry_task, next);
513 * ARM erratum 1418040 handling, affecting the 32bit view of CNTVCT.
514 * Assuming the virtual counter is enabled at the beginning of times:
516 * - disable access when switching from a 64bit task to a 32bit task
517 * - enable access when switching from a 32bit task to a 64bit task
519 static void erratum_1418040_thread_switch(struct task_struct *prev,
520 struct task_struct *next)
525 if (!IS_ENABLED(CONFIG_ARM64_ERRATUM_1418040))
528 prev32 = is_compat_thread(task_thread_info(prev));
529 next32 = is_compat_thread(task_thread_info(next));
531 if (prev32 == next32 || !this_cpu_has_cap(ARM64_WORKAROUND_1418040))
534 val = read_sysreg(cntkctl_el1);
537 val |= ARCH_TIMER_USR_VCT_ACCESS_EN;
539 val &= ~ARCH_TIMER_USR_VCT_ACCESS_EN;
541 write_sysreg(val, cntkctl_el1);
547 __notrace_funcgraph struct task_struct *__switch_to(struct task_struct *prev,
548 struct task_struct *next)
550 struct task_struct *last;
552 fpsimd_thread_switch(next);
553 tls_thread_switch(next);
554 hw_breakpoint_thread_switch(next);
555 contextidr_thread_switch(next);
556 entry_task_switch(next);
557 uao_thread_switch(next);
558 ssbs_thread_switch(next);
559 erratum_1418040_thread_switch(prev, next);
562 * Complete any pending TLB or cache maintenance on this CPU in case
563 * the thread migrates to a different CPU.
564 * This full barrier is also required by the membarrier system
570 * MTE thread switching must happen after the DSB above to ensure that
571 * any asynchronous tag check faults have been logged in the TFSR*_EL1
574 mte_thread_switch(next);
576 /* the actual thread switch */
577 last = cpu_switch_to(prev, next);
582 unsigned long get_wchan(struct task_struct *p)
584 struct stackframe frame;
585 unsigned long stack_page, ret = 0;
587 if (!p || p == current || p->state == TASK_RUNNING)
590 stack_page = (unsigned long)try_get_task_stack(p);
594 start_backtrace(&frame, thread_saved_fp(p), thread_saved_pc(p));
597 if (unwind_frame(p, &frame))
599 if (!in_sched_functions(frame.pc)) {
603 } while (count ++ < 16);
610 unsigned long arch_align_stack(unsigned long sp)
612 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
613 sp -= get_random_int() & ~PAGE_MASK;
618 * Called from setup_new_exec() after (COMPAT_)SET_PERSONALITY.
620 void arch_setup_new_exec(void)
622 current->mm->context.flags = is_compat_task() ? MMCF_AARCH32 : 0;
624 ptrauth_thread_init_user(current);
626 if (task_spec_ssb_noexec(current)) {
627 arch_prctl_spec_ctrl_set(current, PR_SPEC_STORE_BYPASS,
632 #ifdef CONFIG_ARM64_TAGGED_ADDR_ABI
634 * Control the relaxed ABI allowing tagged user addresses into the kernel.
636 static unsigned int tagged_addr_disabled;
638 long set_tagged_addr_ctrl(struct task_struct *task, unsigned long arg)
640 unsigned long valid_mask = PR_TAGGED_ADDR_ENABLE;
641 struct thread_info *ti = task_thread_info(task);
643 if (is_compat_thread(ti))
646 if (system_supports_mte())
647 valid_mask |= PR_MTE_TCF_MASK | PR_MTE_TAG_MASK;
649 if (arg & ~valid_mask)
653 * Do not allow the enabling of the tagged address ABI if globally
654 * disabled via sysctl abi.tagged_addr_disabled.
656 if (arg & PR_TAGGED_ADDR_ENABLE && tagged_addr_disabled)
659 if (set_mte_ctrl(task, arg) != 0)
662 update_ti_thread_flag(ti, TIF_TAGGED_ADDR, arg & PR_TAGGED_ADDR_ENABLE);
667 long get_tagged_addr_ctrl(struct task_struct *task)
670 struct thread_info *ti = task_thread_info(task);
672 if (is_compat_thread(ti))
675 if (test_ti_thread_flag(ti, TIF_TAGGED_ADDR))
676 ret = PR_TAGGED_ADDR_ENABLE;
678 ret |= get_mte_ctrl(task);
684 * Global sysctl to disable the tagged user addresses support. This control
685 * only prevents the tagged address ABI enabling via prctl() and does not
686 * disable it for tasks that already opted in to the relaxed ABI.
689 static struct ctl_table tagged_addr_sysctl_table[] = {
691 .procname = "tagged_addr_disabled",
693 .data = &tagged_addr_disabled,
694 .maxlen = sizeof(int),
695 .proc_handler = proc_dointvec_minmax,
696 .extra1 = SYSCTL_ZERO,
697 .extra2 = SYSCTL_ONE,
702 static int __init tagged_addr_init(void)
704 if (!register_sysctl("abi", tagged_addr_sysctl_table))
709 core_initcall(tagged_addr_init);
710 #endif /* CONFIG_ARM64_TAGGED_ADDR_ABI */
712 asmlinkage void __sched arm64_preempt_schedule_irq(void)
714 lockdep_assert_irqs_disabled();
717 * Preempting a task from an IRQ means we leave copies of PSTATE
718 * on the stack. cpufeature's enable calls may modify PSTATE, but
719 * resuming one of these preempted tasks would undo those changes.
721 * Only allow a task to be preempted once cpufeatures have been
724 if (system_capabilities_finalized())
725 preempt_schedule_irq();
728 #ifdef CONFIG_BINFMT_ELF
729 int arch_elf_adjust_prot(int prot, const struct arch_elf_state *state,
730 bool has_interp, bool is_interp)
733 * For dynamically linked executables the interpreter is
734 * responsible for setting PROT_BTI on everything except
737 if (is_interp != has_interp)
740 if (!(state->flags & ARM64_ELF_BTI))
743 if (prot & PROT_EXEC)